# Account deletion
Source: https://relevanceai.com/docs/admin/account-management/account-deletion

How to delete your Relevance AI account

At this time, the only way to delete your Relevance AI account is to reach out to our support team to request your account deletion.

You can learn how to contact our support team at [here](https://relevanceai.com/docs/get-started/support).

Once your request is processed by our team, your account will be deleted in 30 days.

<Note>Adding the ability to self-serve account deletion is on our roadmap!</Note>


# Changing your email address
Source: https://relevanceai.com/docs/admin/account-management/change-email-address

How to change your Relevance AI account email address

You can now change your Relevance AI account email address directly from your profile settings. This process allows you to update your account email while maintaining access to all your projects and resources.

## How to change your email address

<Warning>
  You cannot change your email address to one that is already associated with another Relevance AI account. The email address you choose must be unique and not currently in use.
</Warning>

<div>
  <iframe title="Changing your email address" />
</div>

Follow these steps to change your email address:

1. **Log into Relevance AI** at [https://app.relevanceai.com](https://app.relevanceai.com)
2. **Click on your profile picture** in the top right corner, then select **Profile**
3. **Click on the Security tab**
4. **On the email address panel**, click the **Edit** button
5. **Enter your current password** for security verification
6. **Enter your new email address** and confirm it by entering it again in the confirmation field
7. **Submit the change** - a verification code will be sent to your new email address
8. **Check your new email inbox** for the verification code
9. **Enter the verification code** on the same screen to complete the email change process

<Warning>
  You will receive a verification code at your new email address. Make sure you can receive emails at the new email address before proceeding, otherwise you may lose access to your account. You can enter the verification code on the same screen to complete the process.
</Warning>

## Who cannot change their email

Some users are unable to change their email address through this process:

### Google sign-in users

If you signed up for Relevance AI using Google authentication (Sign in with Google), you cannot change your email address through this method at this time. This capability may be added in the future.

<Frame>
  <img alt="Email change screen for Google sign-in users" />
</Frame>

<AccordionGroup>
  <Accordion title="Workaround for Google sign-in users">
    1. Create a new account with your new email address at [https://app.relevanceai.com/auth](https://app.relevanceai.com/auth)
    2. After successfully creating a new account with your new email address, log back into your old account and add this account to all of your project(s)
    3. Reach out to our support team to delete your old account and change your billing details to reflect your new email if you have a subscription. You can learn how to reach out to our support team [here](https://relevanceai.com/docs/get-started/support)
  </Accordion>
</AccordionGroup>

### Enterprise users with SSO enforced

If you are an Enterprise user and your organization's administrator has enforced Single Sign-On (SSO), you cannot change your email address through this method. Contact your organization's administrator for assistance with email address changes. SSO enforcement is only available to Enterprise customers.

<Frame>
  <img alt="Email change screen for Enterprise users with SSO enforced" />
</Frame>

<Note>
  Enterprise users who do not have SSO enforced by their administrators can change their email addresses using the standard process described above.
</Note>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Will I lose access to my projects when I change my email?">
    No, all your projects, agents, and resources will remain intact. You'll need to verify your new email address before you can log in with it.
  </Accordion>

  <Accordion title="What happens if I don't verify the new email address?">
    If you don't verify your new email address, the change will not be completed and you'll need to continue using your old email address to log in. If you're experiencing issues with verification, please reach out to our support team [here](https://relevanceai.com/docs/get-started/support).
  </Accordion>

  <Accordion title="Can I change my email back to the original one?">
    Yes, you can change your email address again using the same process. Just make sure you can receive verification emails at the address you want to use.
  </Accordion>

  <Accordion title="What if I signed up with Google and want to change my email?">
    Currently, users who signed up with Google authentication need to follow the workaround process described above. We may add this capability in the future.
  </Accordion>

  <Accordion title="I'm an Enterprise user - can I change my email?">
    If your organization has enforced SSO, you cannot change your email address through this method. Contact your organization's administrator for assistance. If SSO is not enforced, you can change your email address using the standard process described above.
  </Accordion>

  <Accordion title="What if the email address I want to use is already taken?">
    You cannot change your email address to one that is already associated with another Relevance AI account. Each email address must be unique. If you need to use an email address that's already in use, you'll need to first delete or change the email on the existing account that's using it.
  </Accordion>
</AccordionGroup>


# Changing your password
Source: https://relevanceai.com/docs/admin/account-management/change-password

How to change your Relevance AI account password

You can now change your Relevance AI account password directly from your profile settings. This allows you to update your password securely while maintaining access to all your projects and resources.

## How to change your password

<div>
  <iframe title="Changing your password" />
</div>

Follow these steps to change your password:

1. **Log into Relevance AI** at [https://app.relevanceai.com](https://app.relevanceai.com)
2. **Click on your profile picture** in the top right corner, then select **Profile**
3. **Click on the Security tab**
4. **On the password panel**, click the **Edit** button
5. **Enter your current password** for security verification
6. **Enter your new password** and confirm it by entering it again in the confirmation field
7. **Click 'Update Password'** to change your password

### Password requirements

Your new password must meet the following requirements:

* **Minimum 12 characters**
* **At least one lowercase letter**
* **At least one uppercase letter**
* **At least one number**
* **At least one symbol**
* **Maximum 128 characters**
* **Must be different from your previous password**
* **Cannot be a password you've used before**

<Tip>
  We recommend saving your password in a secure password manager app to keep it safe and easily accessible.
</Tip>

## If you've forgotten your password

If you can't remember your current password or are logged out, follow these steps to reset it:

<div>
  <iframe title="How to reset your password" />
</div>

1. Head over to our [log in page](https://app.relevanceai.com/auth/reset)
2. Enter your email address
3. Click **'Forgot password'**
4. Enter your email address again on the reset page, then click **'Forgot password'**
5. You will be sent an email with instructions to reset your account password

## Who cannot change their password

Some users are unable to change their password through this method:

### Google sign-in users

If you signed up for Relevance AI using Google authentication (Sign in with Google), you cannot change your password through this method as your password is managed by Google.

<Frame>
  <img alt="Password change screen for Google sign-in users" />
</Frame>

### Enterprise users with SSO enforced

If you are an Enterprise user and your organization's administrator has enforced Single Sign-On (SSO), you cannot change your password through this method. Your password is managed by your organization's identity provider. Contact your organization's administrator for assistance with password changes. SSO enforcement is only available to Enterprise customers.

<Frame>
  <img alt="Password change screen for Enterprise users with SSO enforced" />
</Frame>

<Note>
  Enterprise users who do not have SSO enforced by their administrators can change their passwords using the standard process described above.
</Note>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What if I can't remember my current password?">
    If you can't remember your current password, follow the "If you've forgotten your password" process described above to reset it via email.
  </Accordion>

  <Accordion title="How often should I change my password?">
    We recommend changing your password regularly for security purposes, especially if you suspect your account may have been compromised.
  </Accordion>

  <Accordion title="What if my new password doesn't meet the requirements?">
    Make sure your password includes at least 12 characters with a mix of uppercase letters, lowercase letters, numbers, and symbols. The system will prompt you if your password doesn't meet these requirements.
  </Accordion>

  <Accordion title="What if I signed up with Google?">
    If you signed up with Google authentication, your password is managed by Google. You'll need to change your Google account password through Google's password management system.
  </Accordion>

  <Accordion title="I'm an Enterprise user - can I change my password?">
    If your organization has enforced SSO, your password is managed by your organization's identity provider. Contact your organization's administrator for assistance. If SSO is not enforced, you can change your password using the standard process described above.
  </Accordion>

  <Accordion title="What if I'm having trouble resetting my password?">
    If you're experiencing issues with password reset, please reach out to our support team [here](https://relevanceai.com/docs/get-started/support).
  </Accordion>
</AccordionGroup>


# Update your profile information
Source: https://relevanceai.com/docs/admin/account-management/update-profile-info

How to update your name and company on your Relevance AI profile

You can update your personal details including your name and company information from your profile settings. These details help personalize your experience across Relevance AI.

## How to update your profile information

<img alt="Personal Details page in Profile settings" />

Follow these steps to update your profile information:

1. **Log into Relevance AI** at [https://app.relevanceai.com](https://app.relevanceai.com)
2. **Click on your profile picture** in the top right corner, then select **Profile**
3. You will be on the **Personal Details** page
4. **Click the Edit button** on the Personal Information section
5. **Update your first name, last name, and/or company** as needed
6. **Click 'Save Changes'** to update your profile

### About your name

Your name appears throughout Relevance AI in several places:

* When you call an Agent
* What your collaborators see when you interact with shared resources
* What admins see on their Members page
* Throughout the platform when your account is referenced

### About your company

The company field updates only your personal profile and does not affect your organization or project names in Relevance AI.

<Tip>
  To change your organization or project name, please see [Change project and organization name](/docs/admin/project-management/change-project-and-org-name).
</Tip>

## Changing your profile picture

Your profile picture appears throughout Relevance AI when you interact with agents, collaborate with team members, and across the platform. How you update your profile picture depends on how you signed up for Relevance AI.

### For Google sign-in users

If you signed up using Google authentication, your profile picture comes from your Google account. To change it, update your Google profile picture and it will automatically sync to Relevance AI.

### For Enterprise users with SSO enforced

If you're an Enterprise user and your organization has enforced Single Sign-On (SSO), your profile picture comes from your organization's identity provider. To change it, update your profile picture in your organization's identity management system.

### For all other users (Gravatar)

If you signed up with an email and password, you can set your profile picture using [Gravatar](https://gravatar.com).

<Note>
  **What is Gravatar?** Gravatar (Globally Recognized Avatar) is a free service that allows you to use a single profile picture across multiple websites and platforms. Once you set up your Gravatar profile with your email address, any website that supports Gravatar will automatically display your chosen picture.
</Note>

#### Setting up your Gravatar profile picture

<Warning>
  Your Gravatar email address must match your Relevance AI account email address. You can check your current email address on the Security page in your Profile settings. If needed, [change your email address](/docs/admin/account-management/change-email-address) to match your Gravatar account.
</Warning>

Follow these steps to set up your profile picture using Gravatar:

<Steps>
  <Step title="Create a Gravatar account">
    Go to [Gravatar](https://gravatar.com) and create an account (if you don't have one). Use the same email address as your Relevance AI account when creating your Gravatar profile.
  </Step>

  <Step title="Upload your profile picture">
    Upload a profile picture to your Gravatar account.
  </Step>

  <Step title="Navigate to Profile Picture settings">
    Return to Relevance AI and navigate to **Profile** → **Personal Details** → **Profile Picture**.
  </Step>

  <Step title="Edit Profile Picture">
    Click the **Edit** button.
  </Step>

  <Step title="Retrieve from Gravatar">
    Click **'Retrieve profile picture'** to sync your Gravatar image with Relevance AI.
  </Step>
</Steps>

#### Updating your profile picture

If you've updated your Gravatar profile picture and want to sync the changes to Relevance AI:

<Tip>
  Any changes you make to your Gravatar profile picture require you to remove and re-retrieve it in Relevance AI to see the updates.
</Tip>

<Steps>
  <Step title="Navigate to Profile Picture settings">
    Go to **Profile** → **Personal Details** → **Profile Picture**.
  </Step>

  <Step title="Edit Profile Picture">
    Click the **Edit** button.
  </Step>

  <Step title="Remove current picture">
    Click **'Remove profile picture'** to remove the current image.
  </Step>

  <Step title="Retrieve updated picture">
    Click **'Retrieve profile picture'** to get your latest Gravatar image.
  </Step>
</Steps>

#### Removing your profile picture

To remove your profile picture:

<Steps>
  <Step title="Navigate to Profile Picture settings">
    Go to **Profile** → **Personal Details** → **Profile Picture**.
  </Step>

  <Step title="Edit Profile Picture">
    Click the **Edit** button.
  </Step>

  <Step title="Remove profile picture">
    Click **'Remove profile picture'**.
  </Step>
</Steps>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Will changing my name affect my existing projects or agents?">
    No, changing your name will only update how your name is displayed across Relevance AI. All your projects, agents, and resources will remain intact and unchanged.
  </Accordion>

  <Accordion title="Can I change my name as many times as I want?">
    Yes, you can update your name at any time. There are no restrictions on how frequently you can change your display name.
  </Accordion>

  <Accordion title="What's the difference between changing my company here and changing my organization name?">
    The company field on your profile is personal information that doesn't affect your organization or project structure. To change your organization or project name, see [Change project and organization name](/docs/admin/project-management/change-project-and-org-name).
  </Accordion>

  <Accordion title="Will my collaborators be notified when I change my name?">
    No, collaborators will not receive a notification. They will simply see your updated name the next time they interact with you or view shared resources.
  </Accordion>

  <Accordion title="My profile picture isn't updating from Gravatar. What should I do?">
    Make sure you're using the same email address for both your Gravatar account and your Relevance AI account. You can check your Relevance AI email on the Security page. If your emails match and it's still not working, try removing your profile picture and retrieving it again.
  </Accordion>

  <Accordion title="Can I upload a profile picture directly to Relevance AI?">
    The ability to upload profile pictures directly to Relevance AI is on our roadmap and will be available in the future. Currently, you need to use Gravatar, Google, or your organization's SSO provider.
  </Accordion>

  <Accordion title="I signed up with Google but want to use a different profile picture. What can I do?">
    If you signed up with Google, your profile picture is synced from your Google account. You'll need to change your Google profile picture to update it in Relevance AI.
  </Accordion>

  <Accordion title="What if I don't want to sign up for Gravatar but want a profile picture?">
    Gravatar is a secure platform owned by Automattic (the company behind WordPress) and is trusted by millions of websites worldwide. However, if you prefer not to use Gravatar, you'll need to wait for us to add the ability to upload profile pictures directly to Relevance AI. This feature is on our roadmap and will be available in the future.
  </Accordion>

  <Accordion title="I'm having trouble updating my profile information. What should I do?">
    If you're experiencing issues updating your profile, please reach out to our support team [here](https://relevanceai.com/docs/get-started/support).
  </Accordion>
</AccordionGroup>


# Add members to your project and organization
Source: https://relevanceai.com/docs/admin/project-management/add-members

Add members to your project to collaborate with your team.

<Info>
  This page describes the standard permission system used by organizations that have not yet migrated to RBAC. If your organization has RBAC enabled, see [Role-based access controls (RBAC)](/docs/enterprise/rbac) for the current permission model and the [transition guide](/docs/enterprise/rbac#transitioning-to-rbac) for what changes during migration.
</Info>

## Add members to your project

<div>
  <iframe title="Invite a user to a project in Relevance AI" />
</div>

1. Click on `Settings` in the left-side pane of the dashboard.
2. In **Projects**, select the project you want to add members to (or create a new one for your team). If the project is collapsed, click the expand button to expand it.
3. Click on `+ Invite user` and enter the email of the member you want to add to your organization.
4. Send invite!

### User roles

Members in your organization can be assigned the following roles: Admin, Editor, Chat, and Viewer. These roles determine what they can do inside of the platform, and what they can do when using the API.

<AccordionGroup>
  <Accordion title="Admin" icon="user-tie-hair-long">
    Has all read and write permissions.
  </Accordion>

  <Accordion title="Editor" icon="typewriter">
    **Has read permissions for:**

    * All datasets
    * All knowledge sets
    * All agents
    * Users

    **Has write permissions for:**

    * All datasets
    * All knowledge sets
    * All agents

    **Other permissions:**

    * Can run agents and tools.
  </Accordion>

  <Accordion title="Chat" icon="message">
    Access [Relevance Chat](/docs/get-started/chat/introduction) only — cannot access the web app. Requires asset-level permissions to run specific agents.
  </Accordion>

  <Accordion title="Viewer" icon="book-open-reader">
    View agents, tools, and knowledge outputs only. Cannot run agents, create assets, or edit anything.
  </Accordion>
</AccordionGroup>

<Tip>
  Assign more than one admin whenever your plan allows it, so access isn't lost if an admin leaves the company or needs to hand over control unexpectedly. Free organizations are limited to a single user, who is always the admin of their organization. On Pro, you can only add two users to your organization, yourself and another. If you're the admin of your organization and there's a chance you will leave your team and need to hand over access to your fellow user, you should promote them to admin of your organization as well. On Team and Enterprise plans, keep two or more admins.
</Tip>

<Info>
  For comprehensive permission details including enterprise RBAC controls, see [Role-based access controls](/docs/enterprise/rbac).
</Info>

### Cancel pending invitations

If you've sent an invitation that hasn't been accepted yet, you can cancel it before the recipient responds. Only the person who created the invitation or a project admin can cancel pending invitations.

1. Click on `Settings` in the left-side pane of the dashboard.
2. In **Projects**, select the project with the pending invitation.
3. Find the pending invitation in the members list.
4. Click the cancel option for that invitation.

<Note>
  Once an invitation has been accepted, you'll need to [remove the member](/docs/admin/project-management/remove-members) instead of canceling the invitation.
</Note>

## Add members to your Organization

<Warning>Your region choice is permanent and cannot be changed after your organization is created. If you need to change regions, [contact support](/docs/get-started/support) to discuss your options. Learn more about [organization regions](/docs/admin/project-management/ids#organization-region).</Warning>

Once you've added someone to one of the projects in your organization, this will add them to the organization itself.

<Note>You cannot add someone to your Organization without adding them to one of your projects.</Note>

Once you've added someone to your organization, you can promote them to an admin of your Organization by going into Settings, clicking 'Organization' in the left-side panel, and then changing their organization role.

## Accepting a project invitation

When you're invited to a project, you'll receive an email with an invitation link. Here's how the acceptance process works:

### From an email invitation

1. **Click the invitation link** in your email. The link will look like: `https://app.relevanceai.com/invite/accept/{region}/{project}/{inviteCode}`

2. **Authenticate** if you're not already logged in. You'll see a loading screen that says "Processing your project invite..."

3. **Automatic acceptance** based on your account status:
   * **New user:** You'll be redirected to complete the onboarding process. Once you finish signing up, the invitation will be automatically accepted and you'll have access to the project.
   * **Already in the project:** You'll be redirected directly to the project dashboard. No further action is needed.
   * **Existing user, not in the project:** The system will automatically process your invitation and grant you access to the project.

<Tip>
  Can't see the agents after being added to a project? You might need to [switch to the correct project](/docs/admin/project-management/switch-projects).
</Tip>

### Already in the same organization

If you're already a member of the same organization as the project, you may be automatically added to the project without receiving an email invitation. In this case, you can access the project immediately by [switching to the project](/docs/admin/project-management/switch-projects).

<Note>
  When you accept a project invitation, you're also automatically added to the organization that owns the project.
</Note>


# Change the name of your project and organization
Source: https://relevanceai.com/docs/admin/project-management/change-project-and-org-name

Rename your project and organization.

## Changing the name of a project

<Info>To change the name of a project, you must be an admin of the project.</Info>

<div>
  <iframe title="Invite a user to a project in Relevance AI" />
</div>

To change the name of a project, you should do the following:

1. Click on your user icon in the top left of your workspace.
2. Click on `Manage organization`.
3. Find the project you want to change the name of and expand it.
4. Click the option menu for your project and select `Edit`.
5. Change the name or description of your project.
6. Click `Save`!

## Changing the name of your organization

<Info>To change the name of your organization, you must be an admin of the organization.</Info>

<div>
  <iframe title="Invite a user to a project in Relevance AI" />
</div>

To change the name of your organization, you should do the following:

1. Click on your user icon in the top left of your workspace.
2. Click on `Manage organization`.
3. Click `Plan & Billing`.
4. Click the pencil icon next to your organization name.
5. Change the name of your organization.
6. Click the tick icon to save your changes!

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Do I have any restrictions on the number of projects I can create?">
    Yes, the number of projects you can create is dependent on your plan. You can learn more about our plans [here](/docs/get-started/pricing).
  </Accordion>
</AccordionGroup>


# Project deletion
Source: https://relevanceai.com/docs/admin/project-management/delete-project

How to delete projects in Relevance AI

If you are an admin of your organization, or the project you wish to delete, you can delete a project by following this tutorial:

<div>
  <iframe title="Invite a user to a project in Relevance AI" />
</div>

1. Click 'Settings' in the sidebar
2. Click 'Projects'
3. Open the Project you want to delete
4. Click the three dots button, then click 'Delete'
5. On the pop-up that opens, click 'Delete Project'
6. On the next screen, type in the name of your project then click 'Delete Project' again to confirm its deletion

After that, you'll be directed to a pop-up showing that your project deletion is processing. You can navigate away from this screen if you wish.

<Warning>Project deletion is **permanent**. If you delete a project, you will lose all of the assets inside of it (Agents, Tools, Workforces, Knowledge, etc.). If you delete a project, our team cannot recover it for you.</Warning>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="How do I delete a project if I'm not an admin?">
    Only organization admins or the project admin can delete a project. If you wish to delete a project but don't have this access, reach out to your organization admin to either give you this access or do this for you.
  </Accordion>

  <Accordion title="Can I delete an organization?">
    No, you cannot delete an organization. You can only delete a project. If you wish to delete an organization, [reach out to our support team to request this](https://relevanceai.com/docs/get-started/support).
  </Accordion>
</AccordionGroup>


# Organization, project and agent IDs
Source: https://relevanceai.com/docs/admin/project-management/ids

Learn how to find your organization, project and agent IDs

You may need to know your organization, project and agent IDs for various reasons. Here's how to find them.

## URL structure

Relevance AI URLs generally follow this pattern:

```
https://app.relevanceai.com/[SECTION]/[REGION_ID]/[PROJECT_ID]/[AGENT_ID]/...
```

You can also tell what region your organization is in by looking at the URL.

* **AU**: `f1db6c`
* **EU**: `d7b62b`
* **US**: `bcbe5a`

Here is an example of a Relevance AI URL, broken down into its sections:

```
https://app.relevanceai.com/agents/f1db6c/123456789012-46bf-9cf4-a98611e97a41/43c3a067-0734-4f31-b4c9-3c72dc816519/edit/instructions
```

* **Section**: `agents`
* **Region**: `f1db6c`
* **Project ID**: `123456789012-46bf-9cf4-a98611e97a41`
* **Agent ID**: `43c3a067-0734-4f31-b4c9-3c72dc816519`

## Organization Region

Your organization's region is selected when you sign up to Relevance AI and create your organization. This determines where your data is stored and **cannot be changed after creation**.

### How to identify your region

You can identify your organization's region in two ways:

<Steps>
  <Step title="From the URL">
    Look at any Relevance AI URL while logged in. The region code appears immediately after the section name:

    ```
    https://app.relevanceai.com/agents/[REGION_CODE]/...
    ```

    The region code will be one of:

    * **US (N. Virginia)**: `bcbe5a`
    * **EU (London)**: `d7b62b`
    * **AU (Sydney)**: `f1db6c`
  </Step>

  <Step title="From the organization switcher">
    Your region is also displayed in the organization switcher in the Relevance AI interface, shown as a two-letter abbreviation (AU, EU, or US).

    <Frame>
      <img alt="Organization switcher showing region abbreviation" />
    </Frame>
  </Step>
</Steps>

### Available regions

Relevance AI operates in three regions:

<CardGroup>
  <Card title="US (N. Virginia)" icon="flag-usa">
    Region code: bcbe5a
  </Card>

  <Card title="EU (London)" icon="landmark">
    Region code: d7b62b
  </Card>

  <Card title="AU (Sydney)" icon="earth-oceania">
    Region code: f1db6c
  </Card>
</CardGroup>

<Warning>
  Your organization's region cannot be changed after creation. If you need to change regions, [contact support](/docs/support) to discuss your options.
</Warning>

## How to find each ID

### Organization ID

<Note>You'll only be able to see your organization ID if you're an organization admin. This section is not available to non-organization admins.</Note>

To find your organization ID, head to the Settings page in the sidebar of Relevance AI. The settings page will show a URL like the following below:

```
https://app.relevanceai.com/settings/organization/REGION_ID/ORG_ID/
```

From here, you should be able get your organization ID.

### Project ID

To find your project ID, head to any page in your project in Relevance AI. The URL will show a URL like the following below:

```
https://app.relevanceai.com/agents/REGION_ID/PROJECT_ID/AGENT_ID/...
```

From here, you should be able get your project ID.

### Agent ID

To find your agent ID, head to any page in your agent in Relevance AI. The URL will show a URL like the following below:

```
https://app.relevanceai.com/agents/REGION_ID/PROJECT_ID/AGENT_ID/...
```

From here, you should be able get your agent ID.

## Use a Tool in Relevance AI to grab your IDs

You can use this tool to get IDs from an Agent or Tool URL.

<Note>This Tool will not access your data or enter your project - it'll just return IDs from the URL you provide.</Note>

<iframe />

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Are IDs case sensitive?">Yes, they are.</Accordion>
  <Accordion title="Are forward slashes a valid character in an ID?">No, these are just part of the URL structure.</Accordion>

  <Accordion title="Can I change my organization's region after signup?">
    No, your organization's region is permanent and cannot be changed after creation. If you need to use a different region, you must contact support to create a new organization in the desired region.
  </Accordion>

  <Accordion title="How do I know which region my organization is in?">
    You can identify your region by looking at the region code in your URL or in the organization switcher. See the [Organization Region](#organization-region) section above for details.
  </Accordion>
</AccordionGroup>


# Remove members from your project and organization
Source: https://relevanceai.com/docs/admin/project-management/remove-members

Remove members from your Relevance AI projects and organizations

## Remove members from your project

If you're an admin, you can remove a member by doing the following:

1. Click 'Settings'
2. Click 'Projects'
3. Open the Project you want to remove a user from
4. Hover over the user you want to remove, and click 'Edit'
5. Then, click 'Remove'

This will remove the user from your project.

<div>
  <iframe title="Remove a user from a project in Relevance AI" />
</div>

## Remove members from your organization

At this time, the only way to remove users from your organization is to remove them from all projects they've been added to in your organization.

<Warning>
  If you need to remove a user who is an organization admin, you must first change their role from admin to member before you can remove them.
</Warning>

### How to remove an admin user

<Steps>
  <Step title="Change admin permissions to member">
    Navigate to the organization where the user has admin permissions. Edit their role and change it from **Admin** to **Member**.
  </Step>

  <Step title="Remove the user">
    Once the permission change is complete, you can now remove the user following the standard removal process outlined above.
  </Step>
</Steps>


# How to switch projects
Source: https://relevanceai.com/docs/admin/project-management/switch-projects

Learn how to switch between projects in both Builder and Chat

If you have access to multiple projects in Relevance AI, you can easily switch between them in both the Builder platform and Chat platform.

## Switching projects in Builder

<div>
  <iframe title="Switching projects in Builder" />
</div>

To switch projects in the Builder platform:

1. Click your **user icon** in the top left of your workspace
2. Select the project you want to switch to from the dropdown menu

When you switch projects, you'll have access to all the tools, agents, and resources available in that project.

## Switching projects in Chat

<div>
  <iframe title="Switching projects in Chat" />
</div>

To switch projects in the Chat platform:

1. Look for the **project selector** in the bottom left of the Chat interface
2. Click on it and select the project you want to switch to

<Tip>
  You can also use **Cmd + K** (Mac) or **Ctrl + K** (Windows) to quickly switch projects in Chat.
</Tip>

When you switch projects in Chat, you'll have access to all agents available in that project.


# Credit and action usage alerts
Source: https://relevanceai.com/docs/admin/project-management/usage-alerts

Set usage alerts to receive email notifications on your credit and action usage

<Info>
  Usage Alerts are separate to Usage Limits, which hard stops usage when you reach a specified limit. You can learn more about Usage Limits [here](/docs/admin/project-management/usage-limits).

  Looking for automatic top-ups instead of just notifications? Check out [Spend Controls](/docs/admin/subscriptions/spend-controls).
</Info>

<Note>Usage alerts are set per calendar month and reset at the beginning of each month.</Note>

Usage alerts allow you to receive email notifications when your credit and action usage reaches a specified limit.

## How do I set up usage alerts?

### Organization level

Organization admins can set usage alerts at the organization level by following these steps from 'Settings' > 'Plan & billing'.

<div>
  <iframe title="Set usage alerts at the organization level" />
</div>

1. Click 'Usage Alerts' in 'Plan & billing'
2. Set the number of Credits you want to be alerted for in 'Set credits alert'
3. Set the number of Actions you want to be alerted for in 'Set actions alert'
4. Enter the email(s) of the user(s) you want to notify once these limits are reached in 'Notify'
5. Click 'Save Alert Settings' to save the alert

The user(s) you've added will be notified once your organization spends the specified amount of credits or actions, across all projects in your organization.

### Project level

Project admins can set usage alerts at the project level by following these steps from 'Settings' > 'Plan & billing'.

<div>
  <iframe title="Set usage alerts at the project level" />
</div>

1. Click 'Usage Alerts' in 'Plan & billing'
2. Select 'Project' at the top to set a project-level alert
3. Set the number of Credits you want to be alerted for in 'Set credits alert'
4. Set the number of Actions you want to be alerted for in 'Set actions alert'
5. Enter the email(s) of the user(s) you want to notify once these limits are reached in 'Notify'
6. Click 'Save Alert Settings' to save the alert

The user(s) you've added will be notified once your project spends the specified amount of credits or actions.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can I set usage alerts for multiple users?">Yes, you can send usage alerts to multiple users.</Accordion>
  <Accordion title="Can I set usage alerts for multiple projects?">Yes, you can set usage alerts for multiple projects.</Accordion>
  <Accordion title="Is this feature only available on Enterprise plans?">No, you can set usage alerts on all plans.</Accordion>
</AccordionGroup>


# Credit and action usage limits
Source: https://relevanceai.com/docs/admin/project-management/usage-limits

Set usage limits to make sure your users don't exceed hard limits

<Info>Usage Limits are separate to Usage Alerts, which only send email notifications when your usage reaches a specified limit. You can learn more about Usage Alerts [here](/docs/admin/project-management/usage-alerts).</Info>

<Note>Usage limits are set per calendar month and reset at the beginning of each month.</Note>

Usage limits allow you to hard stop usage when you reach a specified limit. These limits will also send an email notification to the user(s) you've added so that they're notified that your organization has hit the limit.

Usage limits are a great way to make sure your users don't go over your Credits and Actions available on your current plan, and enable you to have more control over your usage.

## How do I set up usage limits?

### Organization level

Organization admins can set usage limits at the organization level by following these steps from 'Settings' > 'Plan & billing'.

<div>
  <iframe title="Set usage limits at the organization level" />
</div>

1. Click 'Usage Limits' in 'Plan & billing'
2. Set the number of Credits you want to be alerted for in 'Set credits alert'
3. Set the number of Actions you want to be alerted for in 'Set actions alert'
4. Enter the email(s) of the user(s) you want to notify once these limits are reached in 'Notify'
5. Click 'Save Usage Limit' to save the alert

The user(s) you've added will be notified once your organization spends the specified amount of credits or actions, across all projects in your organization.

### Project level

Project admins can set usage limits at the project level by following these steps from 'Settings' > 'Plan & billing'.

<div>
  <iframe title="Set usage limits at the project level" />
</div>

1. Click 'Usage Limits' in 'Plan & billing'
2. Select 'Project' at the top to set a project-level alert
3. Set the number of Credits you want to be alerted for in 'Set credits alert'
4. Set the number of Actions you want to be alerted for in 'Set actions alert'
5. Enter the email(s) of the user(s) you want to notify once these limits are reached in 'Notify'
6. Click 'Save Usage Limit' to save the alert

The user(s) you've added will be notified once your project spends the specified amount of credits or actions.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can I set usage limits for multiple users?">Yes, you can send usage limits to multiple users.</Accordion>
  <Accordion title="Can I set usage limits for multiple projects?">Yes, you can set usage limits for multiple projects.</Accordion>
  <Accordion title="Is this feature only available on Enterprise plans?">No, you can set usage limits on all plans.</Accordion>
</AccordionGroup>


# Security overview
Source: https://relevanceai.com/docs/admin/security

Relevance AI is SOC 2 Type II compliant and works with trusted vendors.

This page provides a security overview of our product and its key features. We are committed to ensuring the security and privacy of our users' data, and we are proud to be a SOC 2 Type II compliant company. We take your data and security seriously. Here's how we keep your information safe, private, and under your control.

## Data ownership and privacy

### You own your data

* All data you upload to Relevance AI remains your property
* We do not use your data to train our models or improve our services unless you have a specific partnership agreement with us
* Metadata may be used to improve your experience (e.g., better search), but never for model training

### Data export and deletion

* You can export your data at any time in standard formats (CSV, Excel, JSON)
* You can request account deletion, and we'll process it within 60 days
* Knowledge bases, agent logs, and files are all under your control for retention and deletion

### Data retention

#### Default retention policies

* **Agent and tool run logs**: 30 days (free tier). For other tiers, the data is stored until you choose to delete it
* **API keys**: Fully self-managed. Relevance will store certs, service accounts, API keys etc securely and encrypted

#### Enterprise Data Retention

For Enterprise customers, Relevance AI offers advanced [Data Retention](/docs/enterprise/data-retention) capabilities that allow organizations to automatically manage the lifecycle of sensitive data. This Enterprise feature enables you to:

* Configure automatic deletion policies for tool runs, agent conversations, workforce tasks, and asset versions
* Set organization-wide retention periods to meet compliance requirements
* Support regulatory frameworks like GDPR, HIPAA, and industry-specific regulations
* Reduce security risks by automatically removing old, sensitive data
* Optimize storage costs and platform performance

<Info>
  Data Retention is an Enterprise feature that must be enabled by the Relevance AI team. [Learn more about Data Retention →](/docs/enterprise/data-retention)
</Info>

## Data residency & storage

### Choose your region

Data is stored in Australia, the US, or the EU/UK, based on your selection at signup.

* US (N. Virginia)
* EU (London)
* AU (Sydney)

You can choose the region that best suits your needs when setting up your Relevance AI account.

<Note>
  Your organization's region is selected during signup and cannot be changed after your organization is created. If you need to change regions, [contact support](/docs/get-started/support) to discuss your options. Learn more about [organization regions and how to identify yours](/docs/admin/project-management/ids#organization-region).
</Note>

### Tenant isolation

* Relevance operates on a multi-tenanted architecture where customer data is logically separated
* A separate service and database exists for Enterprise customers with FGA (Fine Grained Access) controls enabled to further manage access
* Access is restricted to invited users and can be further locked down with SSO
* Single-tenant options are currently in the works

### Network security

* Relevance has network level isolation, utilizing custom VPCs and private subnets
* Restrictive firewall, rate limiting and concurrency rules are in place to protect customer data
* Our infrastructure is continuously scanned for vulnerabilities and patched within strict SLAs

### Endpoint security

* All Relevance issued hardware is pre-configured to comply with our standards of security
* Workstations are configured with encryption by default, data exfiltration prevention and lock when idle
* Up to date software is enforced to prevent malware

## Security certifications and compliance

### SOC2 Type II compliant

Relevance AI is compliant with strict enterprise-grade security and governance standards. Relevance AI is SOC2 Type II compliant, and we operate in a multi-region environment. As part of SOC2, only executive management has any kind of visibility into your customer data.

You are free to voluntarily invite our support / success team to help you check that the messaging is okay and escalations are working okay, with freedom to revoke that access at any time.

### GDPR compliant

Relevance AI is compliant with the General Data Protection Regulation (GDPR) and other relevant data protection laws. We take your data privacy seriously and ensure that all data is processed in accordance with GDPR regulations.

### Security questionnaire and documentation

We offer security questionnaire completion for Enterprise customers. We only share documentation on our compliance to Enterprise customers upon request under NDA.

### Third-party assessments

We regularly undergo third-party assessments to ensure that we are compliant with the latest security standards. These are performed by independent security firms. Reports are available to Enterprise customers under NDA.

## Encryption & key management

### Encryption everywhere

* All data is encrypted in transit and at rest using industry-standard cryptography (TLS 1.2+ for data in transit and AES 256 for data at rest)
* SOC 2 Type II policies define accepted algorithms and key management

### Customer-managed keys

* Relevance provides a mechanism to securely store keys
* Customers must manage the scope of the keys at the underlying integration

## Access management

### Authentication

* SSO and MFA are supported via your identity provider for Enterprise customers
* [Role-based access control (RBAC)](/docs/enterprise/rbac) for organizations and projects is available for Enterprise customers
* [Fine-grained access control (FGA)](/docs/enterprise/asset-controls) is available for Enterprise customers

### Administrative controls

* Fine-grained RBAC and escalation features for sensitive actions is available for Enterprise customers
* Least-privilege access by default

## AI Agent security

### Preventing data leaks and monitoring Agent behavior

* You control data sources and can pre-scrub for PII
* Human-in-the-loop escalation and monitoring features are available
* Governance Agents can be built to proactively manage Agent security
* S3 Audit events can be used to reactively manage Agent security
* [Visual Data Masking](/docs/build/agents/build-your-agent/agent-settings/visual-data-masking) protects sensitive information in the UI during screen sharing and team reviews

### Prompt injection protection

We have multiple controls, including prompt management, escalation, and parameterized inputs to protect you from prompt injection. Enterprise customers can also enable [prompt injection detection](/docs/enterprise/streaming-events#prompt-injection-detection-enterprise-feature), which scans user input for injection attempts and records them on the OTEL traces streamed to your own infrastructure.

## Vendor & supply chain security

### Third-party risk management

* All subprocessors are listed at [https://trust.relevanceai.com/subprocessors](https://trust.relevanceai.com/subprocessors)
* We conduct annual reviews and risk assessments for all vendors

## Disaster recovery & business continuity

### Resilience

* Relevance has automatic, encrypted backups across multiple availability zones
* Backups are regularly tested with failover procedures
* Relevance can operate globally and remotely to minimize disruption to services

## Self-hosted models that are multi-region

We don't train any models on your data, ever. Similarly, usually when you access API endpoints for LLMs, the processing agreement states that they don't train on that data. We take it one step further and host our own OpenAI models, and open source ones like LLaMA and Fireworks that we host within our own AWS and Azure environments which are multi-region as well.

## Key features

Relevance AI offers three key features: Tools, Agents, and Knowledge. Each feature has its own data retention and storage policies, which we will explain in detail below.

### Tools

Tools in Relevance AI are workflows that transform and process input data. It's important to note that neither the input nor the output of these tools is logged by Relevance AI. However, some steps within the tools may require the use of external vendors for processing. For example, LLM steps utilize different vendors depending on the specific model being used.

### Agents

Agents in Relevance AI enable conversations and maintain a history of interactions for the benefit of the user. These conversations are private to your project and can be deleted at any time by you. Conversations are stored securely within the same region you have chosen for your Relevance AI account.

### Knowledge

Knowledge is a feature in Relevance AI that allows you to store data in a table, enabling bulk runs of tools on entire datasets. You have full control over your stored data and can delete it at any time. Similar to agents, your data is stored securely within the region you have selected for your Relevance AI account.

### LLM models

If you provide an OpenAI API Key, it will be passed through OpenAI's API service. No data is stored or trained on during this process. You can review the Data Processing Agreement (DPA) on our website for more information.

Anthropic models are offered through their API, and once again, no data is stored or trained on during this process. You can review the DPA on our website for further details.

The below table describes the LLM models available in Relevance AI and their respective vendors when no API key is provided.

| LLM Model | Vendor    | Data Logged | Used for Training | DPA |
| --------- | --------- | ----------- | ----------------- | --- |
| GPT       | OpenAI    | No          | No                | Yes |
| Claude    | Anthropic | No          | No                | Yes |


# Plans and credits
Source: https://relevanceai.com/docs/admin/subscriptions/plans

Understand different plans at Relevance AI, and learn how to manage your subscription

## Understanding Actions

An Action is a single run of a Tool. Each time a tool runs, it counts as an action — whether it's a simple task like sending one email or running a complex workflow with many steps.

Actions are charged when you run a Tool, or when your Agent / Workforce runs a Tool. If the Tool fails, this will still count as one Action.

## Understanding Vendor Credits

Vendor Credits are the cost of running the AI model. This is the cost of the LLM, and the cost of the tools you use.

Pay only for the LLMs you use, with no markup, and credits that never expire while subscribed.

To bypass Vendor Credits entirely, you can bring your own API keys. This is only available on our paid plans and is not available to free users.

## Purchasing additional Actions and Vendor Credits

If you have a paid plan, you can purchase extra Actions and Vendor Credits to use before your next renewal if you run out.

Both Action top-ups and Vendor Credit top-ups roll over, so you won't lose them at renewal. Action top-ups roll over to your next billing cycle, and Vendor Credit top-ups roll over indefinitely while you're subscribed.

### Pricing for top-ups

* **Actions:** \$80 USD per 1,000 Actions
* **Vendor Credits:** \$20 USD per 10,000 Vendor Credits

### How to purchase additional Actions and Vendor Credits

<div>
  <iframe title="How to purchase additional Actions and Vendor Credits" />
</div>

<Info>
  You'll need to be an Organization admin to purchase additional Actions and Vendor Credits.
</Info>

To purchase additional Actions and Vendor Credits:

1. Click the Credit & Action Counter in the bottom left of the home screen
2. In the top right of the modal, click the **Buy credits** button
3. This will take you to a page where you can buy both Vendor Credits and Actions
4. Enter the amount of Vendor Credits and Actions you want to purchase:
   * Vendor Credits must be in increments of 10,000
   * Actions must be in increments of 1,000
5. Click **Next**
6. Proceed to enter your payment information

## Monitoring Actions and Vendor Credits usage

<Info>
  You can also set usage alerts to receive email notifications when your credit and action usage reaches a specified limit. [Learn more about usage alerts](/docs/admin/project-management/usage-alerts).
</Info>

### At an Organization level

To check your Actions and Vendor Credits consumption at an Organization level:

1. Click 'Settings' in the sidebar
2. Navigate to Plan & Billing

<Warning>
  You'll only have access to this section if you are an admin of your Organization.
</Warning>

<img alt="Marketing image for plan and billing" />

From this page, you will be able to see credit usage across your entire Organization, and view this by Agent or see a detailed view of your credit expenses.

### At an individual Agent level

You can also see how many Actions and Vendor Credits your Agent used, broken down by each Tool and your Agent's LLM cost.

<div>
  <iframe title="How to view the credit usage of an Agent run" />
</div>

1. First, access the run of your Agent you want to view credits on, on the Run screen
2. Click the number next to 'Credits used'
3. This will open a pop-up that will show you how much each Tool cost, as well as your Agent LLM cost and the base run cost

### How many Vendor Credits and Actions do I have left?

<img alt="Image showing the credits counter in Relevance AI, and explaining that the counter shows the number of credits remaining over the number of credits in your plan per month / year" />

You can view how many Actions and Vendor Credits you have left in the bottom left of Relevance AI. The credits counter will show you a total of the **number of Actions and Vendor Credits you have left remaining** over the **number of Actions and Vendor Credits you receive in your plan per month / year**.

The color of the counter also indicates how many credits you have remaining based on how many you have in your billing period. If your counter is green, this means you're in a good position! If the counter turns red, you should consider adding more credits if you think you're going to run out before your next renewal, or consider upgrading to a higher plan in the future.

For example, in the image above, this user is on a Pro plan, billed monthly. They currently receive 2500 Actions per month, and 3000 Vendor Credits. They have 1267 Actions and 1995 Vendor Credits left.

## Monitoring concurrency usage

Concurrency is a system capacity metric — it measures how many tasks are running simultaneously across your Organization, not how many credits or actions you consume. Each subscription tier has a specific concurrent task limit. When you reach this limit, additional tasks are queued until capacity becomes available.

<Note>
  For detailed information about concurrent task limits for each subscription tier, see the [System Quotas](/docs/admin/system-limits#concurrent-tasks) page.
</Note>

A time-series area chart displays your Organization's concurrency usage over time, helping you identify peak usage periods and determine when to upgrade for increased capacity. You can view this same chart in two places:

* On the **Plan & Billing** page: click 'Settings' in the sidebar, then open Plan & Billing.
* On the **Analytics** page, where the same chart shows organization-wide concurrency data, so you can correlate agent activity with overall concurrency patterns.

## Credit allocation and reset

For paid plans, Vendor Credits and Actions will be topped up at your next renewal date.

**Vendor Credits rollover:** All Vendor Credits roll over indefinitely while you're subscribed. This includes both the Vendor Credits included in your plan and any Vendor Credit top-ups you purchase. You'll never lose unused Vendor Credits as long as you maintain an active subscription.

**Actions rollover:** Actions included in your plan will reset to your plan's default amount at each renewal. However, any Action top-ups you purchase will roll over to your next billing cycle. This means purchased Action top-ups carry forward, but your base plan Actions reset.

<Note>
  This only applies to Free, Pro and Team plans. Enterprise customers should contact their Account Manager to check the rollover behavior for their Enterprise Organization.
</Note>

For free users, you will be given 1000 Vendor Credits when you sign up to Relevance AI, and 200 Actions a month. To access more Vendor Credits, you'll need to upgrade to a subscription.

## Paying for plans

At this time, we only offer the ability to pay for subscriptions using a credit card. We do not support debit cards at this time.

## Managing and updating payment methods

<Info>
  You'll need to be an Organization admin to manage payment methods.
</Info>

To add, update, or remove payment methods:

1. Click the **Manage plan / credits counter** section at the bottom left of the dashboard
2. Click **Manage payment methods**
3. From here, you can add a new card, update an existing payment method, or remove old payment methods

## Cancelling your subscription

You can easily cancel your subscription if you are an organization admin by following this tutorial:

<div>
  <iframe title="How to cancel your subscription in Relevance AI" />
</div>

1. Click 'Manage plan' at the bottom of the platform
2. Click 'Cancel subscription'
3. On the pop-up that opens, click 'Cancel subscription' *again* to make sure the cancellation goes through
4. You should then be directed to a pop-up confirming that you've cancelled successfully

## Invoices and receipts

Invoices and receipts for payment will be sent to the user at your company who purchased the subscription / credits at the time of purchase. You'll also be sent a receipt / invoice at every renewal.

To find your invoices and receipts in the platform:

1. Click 'Settings' while logged into Relevance AI
2. Click 'Invoice management'
3. From here, you can view and download your invoices and receipts

<Warning>
  You'll only have access to Invoice management if you are an admin of your Organization.
</Warning>

### Organization name on invoices

The organization name that appears on your invoices is the same as your organization name in Relevance AI. If you need to change the organization name on your invoices, you can [update your organization name](/docs/admin/project-management/change-project-and-org-name) in the platform.

### Updating other invoice details

For changes to other invoice details such as billing addresses or tax information, [contact our support team](/docs/get-started/support). The easiest way to do this is to forward your latest invoice along with the details you'd like updated.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What rolls over and what resets at renewal?">
    Understanding what carries over to your next billing cycle:

    Vendor Credits included in your plan and any Vendor Credit top-ups you purchase will roll over indefinitely while you're subscribed. You'll never lose unused Vendor Credits.

    Actions included in your plan will reset to your plan's default amount at renewal. However, any Action top-ups you purchase will roll over to your next billing cycle.

    This means if you purchase additional Actions or Vendor Credits as top-ups, those will carry forward, but your base plan Actions will reset to the standard amount for your subscription tier.
  </Accordion>

  <Accordion title="Can I purchase additional Actions and Vendor Credits if I run out?">
    Yes! You can purchase additional Actions and Vendor Credits if you have a subscription.

    Actions are priced at $80 USD per 1,000 Actions, and Vendor Credits are priced at $20 USD per 10,000 Vendor Credits.

    Both Action top-ups and Vendor Credit top-ups roll over. This is different from your plan's included Actions, which reset at renewal. Purchased Action top-ups will carry forward to your next billing cycle, and Vendor Credit top-ups roll over indefinitely.
  </Accordion>

  <Accordion title="Can I bring my own API keys to bypass Vendor Credits?">
    Yes! You can bring your own API keys to bypass Vendor Credits entirely. This is only available on our paid plans and is not available to free users.
  </Accordion>

  <Accordion title="I'm on old billing with Credits — what happened to the old pricing model?">
    In September 2025 we moved to a new pricing model that splits Credits into Actions (what your agents do) and Vendor Credits (AI model costs). All customers have now been transitioned to the new pricing model, so you'll see Actions and Vendor Credits in the platform rather than Credits. For a full breakdown of the changes, see our [pricing and packaging changes FAQ](/docs/admin/subscriptions/new-pricing).
  </Accordion>
</AccordionGroup>


# Spend controls
Source: https://relevanceai.com/docs/admin/subscriptions/spend-controls

Configure automatic top-up limits for Actions and Vendor Credits to ensure your account never runs out unexpectedly

<Frame>
  <img alt="Spend Controls dashboard showing automatic top-up configuration" />
</Frame>

Spend Controls allows you to configure automatic top-up limits for both your Actions and Vendor Credits, ensuring your account never runs out of credits unexpectedly.

<Note>
  Spend Controls is available to Pro and Team plan users (monthly and annual).
</Note>

## How to configure spend controls

<div>
  <iframe title="How to configure spend controls" />
</div>

1. Click on **Settings** in the sidebar, then select **Spend controls**
2. Click **Start** on either auto-recharge for **Actions** or **Vendor Credits**
3. Set your minimum threshold (the credit balance that triggers an automatic top-up when your balance falls below this limit)
4. Set your top-up amount (the amount added above your minimum threshold when a top-up occurs)

<Note>
  The same configuration process applies to both Actions and Vendor Credits. You can set up separate limits for each type of credit based on your usage patterns.
</Note>

## How top-ups work

When your balance drops below your minimum threshold, you'll automatically be charged to bring your balance up to your **minimum threshold + top-up amount**.

For example, if you have 1000 credits remaining with a minimum threshold of 10000 and a top-up amount of 1000, you'll be charged for 10000 credits to bring your balance to 11000 — not just 1000 credits.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can I set different limits for Actions and Vendor Credits?">
    Yes! Spend Controls works for both Actions and Vendor Credits independently. You can configure separate minimum thresholds and top-up amounts for each type of credit based on your specific needs and usage patterns.
  </Accordion>

  <Accordion title="What happens if my payment method fails?">
    If your payment method fails when an automatic top-up is triggered, you'll receive a notification and your account may be temporarily limited until the payment issue is resolved. Make sure to keep your payment information up to date.
  </Accordion>

  <Accordion title="When do top-ups occur?">
    Top-ups are triggered automatically when your balance falls below the minimum threshold you've set. The charge will be processed immediately to ensure uninterrupted service.
  </Accordion>

  <Accordion title="Can I modify or disable spend controls after setting it up?">
    Yes, you can modify or disable Spend Controls at any time by going to **Settings > Spend controls**. Changes take effect immediately.
  </Accordion>

  <Accordion title="Will I be notified when a top-up occurs?">
    Yes, you'll receive a notification when an automatic top-up is triggered and processed. You can also view your top-up history in your billing section.
  </Accordion>

  <Accordion title="Is there a limit to how many times I can top up?">
    There's no limit to the number of automatic top-ups. However, you should monitor your usage and adjust your settings if you're experiencing frequent top-ups to ensure you're on the right plan for your needs.
  </Accordion>

  <Accordion title="Can Enterprise customers use spend controls?">
    Spend Controls is currently designed for Pro and Team plan users. Enterprise customers have custom billing arrangements and should contact their account manager for credit management options.
  </Accordion>
</AccordionGroup>


# System quotas
Source: https://relevanceai.com/docs/admin/system-limits

Maximum system resource allocations

Relevance provides a variety of resources from space to temporarily or permanently store data to in-cloud analysis resources.

Below is a brief overview of maximum system resource allocations:

## Maximum allowed file size

| Resource                                    | Quota  |
| ------------------------------------------- | ------ |
| Upload space                                | 100 MB |
| Upload space for CSV file - number of rows  | 50 K   |
| Upload space for raw text size as knowledge | 10 MB  |

## Maximum allowed run time

| Resource                      | Quota      |
| ----------------------------- | ---------- |
| Tools (15 minutes)            | 15 minutes |
| Tools (24 hour)               | 24 hours   |
| Tools (deprecated)            | 15 minutes |
| Agents after each user action | 15 minutes |
| Bulk run                      | 24 hours   |

## Concurrent Tasks

In the context of Relevance AI, task concurrency refers to the number of tasks that can be executed simultaneously by an agent. This is an important consideration for performance, especially when dealing with webhooks, multi-agent systems, and high-volume scenarios.

Concurrency limits are determined by your subscription tier. Each tier provides different capacity for concurrent agent tasks, with higher tiers offering increased capacity to support more simultaneous operations. Enterprise plans can include custom concurrency limits tailored to your specific needs.

<Note>
  Each subscription tier includes a specific concurrent task capacity. If you need higher capacity for your use case, upgrading your subscription tier provides increased concurrent task limits. For Enterprise plans with custom requirements, please [book a demo](https://relevanceai.com/book-a-demo) with our sales team to discuss your specific needs.
</Note>

### How It Works

* The number of concurrent tasks you are allowed is determined by your subscription tier, **not** the type of trigger (webhook, API, or scheduled).
* When your tier's capacity is reached, additional tasks will be queued until resources become available.

### Monitoring Your Concurrency Usage

To monitor your real-time concurrency usage, view the concurrency usage chart in **Settings > Plan & Billing**. This time-series chart shows how many tasks are running simultaneously across your Organization, helping you identify peak usage periods and determine when you might need to upgrade for higher capacity.

[Learn more about monitoring concurrency usage](/docs/admin/subscriptions/plans#monitoring-concurrency-usage)

### Additional Resources

For a detailed comparison of subscription tiers and their respective features, please refer to our [Pricing Page](/docs/get-started/pricing).


# On-Call AI Commands
Source: https://relevanceai.com/docs/build/agents/advanced-features/on-call-ai-commands

Control your meeting agent in real-time with trigger words

Elevate your Meeting Notetaker Agent by enabling real-time commands. Control your agent during live meetings by typing keywords in the chat. The agent responds instantly, whether you need to pause recording, search for information, or flag important moments.

<iframe />

## Example commands

There are pre-built tools available for these commands in your agent's tool library.

<CardGroup>
  <Card title="Pause" icon="pause" />

  <Card title="Resume" icon="play" />

  <Card title="Search" icon="magnifying-glass" />

  <Card title="Note" icon="pen" />
</CardGroup>

| Command                         | Action                                   |
| ------------------------------- | ---------------------------------------- |
| `[trigger word] pause`          | Stop recording for sensitive discussions |
| `[trigger word] resume`         | Restart recording                        |
| `[trigger word] search [query]` | Get instant research during the call     |
| `[trigger word] note: [text]`   | Flag important items                     |

<Note>
  Configure any trigger word you want for your agent.
</Note>

## How to setup on-call AI commands

<Steps>
  <Step title="Enable interaction webhooks">
    <div>
      <iframe title="Enable Interaction Webhooks" />
    </div>
  </Step>

  <Step title="Add required tools (these are pre-built tools available in your tool library)">
    Go to your Meeting Notetaker agent's **Tools** tab and click **+ Add Tool**. Search for and add these four tools:

    1. **Resume meeting bot recording**
    2. **Pause meeting bot recording**
    3. **Google Search**
    4. **Send meeting bot message**
  </Step>

  <Step title="Configure prompt">
    Add to your agent prompt:

    ```
    You have four tools:
    - Resume meeting bot recording
    - Pause meeting bot recording
    - Google Search
    - Send meeting bot message

    Use them when requested.
    ```
  </Step>
</Steps>


# Phone Agents
Source: https://relevanceai.com/docs/build/agents/build-your-agent/agent-modes/phone-calls

Agents that can speak, listen, and take action during a call

A Phone Agent is a type of agent that can make outbound calls on your behalf. During a call, it speaks naturally with the person on the other end while also being able to take action — running tools, looking up information, and responding based on what it hears in real-time. This makes it useful for tasks like outbound sales, appointment reminders, or any workflow where a voice interaction is part of the process.

## Creating a Phone Agent

There are two ways to create a Phone Agent:

* From the **Agents** tab, click **Phone** to create one directly.
* Click **+ New Agent** from the Agents tab and select **Phone** as the agent type.

<Info>
  Your Phone Agent can use tools during a call, but doing so will pause the conversation while the tool runs. To keep calls smooth, we recommend adding your Phone Agent to a [Workforce](/docs/build/workforces/create-a-workforce) so a parent agent can complete any research or data gathering before the call starts.
</Info>

## Phone Agent settings

### Greeting

| Setting                     | Description                                                                                                                                                                                          |
| --------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| First message mode          | How the agent behaves at the start of the call. Options: agent speaks first with a model-generated message, agent speaks first with a message you write, or agent waits for the user to speak first. |
| First message               | The first thing your agent says to start the conversation.                                                                                                                                           |
| First message interruptions | When enabled, allows the user to interrupt the agent's first message.                                                                                                                                |

### Outbound phone number

<CardGroup>
  <Card title="Provided by Relevance" icon="phone">
    Pick a phone number provided by Relevance AI.
  </Card>

  <Card title="Twilio" icon="phone-arrow-up-right">
    If you have integrated Twilio in Integrations & API Keys, select a Twilio phone number.
  </Card>

  <Card title="Based on country code" icon="earth-americas">
    Automatically selects the appropriate phone number based on the country code of the number being called. Configure country mappings and set a default fallback for all others.
  </Card>
</CardGroup>

## Voice and transcription

| Setting              | Description                                                                           |
| -------------------- | ------------------------------------------------------------------------------------- |
| Voice provider       | The service powering your agent's voice on calls. Currently supports ElevenLabs only. |
| Voice model          | The voice model your agent uses when speaking.                                        |
| Agent voice          | The specific voice your agent uses when speaking.                                     |
| Voice speed          | Adjusts the speaking speed of the agent's voice.                                      |
| Background sound     | Adds a background sound to the call.                                                  |
| Transcriber provider | The service for speech recognition. Currently supports Deepgram only.                 |
| Transcriber model    | The model used for speech recognition.                                                |
| Transcriber language | The language for speech recognition.                                                  |
| Background denoising | Enables background noise reduction while the user is speaking.                        |

## Call behavior

| Setting             | Description                                                                                               |
| ------------------- | --------------------------------------------------------------------------------------------------------- |
| End call            | Allows the agent to end the call when appropriate.                                                        |
| Silence timeout     | Number of seconds of silence before ending the call. Maximum 3600 seconds.                                |
| Voicemail detection | Enables the agent to detect and respond to voicemail. Set the message it says when voicemail is detected. |
| Audio recording     | Enables recording of the phone call.                                                                      |
| Summary prompt      | Enables automatic call summary generation. Customize the prompt used to generate summaries.               |

## Call transcripts

After a call completes, the transcript is available in the call record.

| Action              | Description                                                                                            |
| ------------------- | ------------------------------------------------------------------------------------------------------ |
| Download transcript | Saves the transcript as a file to your device.                                                         |
| Copy transcript     | Copies the full transcript to your clipboard. A "Copied!" confirmation appears briefly after clicking. |
| Download recording  | Saves the audio recording of the call to your device. Only available if audio recording is enabled.    |

***

## Follow along on YouTube

### Getting started with voice agents

<Frame>
  <iframe title="YouTube video player" />
</Frame>

***

### Building a customer support Phone Agent

<Frame>
  <iframe title="Build an AI Phone Agent for Customer Support (Full Demo)" />
</Frame>


# Flow Builder
Source: https://relevanceai.com/docs/build/agents/build-your-agent/agent-settings/flow-builder

Carefully craft decisions for your Agent

<Warning>This feature is a legacy feature and is found in the Advanced Settings for only some users. If you cannot see the Flow Builder in your account, this means that the feature has been depreciated for you and you will need to use the [Prompt](/docs/build/agents/build-your-agent/prompt) to guide your Agent moving forward. We depreciated this feature at while this format worked well with older models, it was not compatible with newer models that prefer a text Prompt.</Warning>

## Overview

The Flow Builder empowers you to give your agent more precise guidance about its tasks by creating conditional logic pathways. Rather than relying solely on general instructions, you can design complex, branching decision trees that help your agent navigate different scenarios with tailored responses and actions.

With Flow Builder, you can:

* Create conditional logic using "if-then" statements
* Build branching decision trees for different conversation paths
* Provide specific instructions for various scenarios
* Enhance your agent's ability to handle complex interactions

## How Flow Builder Works

Flow Builder operates on a simple yet powerful premise: when certain conditions are met, your agent will follow specific instructions. This conditional logic framework allows your agent to adapt its responses based on the context of the conversation, user inputs, or other variables.

### Creating Decision Trees

The Flow Builder interface allows you to visually construct decision pathways that your agent will follow. Each pathway consists of:

1. A condition that triggers the pathway
2. Specific instructions for the agent to follow when that condition is met
3. Optional branching paths for further conditional responses

By creating these structured decision trees, you can ensure your agent responds appropriately to different scenarios without having to include every possible variation in your core instructions.

## When to Use Flow Builder

Flow Builder is particularly valuable in the following situations:

* **Complex Customer Support Scenarios**: Create different response paths based on the type of issue a customer reports


# General
Source: https://relevanceai.com/docs/build/agents/build-your-agent/agent-settings/general



<img />

### Suggest Replies

**What it is:** A toggle that enables automatic reply suggestions after each agent response.

**When to use it:** Enable this feature when you want to:

* Keep conversations flowing naturally
* Save time on typing common responses
* Provide users with quick response options

Simply toggle this on, and your agent will offer clickable reply suggestions after each of its messages.

***

### Agent Timeout Time

**What it is:** The maximum time your agent can work on a task before timing out.

**How to choose:**

* **Shorter timeouts:** More efficient for simple tasks, saves resources
* **Longer timeouts:** Necessary for complex reasoning, research tasks, or when working with multiple tools

Consider the complexity of your agent's typical tasks when setting this value.

***

### Instructions for Naming Tasks

**What it is:** Guidelines that help your agent create clear, consistent names for new tasks.

#### How It Works

When your agent creates a task, it uses the LLM to generate a name based on your instructions. The agent considers the full context—trigger data, workflow, and tools—to create meaningful names. These instructions act as a prompt template guiding the LLM's naming decision.

#### Writing Effective Instructions

* **Think about scanning:** Put the most important information first
* **Consider your workflow:** Align names with how you organize work
* **Be consistent:** Use the same format across similar agents
* **Include key identifiers:** Customer names, dates, action types, or outcomes
* **Test and iterate:** Run a few tasks and refine based on results

#### Examples by Use Case

<AccordionGroup>
  <Accordion title="Customer Support Agents">
    ```
    Name tasks using this format: [Customer Name] - [Request Type] - [Status]
    Example: John Smith - Refund Request - Completed
    ```
  </Accordion>

  <Accordion title="Content Creation Agents">
    ```
    Start with the content type, followed by the topic and date.
    Example: Blog Post: AI Agents Guide - 2024-11-06
    ```
  </Accordion>

  <Accordion title="Data Processing Agents">
    ```
    Include the data source and action performed.
    Example: Processed: LinkedIn Data Export - 150 records
    ```
  </Accordion>

  <Accordion title="Sales/Lead Management">
    ```
    Use format: [Company Name] - [Action] - [Outcome]
    Example: Acme Corp - Outreach Email - Sent
    ```
  </Accordion>
</AccordionGroup>

#### Best Practices

<Columns>
  <Card title="Do" icon="circle-check">
    Provide a clear format or structure for your task names, and make sure to include the most important identifying information first. Use consistent terminology that matches your workflow, and keep your instructions concise but specific. Don't forget to test with a few tasks and refine your instructions based on what works best for your needs.
  </Card>

  <Card title="Don't" icon="circle-xmark">
    Avoid making your instructions too vague or generic—the agent needs clear guidance to create useful names. Don't include variables that might not always be available in your workflow, and steer clear of complex conditional logic that could confuse the naming process. Remember to specify the order of information so your task names stay consistent.
  </Card>
</Columns>

#### Common Naming Patterns

<AccordionGroup>
  <Accordion title="Chronological">
    **Format:** `[Date] - [Action] - [Subject]`

    **Example:** "2024-11-06 - Email Sent - Product Demo"

    **Best for:** Time-sensitive workflows, audit trails, sequential processes
  </Accordion>

  <Accordion title="Hierarchical">
    **Format:** `[Project] > [Phase] > [Task]`

    **Example:** "Website Redesign > Development > API Integration"

    **Best for:** Complex projects with multiple levels, nested workflows
  </Accordion>

  <Accordion title="Status-based">
    **Format:** `[Status] [Subject] - [Details]`

    **Example:** "Completed: Customer Onboarding - Acme Corp"

    **Best for:** Task tracking, progress monitoring, filtering by completion
  </Accordion>

  <Accordion title="Actor-focused">
    **Format:** `[Who] - [What] - [When]`

    **Example:** "Sarah Johnson - Contract Review - Nov 6"

    **Best for:** Team collaboration, assignment tracking, accountability
  </Accordion>
</AccordionGroup>

***

### Guide for Using Agent

**What it is:** Simple instructions shown to users when they start a conversation with your agent.

**Best practices:**

* Keep it brief (1-3 sentences)
* Explain what the agent can help with
* Include any specific information users should provide

**Example:** "I'm your research assistant. Tell me what topic you're exploring, and I'll help find relevant information and summarize key points."

***

### Welcome Message

**What it is:** The first message your agent automatically sends at the start of every new task.

**Tips for effective welcome messages:**

* Introduce the agent's purpose
* Set clear expectations
* Include a prompt for the user to get started

**Example:** "Hi there! I'm Maya, your marketing assistant. I can help draft social posts, brainstorm campaign ideas, or analyze content performance. What would you like to work on today?"

***

### Default Prompt for Parent Agents

**What it is:** Instructions that guide how other agents should interact with this agent when working together.

**When to use it:** Configure this when building multi-agent workflows where agents need to collaborate.

This ensures consistent communication between agents and helps maintain workflow integrity.

***

### Visible in Chat

**What it is:** A toggle that controls whether the agent appears in Chat App discovery features — the @-mention menu and the "Browse agents" list. This is on by default.

**When to use it:** Turn this off when you want to:

* Hide internal or backend agents from casual discovery
* Keep API-only agents out of the Chat UI
* Maintain a cleaner agent list for end users
* Prevent confusion when an agent is designed for programmatic use only

**How to configure:**

1. Open your agent in Agent Builder
2. Navigate to **Advanced Settings** → **General**
3. Toggle **Visible in Chat** off

<Info>
  Turning off "Visible in Chat" only removes the agent from Chat discovery. Hidden agents remain fully accessible via API and direct links — this is a visibility setting, not an access control.
</Info>

***

### Visual Data Masking

**What it is:** A front-end privacy feature that automatically masks PII (email addresses, phone numbers, credit card numbers, person names) in conversation histories.

**When to use it:** Enable this when handling sensitive customer information, during screen sharing, or when multiple team members review agent conversations.

<Info>
  Visual Data Masking protects what you see in the UI while agents continue to work with actual, unmasked data. [Learn more about Visual Data Masking →](/docs/build/agents/build-your-agent/agent-settings/visual-data-masking)
</Info>

***

### Export Agent

**What it is:** A button that downloads your agent's complete configuration as a .rai file (JSON format).

**When to use it:**

* Create backups before making major changes
* Share agent configurations with team members
* Transfer agents between environments
* Create templates for similar agents

Your exported file contains all settings, tools, and configurations.

***

### Agent Autonomy Limit

**What it is:** The maximum number of consecutive tool executions your agent can perform without human approval.

<Warning>
  The autonomy limit tracks **all consecutive tool executions**, not distinct tool invocations. This means tool retries count as separate executions toward your autonomy limit.
</Warning>

#### How It Works

The autonomy limit counts every tool execution, including retries. If a tool is configured to retry on failure, each retry counts as a separate execution toward your limit. For example, if Tool A runs 3 times and Tool B runs 7 times, your agent uses 10 of its autonomy limit. Retries also count toward credit usage.

#### How to Set It

Navigate to **Agent settings** → **General** → **Agent Autonomy Limit** to configure this value.

<div>
  <iframe title="How to set Agent Autonomy Limit" />
</div>

<Tip>
  If your tools have retry settings enabled, factor those potential retries into your autonomy limit. A tool configured to retry 3 times could consume up to 3 executions of your autonomy limit if it continues to fail.
</Tip>

**Important:** Higher autonomy limits may require longer timeout settings to allow the agent to complete its work.


# Language Model
Source: https://relevanceai.com/docs/build/agents/build-your-agent/agent-settings/language-model

Access advanced settings related to the Language Model for your Agent

### Language Model

**What it is:** The AI engine that powers your agent's thinking and communication.

**How to choose:**

* **Default model:** Best for most use cases - balanced performance and cost
* **Advanced models:** May offer improved reasoning but at higher cost
* **Specialized models:** Optimized for specific tasks like coding or creative writing

<Tip>
  If you're just getting started, stick with the default model. You can always upgrade later as your needs evolve.
</Tip>

***

### Maximum output tokens

**What it is:** Override how much content an agent's model can generate as part of its reasoning, decision making, and text generation. Explicitly setting a larger limit (8,000 or higher) may be required for agents performing complex tasks with many tools and large inputs.

**How to configure:**

* **Lower limits:** More concise responses, faster performance, lower cost
* **Higher limits:** More detailed responses, but may increase processing time

Adjust based on whether you need brief updates or comprehensive explanations from your agent.

<Warning>
  Higher limits may incur more cost, and setting a value that exceeds the model's limit may cause the agent to fail.
</Warning>

***

### Temperature

**What it is:** A slider that controls how creative versus predictable your agent's responses will be.

**How to set it:**

* **Low (0-0.3):** More focused, consistent responses - ideal for factual tasks, customer support, or data analysis
* **Medium (0.4-0.7):** Balanced creativity and precision - good for general conversation
* **High (0.8-1.0):** More varied, creative responses - better for brainstorming, storytelling, or generating diverse ideas

**Example:** A sales agent might use lower temperature for explaining product specs, but higher temperature when brainstorming marketing ideas.

***

### Reasoning / Thinking

For supported models, you can configure how much the model reasons before responding. Reasoning uses output tokens to work through complex problems, which improves quality on hard tasks but is slower and costs more.

#### Thinking type and effort level

For supported Claude models, the Thinking setting has two controls: set **Thinking Type** to **Adaptive** to enable reasoning, and use **Effort Level** (Low, Medium, or High — up to Max on some models) to balance quality against speed and cost.

<img />

To reach these settings, go to **Agent → Settings → Advanced settings → Language Model**.

These controls appear on:

* Claude Opus 4.6, 4.7, and 4.8
* Claude Sonnet 4.6

<Warning>
  Thinking tokens count as output tokens, not a separate category, so enabling reasoning increases your cost. Monitor usage when you turn it on.
</Warning>

***

### Parallel Tool Calls <Badge>Beta</Badge>

**What it is:** Allows an agent to execute multiple tools or hand off multiple tasks to other agents at the same time, rather than one at a time. When enabled, the agent can issue parallel tool calls in a single step, which can significantly reduce overall execution time for workflows with independent operations.

**When to use it:**

* Running multiple independent operations at once (for example, querying several data sources in a single step)
* Gathering information from multiple sources simultaneously
* Reducing overall execution time when individual tool calls don't depend on each other's results
* Parallel agent workflows in Workforces where sub-agents can operate independently

**How to configure:**

1. Open your agent and click **Advanced** in the bottom-left of the agent builder
2. In Advanced settings, go to the **Language Model** tab
3. Toggle on **Parallel Tool Calls**
4. For agent-to-agent handovers in the Workforce Builder: select the connection, open **Edge Settings**, and set **Task Behavior** to **Create new task**

<Note>
  If the agent is part of a Workforce, set all of its outbound connections to use **Create new task** as their Task Behavior to get the parallel speedup. The Workforce Builder highlights conflicting connections in yellow as a reminder. See [Edge Settings](/docs/build/workforces/build-an-ai-workforce/edge-settings) for details on configuring Task Behavior for Workforce connections.
</Note>

***

### Fallback model

**What it is:** A backup language model that automatically retries failed tasks when your primary model encounters errors or fails to respond. If your agent's primary model fails to complete a task, the fallback model will automatically retry the task **one time**. If the fallback model also fails, the task will fail as usual and follow your configured error handling behavior.

**When to use it:**

* Using models that occasionally have reliability issues (e.g., Gemini)
* Running critical workflows where you need higher reliability
* You want to reduce task failures due to temporary LLM provider issues

**How to configure it:**

1. Open your agent
2. Click **Advanced** in the bottom-left of the agent builder
3. In Advanced settings, go to the **Language Model** tab
4. Under **Fallback model**, select a model to rerun failed tasks on

<Tip>
  Choose a fallback model from a **different provider** than your primary model. For example, if you're using Gemini as your primary model, select an OpenAI model (like GPT-5) as your fallback. This eliminates the chance of another provider-specific error and maximizes reliability.
</Tip>


# Subagents
Source: https://relevanceai.com/docs/build/agents/build-your-agent/agent-settings/subagents

Add Subagents to your Agent 

<img />

<Warning>
  This feature is a legacy feature and may be deprecated soon. If you still have access to this feature, you can find this in Advanced Settings. Otherwise, if you cannot see this feature in your account, you will not be able to use it moving forward. If you wish to build a multi-agent system, we recommend using the [Workforce](/docs/build/workforces/create-a-workforce) feature instead.
</Warning>

## Sub-Agent Settings

### Approval Mode

Choose how your parent Agent gets permission to use this Sub-agent:

* **Auto Run**: The Sub-agent runs automatically whenever needed, with no interruptions
* **Approval Required**: You'll be asked for permission each time before the Sub-agent runs

**When to use each option:**

* Choose **Auto Run** for routine, low-risk tasks you're comfortable automating completely
* Choose **Approval Required** when you need oversight for sensitive operations or actions with significant impact

### Max Auto-runs Before Approval

Set a limit on consecutive automatic runs before requiring your approval:

* **No limit**: The Sub-agent can run as many times as needed without stopping for approval
* **Custom number**: After reaching this number of runs within one task, the system will pause and ask for your permission

This setting helps you maintain control over resource usage and repetitive actions while still allowing automation where appropriate.

### Error Handling Behavior

Decide how errors should be managed when they occur:

* **Use Default Settings**: Follow your workspace's standard error handling rules
* **Use Custom Settings**: Create specific error handling instructions just for this Sub-agent

Custom settings are particularly useful when a Sub-agent needs special treatment for retries, escalations, or alternative workflows when something goes wrong.

## Communication Settings

### Usage Instructions

Write clear directions for when and how the parent Agent should use this Sub-agent:

```
Example: "Use this Research Sub-agent whenever you need to gather competitive information or market data. Provide specific search terms and the type of information needed."
```

Good instructions ensure the parent Agent knows exactly when to call on this Sub-agent and what information to provide.

### Message Template

Create a standardized format for requests sent to the Sub-agent:

```
Example: "Please research {{topic}} and focus on {{specific_aspect}}. The information will be used for {{purpose}}."
```

Variables like `{{topic}}` will be filled in by the parent Agent when making requests. This creates consistency and ensures all necessary context is included with each request.

## Task View Settings

### View Title

Name your custom dashboard view with a clear, descriptive title:

```
Examples: "Customer Support Interactions" or "Document Processing Tasks"
```

A good title helps you quickly identify and switch between different monitoring views.

### Information to Include

Select which tools or actions you want to track in this view:

* Search for specific tools (like "Send Email" or "Generate Report")
* Only tasks using these selected tools will appear in this view
* This filtering helps you focus on the specific workflows you need to monitor

**Pro tip:** Create separate views for different workflows to avoid information overload and make monitoring more efficient.

## Why Task Views Matter

Task Views help you organize and monitor complex Agent workflows by:

* Separating different types of work into focused dashboards
* Reducing noise from unrelated activities
* Making it easier to track progress on specific processes
* Allowing different team members to monitor the aspects most relevant to them


# Give Your Agent Tasks
Source: https://relevanceai.com/docs/build/agents/build-your-agent/agent-settings/task-view



<img />

## Task View Settings

### View Title

Enter a clear, descriptive name for your custom Task View. This title will appear in the Task View selector, making it easy to find and switch between your different task dashboards.

**Examples:**

* "Customer Support Tasks"
* "Sales Lead Follow-ups"
* "Content Approval Queue"

**Best Practice:** Choose a name that instantly communicates what type of tasks this view will display.

### Which Information to Include in the View

Use this search box to select specific tools or information sources you want to monitor in this view. Your custom view will only display tasks that involve the tools you've selected here.

**How it works:**

1. Type the name of a tool your Agent uses (e.g., "Send Email", "Extract Data", "Escalate to Slack")
2. Select the relevant tools from the dropdown results
3. Your view will automatically filter to show only tasks using these tools

**Why this matters:** By focusing on specific tools, you can create specialized dashboards that show only the task activity relevant to particular workflows or responsibilities.

### When to use custom Task Views

Task Views are particularly valuable when:

* Your Agent handles multiple types of tasks using different tools
* You need to monitor specific workflows separately
* You want to reduce clutter and focus only on certain types of Agent activity
* Different team members need to track different aspects of your Agent's work

By creating targeted views, you can efficiently monitor exactly what matters most to you without being overwhelmed by unrelated task activity.


# Visual Data Masking
Source: https://relevanceai.com/docs/build/agents/build-your-agent/agent-settings/visual-data-masking

Protect sensitive information in the UI while agents continue to work with actual data

<Warning>
  Visual Data Masking is an Enterprise-only feature currently in limited beta. It is not available on Free, Pro, or Team plans. To have this feature enabled for your organization, contact your account manager or [reach out to support](/docs/get-started/support).
</Warning>

<Warning>
  The AI model used for PII detection can occasionally miss sensitive information. Do not rely on Visual Data Masking as your sole privacy control for highly sensitive data. This is a known limitation of the beta.
</Warning>

Visual Data Masking is a front-end privacy feature that automatically masks standard PII (Personally Identifiable Information) fields in conversation histories and task views. This protects confidential information when viewing agent interactions in the UI, during screen sharing, or when reviewing tasks with your team.

## How Visual Data Masking works

Visual Data Masking operates at the UI layer only. When enabled, sensitive information is masked in what you see on screen, but agents continue to access and process the actual, unmasked data.

<Columns>
  <Card title="What you see" icon="eye-slash">
    Masked values like `***@***.com` or `***-***-1234` in conversation histories
  </Card>

  <Card title="What agents see" icon="robot">
    Full, unmasked data for processing, reasoning, and tool execution
  </Card>
</Columns>

This approach ensures privacy during UI viewing while maintaining full agent functionality. The backend data remains completely unchanged—Visual Data Masking only affects the display layer.

## When to use Visual Data Masking

Enable Visual Data Masking when you need to protect sensitive information from being visible in the UI:

<AccordionGroup>
  <Accordion title="Handling customer data" icon="users">
    When agents process customer information like email addresses, phone numbers, or payment details, Visual Data Masking prevents this data from being visible to team members reviewing conversations.
  </Accordion>

  <Accordion title="Screen sharing and demos" icon="screen-users">
    During product demonstrations, training sessions, or customer calls where you're sharing your screen, Visual Data Masking ensures sensitive information isn't accidentally exposed.
  </Accordion>

  <Accordion title="Team reviews and QA" icon="clipboard-check">
    When multiple team members review agent conversations for quality assurance or training purposes, Visual Data Masking protects customer privacy while allowing evaluation of agent performance.
  </Accordion>

  <Accordion title="Compliance requirements" icon="shield-check">
    For organizations with regulatory requirements around data visibility, Visual Data Masking adds an extra layer of protection when viewing agent interactions. However, it should be used alongside other privacy controls, not as the sole protection mechanism.
  </Accordion>
</AccordionGroup>

## Supported PII types

Visual Data Masking automatically detects and masks the following types of sensitive information:

| PII type            | Example input         | Masked output         |
| ------------------- | --------------------- | --------------------- |
| Email addresses     | `user@example.com`    | `***@***.com`         |
| Phone numbers       | `(555) 123-4567`      | `***-***-1234`        |
| Credit card numbers | `4532 1234 5678 9010` | `****-****-****-9010` |
| Person names        | `John Smith`          | `**** *****`          |

Detection happens automatically in real-time as conversations are displayed in the UI using AI-powered pattern recognition.

## Configuration options

Visual Data Masking is a simple toggle feature with **no advanced configuration options**. When enabled:

* **All four PII types are automatically masked** - You cannot select which types to mask
* **Masking format is fixed** - You cannot customize the placeholder text or masking pattern
* **No detection sensitivity controls** - You cannot adjust the AI detection threshold or rules
* **No custom PII types** - You cannot add organization-specific sensitive data patterns
* **Same rules for all data** - All detected PII is masked using the same approach

This simplicity makes Visual Data Masking easy to enable, but limits flexibility for organizations with specific masking requirements.

## Enabling Visual Data Masking

### For agents

To enable Visual Data Masking on an agent:

<Steps>
  <Step title="Open agent settings">
    Navigate to your agent in the Agent Builder.
  </Step>

  <Step title="Access Advanced settings">
    Click **Advanced** in the bottom-left navigation panel.
  </Step>

  <Step title="Open the General tab">
    In the Advanced settings panel, click the **General** tab.
  </Step>

  <Step title="Enable Visual Data Masking">
    Toggle on **Visual Data Masking** to enable the feature for this agent.
  </Step>

  <Step title="Save changes">
    Your changes are saved automatically. Visual Data Masking will now apply to all conversation histories for this agent.
  </Step>
</Steps>

### For workforces

<Warning>
  If any agent in a workforce has Visual Data Masking enabled, the entire workforce will mask PII across all task views and conversation histories — not just that individual agent. Enable this feature at the agent level only when you intend this cascading behavior.
</Warning>

To enable Visual Data Masking on a workforce:

<Steps>
  <Step title="Open workforce settings">
    Navigate to your workforce and click the **Settings** icon in the top right.
  </Step>

  <Step title="Access Advanced settings">
    Click **Advanced** in the settings panel.
  </Step>

  <Step title="Enable Visual Data Masking">
    Toggle on **Visual Data Masking** to enable the feature for this workforce.
  </Step>

  <Step title="Save changes">
    Your changes are saved automatically. Visual Data Masking will now apply to all task views and conversation histories for this workforce.
  </Step>
</Steps>

## What is and isn't masked

### What is masked

Visual Data Masking applies to these UI elements:

<CardGroup>
  <Card title="Agent conversation histories" icon="messages">
    Conversations shown in the Task View
  </Card>

  <Card title="Workforce task displays" icon="list-check">
    Task views showing agent interactions
  </Card>

  <Card title="Message content" icon="comment">
    Messages displayed in conversation logs
  </Card>

  <Card title="Tool inputs and outputs" icon="wrench">
    Tool data visible in the UI conversation flow
  </Card>

  <Card title="Front-end conversation displays" icon="display">
    Anywhere conversation data appears in the Relevance AI platform UI
  </Card>
</CardGroup>

### What isn't masked

Visual Data Masking does NOT apply to:

<CardGroup>
  <Card title="Backend data storage" icon="database">
    Data is stored unmasked in the database
  </Card>

  <Card title="Agent processing" icon="robot">
    Agents always work with unmasked, actual data
  </Card>

  <Card title="Tool execution" icon="gear">
    Tools receive and process unmasked data
  </Card>

  <Card title="LLM requests" icon="brain">
    Language models receive unmasked data for processing
  </Card>

  <Card title="API responses" icon="code">
    Data returned via API is unmasked
  </Card>

  <Card title="Backend logs" icon="file-lines">
    Internal system logs contain unmasked data
  </Card>

  <Card title="Exported message content" icon="file-export">
    Conversation message bodies in data exports are unmasked. The one exception is the CSV conversation export, which masks the agent name ("Run By") column when Visual Data Masking is enabled
  </Card>

  <Card title="Event streaming" icon="cloud-arrow-up">
    OTEL events exported to S3 are unmasked, except structured fields like email and IP that are always scrubbed. Unstructured content is scrubbed only when PII Redaction is enabled
  </Card>
</CardGroup>

<Warning>
  Visual Data Masking is a UI-only feature. It does not modify, encrypt, or redact data at the storage or processing level. Backend data remains completely unchanged. For backend data protection, see [PII Redaction](#visual-data-masking-vs-pii-redaction).
</Warning>

## Visual Data Masking vs PII Redaction

Visual Data Masking and PII Redaction are **complementary features** that protect sensitive information at different stages. They serve different purposes and are designed to work together for comprehensive privacy protection.

| Feature               | Visual Data Masking                      | PII Redaction                                |
| --------------------- | ---------------------------------------- | -------------------------------------------- |
| **Availability**      | Enterprise only (limited beta)           | Enterprise only                              |
| **Where it applies**  | Front-end UI display                     | Backend data export to S3                    |
| **What it does**      | Visual masking only                      | Data is scrubbed before export               |
| **Data modification** | No - backend data unchanged              | Yes - PII removed from exports               |
| **Use case**          | UI privacy, screen sharing, team reviews | Compliance, audit logs, observability        |
| **Applies to**        | Task View, conversation UI               | S3 event streaming (OTEL)                    |
| **Configuration**     | Toggle only (no options)                 | Advanced (entity types, actions, thresholds) |
| **Detection method**  | AI detection (can miss PII)              | Presidio-powered ML engine                   |
| **Agent access**      | Agents see unmasked data                 | Agents see unmasked data                     |

### Complementary use

These features are designed to work together, not replace each other:

<Columns>
  <Card title="Visual Data Masking" icon="eye-slash">
    Protects privacy when humans view agent conversations in the UI. Use for screen sharing, team collaboration, and internal reviews.
  </Card>

  <Card title="PII Redaction" icon="database">
    Protects data exported to your observability infrastructure. Use for compliance, audit logs, and external data analysis.
  </Card>
</Columns>

**Recommended approach:** Enable both features for comprehensive protection:

* **Visual Data Masking** (Enterprise) protects what your team sees in the platform UI
* **PII Redaction** (Enterprise) protects what gets exported to your S3 bucket for compliance

Together, these features provide defense in depth: UI privacy for internal collaboration and backend data protection for external systems.

<Info>
  Learn more about [PII Redaction for Event Streaming](/docs/enterprise/streaming-events#pii-redaction-enterprise-feature).
</Info>

## Best practices

<Tabs>
  <Tab title="When to enable">
    ### Enable Visual Data Masking when

    * Your agents handle customer PII regularly
    * Multiple team members review agent conversations
    * You frequently share your screen during demos or training
    * You need an extra layer of privacy protection in the UI
    * Compliance policies require limiting PII visibility

    ### Consider not enabling when

    * You need to see actual values for debugging or troubleshooting
    * Your agents don't process sensitive information
    * You're the only person accessing the agent
    * Masked values would make conversation review difficult

    ### Important limitations to consider

    * **Do not rely solely on Visual Data Masking** for critical privacy requirements
    * **Test with sample data** to understand detection accuracy for your use case
    * **Combine with PII Redaction** (Enterprise) for comprehensive privacy protection
    * **Remember AI detection can miss PII** - this is not a foolproof solution
  </Tab>

  <Tab title="Combining with other features">
    ### Layer your security

    Visual Data Masking works best as part of a comprehensive security strategy:

    * **[PII Redaction](/docs/enterprise/streaming-events#pii-redaction-enterprise-feature)** (Enterprise) - Protect exported audit logs and telemetry data
    * **[User Level Authentication](/docs/enterprise/user-level-authentication)** - Ensure users only access their own data in integrations
    * **[RBAC](/docs/enterprise/rbac)** - Control who can access agents and view conversations
    * **[Data Retention](/docs/enterprise/data-retention)** - Automatically delete old sensitive data

    Each feature protects data at a different layer, creating defense in depth.
  </Tab>

  <Tab title="Testing and validation">
    ### Verify masking behavior

    Before rolling out Visual Data Masking to production agents:

    1. **Test with sample data** - Run test conversations with known PII to verify masking works as expected
    2. **Check detection accuracy** - Verify that the AI model detects the PII formats you use
    3. **Test edge cases** - Try international phone numbers, unusual email formats, and context-specific data
    4. **Check all views** - Verify masking appears in task views, conversation histories, and workforce views
    5. **Confirm agent functionality** - Ensure agents can still process data correctly despite UI masking
    6. **Review with your team** - Have team members confirm masked values are sufficiently protected

    ### Monitor for edge cases

    * Some PII formats may not be detected (e.g., international phone numbers)
    * Custom data formats may require additional protection
    * Context-specific sensitive information may not be automatically masked
    * AI detection can occasionally miss PII that matches standard patterns
  </Tab>
</Tabs>

## Beta limitations

As an Enterprise limited beta feature, Visual Data Masking has current limitations that users should understand:

<AccordionGroup>
  <Accordion title="AI detection can occasionally miss PII">
    The AI model used for PII detection can occasionally miss sensitive information. Detection accuracy varies based on data format, context, and PII type. This is a known limitation of the beta. Do not rely on Visual Data Masking as your sole privacy control for highly sensitive data.
  </Accordion>

  <Accordion title="No configuration options">
    Visual Data Masking is a simple toggle with no advanced configuration. You cannot choose which PII types to mask, customize the masking format, adjust detection sensitivity, or add custom PII patterns. All four supported PII types are automatically masked when enabled.
  </Accordion>

  <Accordion title="Fixed masking format">
    The masking format (e.g., `***@***.com`) is fixed and cannot be customized. Future versions may offer configurable masking patterns, but currently all PII is masked using the same placeholder style.
  </Accordion>

  <Accordion title="Limited PII type support">
    Only four standard PII types are supported: email addresses, phone numbers, credit card numbers, and person names. Custom sensitive data types specific to your organization cannot be configured for masking.
  </Accordion>

  <Accordion title="Visual only - no data modification">
    Visual Data Masking only affects the UI display. Backend data storage, agent processing, API responses, and data exports all contain unmasked data. This is by design to ensure agents can function properly, but means the feature provides no backend data protection.
  </Accordion>

  <Accordion title="Performance impact on long conversations">
    On very long conversations with extensive PII, there may be a slight delay in rendering masked content. This is typically imperceptible but may be noticeable in extreme cases with thousands of messages.
  </Accordion>
</AccordionGroup>

<Info>
  We're actively improving Visual Data Masking based on user feedback. If you encounter issues or have feature requests, contact your sales representative or [reach out to support](/docs/get-started/support).
</Info>

## Related features

<CardGroup>
  <Card title="Security overview" icon="shield-halved" href="/admin/security">
    Learn about Relevance AI's comprehensive security features and compliance
  </Card>

  <Card title="PII Redaction" icon="database" href="/enterprise/streaming-events#pii-redaction-enterprise-feature">
    Protect sensitive data in exported OTEL logs and audit trails (Enterprise)
  </Card>

  <Card title="User Level Authentication" icon="user-lock" href="/enterprise/user-level-authentication">
    Enable users to authenticate with their own credentials for integrations
  </Card>

  <Card title="Workforce Task View" icon="list-check" href="/build/workforces/workforce-features/workforce-task-view">
    Monitor and manage workforce tasks with Visual Data Masking enabled
  </Card>
</CardGroup>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Does Visual Data Masking affect agent performance?">
    No. Visual Data Masking only affects what you see in the UI. Agents continue to process unmasked data with no performance impact. The masking happens during UI rendering, not during agent execution.
  </Accordion>

  <Accordion title="How reliable is the AI detection?">
    The AI model can occasionally miss PII, especially with non-standard formats, international data, or context-specific sensitive information. Detection accuracy is optimized for standard US formats. Do not rely on Visual Data Masking as your sole privacy control for highly sensitive data. Always test with your specific data formats to understand detection accuracy.
  </Accordion>

  <Accordion title="Can I customize which PII types are masked?">
    No. Visual Data Masking is a simple toggle feature with no configuration options. All four supported PII types (email, phone, credit card, person names) are automatically masked when enabled. You cannot select specific types or add custom patterns.
  </Accordion>

  <Accordion title="Can I customize the masking format?">
    No. The masking format is fixed (e.g., `***@***.com` for emails). You cannot customize the placeholder text or masking pattern. This may be available in future releases.
  </Accordion>

  <Accordion title="Will masked data appear in API responses?">
    No. Visual Data Masking only applies to the UI. Data returned via API calls is unmasked. If you need to protect data in API responses, implement masking in your application layer.
  </Accordion>

  <Accordion title="Can I temporarily disable masking to see actual values?">
    Yes. You can toggle Visual Data Masking off in agent or workforce settings at any time. Changes take effect immediately for new page loads.
  </Accordion>

  <Accordion title="Does Visual Data Masking work with embedded chat?">
    Visual Data Masking applies to the Relevance AI platform UI only. If you've embedded agents on external websites, Visual Data Masking does not apply to those embedded interfaces.
  </Accordion>

  <Accordion title="How is this different from PII Redaction?">
    Visual Data Masking protects what you see in the UI (front-end only). PII Redaction protects what gets exported to your S3 bucket for audit logs and observability (backend data scrubbing). Both are Enterprise features: Visual Data Masking is a simple toggle currently in limited beta; PII Redaction has advanced configuration options. Use both together for comprehensive protection.
  </Accordion>

  <Accordion title="What happens to data that's already in conversation history?">
    When you enable Visual Data Masking, it applies to all conversation histories immediately, including past conversations. The masking is applied during rendering, so historical data is protected retroactively in the UI. However, the backend data remains unchanged.
  </Accordion>

  <Accordion title="Can I export unmasked data for analysis?">
    Yes. Visual Data Masking primarily affects the UI display, so API responses and backend storage always contain unmasked data, and conversation message content in exports is unmasked. The one exception is the CSV conversation export, which masks the agent name ("Run By") column when Visual Data Masking is enabled. If you need broader masking of exported data, implement masking in your export pipeline or use PII Redaction (Enterprise) for S3 event streaming.
  </Accordion>

  <Accordion title="Does this protect data sent to LLMs?">
    No. Visual Data Masking is UI-only. Agents send unmasked data to LLMs for processing. This is necessary for agents to function correctly. If you need to protect data sent to LLMs, you must implement pre-processing to scrub data before it reaches the agent.
  </Accordion>
</AccordionGroup>


# API Triggers
Source: https://relevanceai.com/docs/build/agents/build-your-agent/agent-triggers/api-trigger

Trigger your Agent with direct HTTP requests

API Triggers let you trigger your Agent programmatically with direct HTTP requests to the Relevance AI API. Use them to integrate from custom backends, automation scripts, or any environment that can make HTTP calls.

<Callout icon="code">
  For a higher-level interface in JavaScript or TypeScript, use the [Relevance AI SDK](/docs/sdk/introduction) — it wraps this endpoint with type-safe primitives, automatic event handling, and runtime support across Node.js, Deno, Bun, Cloudflare Workers, and the browser.
</Callout>

<Warning>
  The Relevance AI API is officially supported only for triggering Agents and Tools. All other usage is currently unsupported.
</Warning>

## Setting up an API trigger

1. Navigate to the Agents page and select the agent you want to configure
2. In the sidebar on the left, go to **Triggers** and click the **Add trigger** button
3. Under the **Build your own triggers** section, click the **API** trigger

The API trigger configuration popup displays the endpoint, a button to generate your API key, and a pre-filled request body with your `agent_id`. Code examples in cURL, JavaScript, and Python are also provided.

<Warning>
  Save your API key somewhere safe — once generated, you won't be able to view it again. You can regenerate one at any time.
</Warning>

## API reference

**Endpoint:**

```http theme={null}
POST https://api-f1db6c.stack.tryrelevance.com/latest/agents/trigger
```

<Info>
  The endpoint URL differs based on your organization's region. The example above shows the AU region endpoint. When you set up your API trigger in the UI, the correct endpoint for your region will be displayed in the configuration popup.
</Info>

**Request body:**

```json theme={null}
{
  "message": {
    "role": "user",
    "content": "Hello"
  },
  "agent_id": "[YOUR_AGENT_ID_GOES_HERE]"
}
```

<Note>
  The response isn't returned directly. The agent's result must be sent to a destination you configure, such as Google Sheets, a webhook, or another integration.
</Note>

## Code examples

**cURL**

```bash theme={null}
curl -X POST \
  -H "Content-Type: application/json" \
  -H "Authorization: [YOUR_API_KEY]" \
  -d '{"message":{"role":"user","content":"Hello"},"agent_id":"[YOUR_AGENT_ID_GOES_HERE]"}' \
  https://api-f1db6c.stack.tryrelevance.com/latest/agents/trigger
```

**JavaScript (Fetch)**

```js theme={null}
fetch('https://api-f1db6c.stack.tryrelevance.com/latest/agents/trigger', {
  method: "POST",
  headers: {
    "Content-Type": "application/json",
    "Authorization": "[YOUR_API_KEY]"
  },
  body: JSON.stringify({
    message: { role: "user", content: "Hello" },
    agent_id: "[YOUR_AGENT_ID_GOES_HERE]"
  })
});
```

**Python (requests)**

```python theme={null}
import requests
import json

requests.post(
  'https://api-f1db6c.stack.tryrelevance.com/latest/agents/trigger',
  headers={
    "Content-Type": "application/json",
    "Authorization": "[YOUR_API_KEY]"
  },
  data=json.dumps({
    "message": { "role": "user", "content": "Hello" },
    "agent_id": "[YOUR_AGENT_ID_GOES_HERE]"
  })
)
```

## Best practices

* **Error handling**: implement error handling in your integration code
* **Rate limiting**: be mindful of API rate limits and throttle accordingly
* **Security**: store API keys in environment variables rather than hardcoding them
* **Monitoring**: set up monitoring for your trigger usage to detect issues early
* **Testing**: test your implementation in a development environment before deploying to production

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What programming languages can I use?">
    The API trigger works with any language capable of making HTTP requests. For JavaScript and TypeScript, the [Relevance AI SDK](/docs/sdk/introduction) provides a higher-level interface with built-in authentication and event handling.
  </Accordion>

  <Accordion title="Are there rate limits?">
    Yes, usage is subject to rate limits based on your subscription plan. Refer to your plan details for specific limitations.
  </Accordion>

  <Accordion title="Can I use this with serverless functions?">
    Yes — the API works with serverless architectures like AWS Lambda, Google Cloud Functions, and Azure Functions.
  </Accordion>

  <Accordion title="How secure is the API trigger?">
    All requests must be authenticated with your API key, and all data is transmitted over HTTPS.
  </Accordion>
</AccordionGroup>


# Webhook Triggers
Source: https://relevanceai.com/docs/build/agents/build-your-agent/agent-triggers/custom-webhook

Connect to a custom webhook to trigger your Agent

Webhook Triggers let you trigger your Agent via HTTP requests from any external system — no native integration required. When your webhook receives a request, it sends the payload to your Agent as a message.

### Setting up a Webhook Trigger

1. Navigate to your Agent's settings and select the **Triggers** section
2. Choose **Webhook Trigger** from the available options
3. Copy your unique webhook URL
4. Configure your webhook settings:
   * **Message template**: Use template variables with double brackets (e.g., `{{first_name}}`) to format incoming data, or use `{{$}}` to pass the entire payload
   * **Thread ID mapping**: Map a field from your payload to a Thread ID to keep related messages in the same conversation
   * **Unique ID mapping**: Map a field to prevent duplicate processing of the same event

### Common use cases

* Process form submissions from your website
* Trigger responses when new support tickets are created
* Connect third-party platforms that support outgoing webhooks
* Build automated workflows between different systems

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can I use webhook triggers with any external system?">
    Yes, any system capable of sending HTTP requests can integrate with Relevance AI's webhook triggers, making them extremely versatile for various integration scenarios.
  </Accordion>

  <Accordion title="How secure are webhook triggers?">
    Webhook triggers use unique URLs that are difficult to guess. For additional security, you can implement Unique ID mapping to prevent duplicate processing and validate incoming requests.
  </Accordion>

  <Accordion title="What happens if the same webhook is triggered multiple times?">
    By mapping a Unique ID field, you can prevent duplicate processing. If an event with the same Unique ID is received again, it will be ignored, ensuring data integrity.
  </Accordion>
</AccordionGroup>


# Integration Triggers
Source: https://relevanceai.com/docs/build/agents/build-your-agent/agent-triggers/integrations

Trigger your Agents from 1,000+ integrations

Relevance AI allows you to trigger your Agents from 1,000+ integrations. Not only can you integrate your favourite products and platforms to your Agents with Tools, you can also trigger your Agents from your favourite integrations.

For each trigger, connect to the integration using your associated credentials. For more information about setting up specific triggers, refer to the integration-specific documentation.

### Premium Triggers

The Premium triggers below are available on the Pro plan and above.

<Columns>
  <Card title="LinkedIn" href="/integrations/popular-integrations/linkedin#setting-up-triggers" icon="linkedin">Initiate your AI agents based on specific LinkedIn events</Card>
  <Card title="WhatsApp" href="/integrations/popular-integrations/whatsapp#setting-up-triggers" icon="whatsapp">Use WhatsApp (personal) to trigger your Agent</Card>
</Columns>

<Columns>
  <Card title="WhatsApp for Business" href="/integrations/popular-integrations/whatsapp-for-business#setting-up-triggers" icon="whatsapp">Use WhatsApp for Business to trigger your Agent</Card>
  <Card title="Telegram" href="/integrations/popular-integrations/telegram#setting-up-triggers" icon="telegram">Trigger your Agents from Telegram messages</Card>
</Columns>

### External Triggers

External triggers like **Zapier** are configured and managed outside of Relevance AI. These allow you to connect your Agent to third-party automation platforms without any setup within Relevance AI itself.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="How do I connect an integration trigger?">
    Navigate to your agent's Triggers section, select the integration you want to connect, and authenticate with your credentials for that service.
  </Accordion>

  <Accordion title="Are premium triggers included in my plan?">
    Premium triggers (LinkedIn, WhatsApp, WhatsApp for Business, and Telegram) are available on the Pro plan and above.
  </Accordion>
</AccordionGroup>


# Scheduled Triggers
Source: https://relevanceai.com/docs/build/agents/build-your-agent/agent-triggers/scheduled-triggers

Set your Agent to run on a schedule or send messages at specific times

Scheduled triggers let your Agent act on its own based on time — either by running tasks on a recurring schedule or by sending messages at specific future times. Use them to automate reports, follow-ups, reminders, and any workflow that needs to happen without manual input.

## Recurring schedules

<img />

Recurring schedules run your Agent automatically at set intervals — daily, weekly, monthly, or on a custom cadence.

### Setting up a recurring schedule

1. Go to your agent in the Relevance AI dashboard
2. Click 'Edit Agent'
3. Navigate to the 'Triggers' section
4. Select 'Recurring Schedule'
5. Choose your schedule type

### Schedule types

#### Simple schedule options

| Schedule Type | What It Does                        | Perfect For                          |
| ------------- | ----------------------------------- | ------------------------------------ |
| **Daily**     | Runs every day at a time you choose | Morning reports, daily check-ins     |
| **Weekly**    | Runs on specific days each week     | Team updates, weekly newsletters     |
| **Monthly**   | Runs on chosen dates each month     | Monthly reports, regular maintenance |
| **Custom**    | Runs on specific dates you select   | One-off tasks or irregular schedules |

#### Advanced: cron expressions

For more precise control over scheduling, you can use **Cron expressions**. Relevance AI uses **AWS EventBridge cron format**, which requires **six fields**:

```
minute hour day-of-month month day-of-week year
```

**Cron Expression Format:**

| Field            | Values          | Wildcards      |
| ---------------- | --------------- | -------------- |
| **Minute**       | 0-59            | , - \* /       |
| **Hour**         | 0-23            | , - \* /       |
| **Day-of-month** | 1-31            | , - \* ? / L W |
| **Month**        | 1-12 or JAN-DEC | , - \* /       |
| **Day-of-week**  | 1-7 or SUN-SAT  | , - \* ? L #   |
| **Year**         | 1970-2199       | , - \* /       |

**Special Characters:**

* `*` (asterisk) - all values
* `?` (question mark) - no specific value (used in day-of-month or day-of-week)
* `-` (dash) - range of values
* `,` (comma) - additional values
* `/` (forward slash) - increments
* `L` - last day of month or week
* `W` - nearest weekday
* `#` - nth occurrence of a day in a month

**Common Cron Expression Examples:**

```
0 10 * * ? *        # Every day at 10:00 AM
0 0 * * MON *       # Every Monday at midnight
0 9 1 * ? *         # First day of every month at 9:00 AM
0 0 * * MON-FRI *   # Weekdays at midnight
*/15 * * * ? *      # Every 15 minutes
0 9,17 * * ? *      # Daily at 9:00 AM and 5:00 PM
0 0 1,15 * ? *      # 1st and 15th of every month at midnight
0 8 ? * 2 *         # Every Monday (day 2) at 8:00 AM
0 10 L * ? *        # Last day of every month at 10:00 AM
0 9 ? * 2#1 *       # First Monday of every month at 9:00 AM
```

<Note>
  AWS EventBridge cron expressions use 6 fields, unlike traditional Unix cron which uses 5 fields. Always include the year field, and use `?` for either day-of-month or day-of-week when specifying the other.
</Note>

### Tips

* **Start small**: Begin with one simple recurring task before setting up complex schedules
* **Test thoroughly**: Run your agent manually first to ensure it works as expected
* **Monitor initially**: Check the first few automated runs to confirm everything works correctly
* **Use cron for precision**: When you need exact timing control, use cron expressions instead of simple schedules

***

## Scheduled messages

<img alt="" />

Scheduled messages let your Agent send messages at specific future times — follow-ups, reminders, check-ins, or any time-based outreach. You can edit or reschedule these messages directly in Task View.

### Setting up Scheduled Messages

1. Navigate to the **Build** page
2. Look for the **Triggers** section
3. Scroll to the bottom and turn on the **Enabled Scheduled Messages** toggle
4. Add instructions in your agent's core instructions about when to schedule messages

Once turned on, your agent will understand time-based requests like "send a follow-up tomorrow at 9 AM" or "remind me about this next Tuesday."

You can guide your agent's scheduling behaviour by adding instructions to your agent's **Prompt** like:

```
When someone asks about upcoming events, schedule a reminder 24 hours before each event.

If someone has a deadline, schedule a check-in halfway to the deadline.

For multi-step processes, schedule messages at each milestone to keep people on track.
```

### Scheduled messages settings

#### Enabled Scheduled Messages (toggle)

Turn this on to allow your Agent to schedule actions or messages for the future. Without this enabled, your Agent can only respond in real-time and won't be able to set future reminders, follow-ups, or scheduled notifications.

#### Max daily Scheduled Triggers

Set a limit on how many scheduled actions your Agent can process in a single day. Leaving it at "No limit" means the Agent can schedule and send as many future messages as needed. If you want to control costs, manage system load, or prevent overactivity, you can set a specific maximum number.

#### Prevent Scheduled Messages from interrupting a Task

When this is enabled, your Agent won't interrupt an active task (like when it's waiting for approvals or processing another action) just to send a scheduled message. Instead, the scheduled message will wait and be delayed by at least the time you specify (default is 1,440 minutes, or 24 hours).

This prevents your Agent from getting confused, stuck, or behaving unpredictably when juggling multiple actions.

#### Schedule Message Tool

A built-in tool your Agent can use when writing prompts. Use the `/schedule message` shortcut inside your Agent's instructions to tell it exactly when and what to schedule automatically during conversations.

**Example:**

> "After summarizing today's emails, schedule a check-in message tomorrow at 10 AM."

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What if my agent encounters an error during a scheduled run?">
    The system logs errors and can send notifications based on your settings. You can view error details in your agent's task history.
  </Accordion>

  <Accordion title="Can I temporarily pause a recurring schedule?">
    Yes, you can disable schedules without deleting them when you need a pause. Simply toggle off the schedule in your agent's Triggers settings.
  </Accordion>

  <Accordion title="How do I change an existing schedule?">
    Go back to the Recurring Schedule section in your agent's Triggers settings and update the schedule as needed. Changes take effect immediately.
  </Accordion>

  <Accordion title="What timezone are scheduled runs based on?">
    Scheduled runs use your account's configured timezone. You can adjust this in your agent's schedule settings to ensure tasks run at the correct local time.
  </Accordion>
</AccordionGroup>


# Tools as Triggers
Source: https://relevanceai.com/docs/build/agents/build-your-agent/agent-triggers/tools-as-triggers

Use a Tool as a trigger for your Agent

Tools as Triggers let you use the output of a Tool to trigger your Agent. This is useful when you need to combine data from multiple sources — like Salesforce and Airtable — to filter and process only the most relevant records before your Agent acts on them.

<Note>The Tool you use as a Trigger must return an **array** as its output.</Note>

### Setting up a Tool Trigger

1. Navigate to your Agent's **Triggers** section and select **Tool Trigger**
2. **Name your Trigger** — give it a clear name so you can identify it later
3. **Select the Tool** you want to use as a trigger
4. Click **Continue**
5. **Run a test** of the Tool — this produces a sample output so you can configure what gets sent to your Agent
6. **Select the output field** from the Tool's results that you want to use as the Agent's input
7. Optionally, enable **deduping** (see below)
8. **Set a schedule** — choose how often the Tool runs (daily, weekly, etc.) and the time
9. Optionally, configure **queue work hours** if your plan supports it
10. Click **Setup** to finish

### Deduping

Deduping prevents duplicate messages from being added to your Agent's queue. When enabled, you specify a field to use for deduplication.

For example, if your Tool output includes a unique field like `_id`, messages with the same value in that field won't be added to the queue more than once — regardless of whether the rest of the data has changed.

### Common use cases

* Pull new leads from a CRM and filter by criteria before triggering outreach
* Combine data from multiple platforms into a single, filtered trigger
* Process only records that meet specific conditions using custom Tool logic

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can I change the Tool's inputs for each run?">
    No. You can only set default values for the Tool's inputs, which will be used every time the Trigger fires. Dynamic inputs per run are not supported.
  </Accordion>
</AccordionGroup>


# Alerts
Source: https://relevanceai.com/docs/build/agents/build-your-agent/alerts

Set up rules to get notified when your agent encounters an issue, reaches a threshold, or gets updated.

Set up rules to get notified when your agent encounters an issue, reaches a threshold, or gets updated.

<Note>
  Alerts are available on paid plans only.
</Note>

## Alert rules

Stay on top of agent issues by creating alert rules that notify you automatically when specific events occur. Each rule has a **trigger condition** (what happened) and one or more **notification channels** (where to send it).

### Alert types

<CardGroup>
  <Card title="Status change" icon="rotate">
    Triggers when an agent task enters one or more specific statuses you select.
  </Card>

  <Card title="Tool error" icon="wrench">
    Triggers when a tool fails during execution, optionally scoped to specific tools.
  </Card>

  <Card title="Agent published" icon="rocket">
    Triggers when the agent is published. No additional configuration needed.
  </Card>
</CardGroup>

#### Status change

Triggers when an agent task enters one or more specific statuses. Select which statuses you want to be alerted on from the multi-select dropdown.

<Accordion title="Available statuses">
  Starting up, Waiting for capacity, Pending approval, Running, Cancelled, Idle, Timed out, Escalated, Unrecoverable error, Paused, Completed, Exhausted retries, Queued for approval, and Queued for rerun.
</Accordion>

<Tip>
  A common configuration is to alert on failure states like **Timed out**, **Unrecoverable error**, and **Exhausted retries** so your team can respond quickly when something goes wrong.
</Tip>

#### Tool error

Triggers when a tool fails during execution. You can optionally scope this to specific tools — leave the selection empty to watch all tools, or pick individual tools to monitor. This is useful for critical integrations where a tool failure could block your agent's workflow.

#### Agent published

Triggers when the agent is published. No additional configuration is needed — select a notification channel and you'll be alerted whenever a new version goes live.

### Notification channels

Each alert rule can send notifications to one or more channels:

<CardGroup>
  <Card title="Email" icon="envelope">
    Add one or more email addresses. No OAuth connection required.
  </Card>

  <Card title="Slack" icon="slack">
    Connect your Slack workspace via OAuth, then select a channel.
  </Card>

  <Card title="Microsoft Teams" icon="microsoft">
    Connect your Microsoft Teams account via OAuth, then select a team and channel.
  </Card>
</CardGroup>

For Slack setup details, see [Slack integration](/docs/integrations/popular-integrations/slack). For Teams, see [Microsoft Teams integration](/docs/integrations/popular-integrations/microsoft-teams).

### Setting up an alert rule

1. Navigate to your agent and open the **Alerts** tab
2. Click **Create alert rule**
3. Select a trigger type from the left panel
4. Configure the trigger conditions (if applicable)
5. Preview how the notification will look
6. Add at least one notification channel
7. Click **Save**

You can create multiple alert rules, each with different triggers and channels. Existing rules appear as cards showing the trigger description, active channels, and controls to edit, delete, or toggle them on/off.

***

## Agent escalations

When your agent encounters a situation it cannot handle, it can escalate the conversation to a human manager using the built-in **Escalate to Manager** tool.

This is different from alert rules — escalation is an in-conversation handoff where the agent actively routes a conversation to a human, rather than a background notification about an event.

### Enabling the escalation tool

In the **Agent escalations** section at the bottom of the Alerts page, click **Enable escalation tool**. This adds the "Escalate to Manager" tool to your agent's available tools.

Once enabled, you can reference the tool in your agent's prompt using the **/** command to give the agent guidance on when to escalate. Click **Edit prompt** to go directly to the prompt editor.

### When to use escalation

<CardGroup>
  <Card title="Complex technical support" icon="headset">
    Issues requiring specialized knowledge beyond the agent's training.
  </Card>

  <Card title="Sensitive situations" icon="hand-holding-heart">
    Customer service scenarios that need human empathy and judgment.
  </Card>

  <Card title="Product feedback" icon="comment">
    Detailed feedback that should reach product managers directly.
  </Card>

  <Card title="High-value sales" icon="handshake">
    Opportunities requiring personalized human attention.
  </Card>
</CardGroup>


# Evals
Source: https://relevanceai.com/docs/build/agents/build-your-agent/evals

Test and evaluate your AI Agents with scenario-based evaluations and reusable Checks

<Info>
  **Rollout Status**: Evals is rolling out progressively, starting with Enterprise customers. If you don't see this feature in your account yet, reach out to your account manager to discuss access.
</Info>

The Evals section is your command center for testing and evaluating AI Agent performance. Located in the **Evaluate** tab (next to the Build and Use tabs) in the Agent builder, Evals lets you create test sets, define reusable Checks, run automated evaluations, and monitor live Agent quality — all without manual testing. Evals apply to both individual Agents and Workforces, including the sub-agents and tools inside a Workforce.

<div>
  <iframe title="Meet Evals — set the bar for your AI Agents" />
</div>

## What you can do with Evals

<CardGroup>
  <Card title="Run tests" icon="flask-vial">
    Build test sets with scenarios that simulate real user interactions, then attach Checks to score every conversation automatically.
  </Card>

  <Card title="Reuse Checks" icon="scale-balanced">
    Define evaluation criteria once in the Checks tab and attach them to scenarios, Monitor dashboards, or ad-hoc evaluations of completed tasks.
  </Card>

  <Card title="Monitor live tasks" icon="chart-line">
    Create Monitor dashboards that score live Agent tasks against your Checks, with sample-rate controls and per-Check trend charts over time.
  </Card>
</CardGroup>

***

## Why use Evals?

Any time we are using agents we want to be sure that they are accomplishing our business outcomes. This might be increasing Marketing Qualified Leads, or reducing the time it takes to respond to a lead. Evals give you the confidence that your agents are hitting the mark with their target audience.

On top of that, Evals enables testing agents across a spectrum of LLMs in order to optimize costs while maintaining a high degree of effectiveness. Fine tuning agent cost effectiveness can have an immediate impact on a company's bottom line. And, because you can see that agents can run effectively on lower cost models, you can broaden the scope of where you can deploy agents. You can now have agents focus on additional business outcomes without worrying about costs as much.

If you have questions on how to track value effectively with Evals, please [contact Relevance](/docs/get-started/support).

***

## Evals sections

The Evals area has five sections, shown in the left sidebar of the Evaluate tab:

* **Test** — Create and manage test sets. Each test set holds scenarios that simulate users; running a scenario produces a conversation with your Agent that gets scored by attached Checks.
* **Runs** — Past evaluation run results. Browse average scores, tasks evaluated, progress status, cost (Credits and Actions), and creation date for every run.
* **Checks** — The reusable set of evaluation criteria. Create a Check once, then attach it to scenarios, to Monitor dashboards, or to one-off evaluations of completed tasks.
* **Publish** — Choose which test sets must pass before your Agent can be published. Set a minimum pass rate and optionally block publishing on failure.
* **Monitor** — Track live Agent quality on real tasks. Create one or more Monitor dashboards, attach Checks, set a sample rate, and watch scores trend over time.

***

## Understanding Checks

Checks are the reusable evaluation criteria that score Agent conversations. You create a Check once in the **Checks** tab and then attach it wherever you need it:

* **To a scenario** in a test set — the Check runs every time that scenario is evaluated.
* **To a Monitor dashboard** — the Check runs on a sampled portion of live Agent tasks.
* **To a one-off evaluation** of already-completed tasks selected from the Agent's task list.

The Checks tab has filters that show where each Check is currently used — **All checks**, **Scenarios**, **Dashboard**, and **Unused** — so you can quickly find Checks that aren't attached anywhere yet.

### Check types

When creating a Check, you choose one of the following types:

<AccordionGroup>
  <Accordion title="LLM Judge" icon="brain-circuit">
    Uses an LLM to evaluate conversations against a prompt you define.

    | Field                           | Description                                                                                                                                                                                                                                                                                                                                                                            |
    | ------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
    | **Evaluation Prompt**           | Describe the criteria for passing                                                                                                                                                                                                                                                                                                                                                      |
    | **Judge model**                 | Select which model evaluates the conversation                                                                                                                                                                                                                                                                                                                                          |
    | **Truncate long conversations** | When enabled, conversations that exceed the judge model's context window are trimmed from the oldest messages first, and the eval runs on the remaining portion. When disabled, oversized conversations fail with an error instead. Note that trimming removes early context, which can affect score accuracy if your evaluation criteria depend on the beginning of the conversation. |
  </Accordion>

  <Accordion title="Text Includes" icon="text">
    Checks whether the Agent's response includes specific text.

    | Field             | Description                               |
    | ----------------- | ----------------------------------------- |
    | **Required text** | The text that must appear in the response |
  </Accordion>

  <Accordion title="Text Equals" icon="equals">
    Checks whether the Agent's response exactly matches an expected value.

    | Field              | Description                                  |
    | ------------------ | -------------------------------------------- |
    | **Expected value** | The exact message the Agent should have sent |
  </Accordion>

  <Accordion title="Tool Usage" icon="screwdriver-wrench">
    Checks whether a specific tool was used during the conversation.

    | Field          | Description                                                      |
    | -------------- | ---------------------------------------------------------------- |
    | **Tool**       | Select the tool to check for                                     |
    | **Position**   | Whether the tool was used anywhere, used first, or used last     |
    | **Comparison** | Check if the tool was used at least, exactly, or at most X times |

    When evaluating a Workforce, you can scope a Tool Usage Check to a specific node — a sub-agent or tool in the Workforce — so you can assert that a particular sub-agent used a given tool.
  </Accordion>
</AccordionGroup>

To create a Check from the Checks tab:

<div>
  <iframe title="Creating a Check" />
</div>

1. Go to the **Evaluate** tab and select **Checks** from the left sidebar.
2. Click **+ New Check**.
3. Select a **Type** (LLM Judge, Text Includes, Text Equals, or Tool Usage).
4. Enter a **Name** for the Check (e.g., "Professional tone").
5. Configure the type-specific settings (see table above).
6. Click **Create Check**.

<Note>
  Checks attached to a scenario are always included when you run that scenario. Additional Checks from the Checks tab are not auto-included — select the ones you want under **Additional global checks** in the run modal (Run Test Set, Run Scenario, or Evaluate Selected Tasks) before kicking off the run.
</Note>

***

## Creating a test set with a scenario

<div>
  <iframe title="Creating a test set" />
</div>

Follow these steps to create your first test set:

1. Open your Agent in the builder and click the **Evaluate** tab. Select **Test** from the left sidebar.

2. Click the **+ New test set** button. Enter a name for your test set and click **Create**.

3. Click on the test set you just created to open it.

4. Click the **+ Add scenario** button to add a scenario to your test set.

5. Fill in the scenario details:

   | Field                         | Description                                                                                                    | Example                                                                   |
   | ----------------------------- | -------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------- |
   | **Scenario name**             | A descriptive name for this scenario                                                                           | "Response empathy"                                                        |
   | **Scenario description**      | Describe a persona and situation — the AI generates realistic messages from this                               | "You are an impatient customer who wants quick answers about their bill." |
   | **Run X times**               | How many times to execute this scenario                                                                        | 3                                                                         |
   | **Up to X messages**          | Maximum conversation length, where each message is one back-and-forth between the simulated user and the Agent | 10                                                                        |
   | **+ Set exact first message** | Optional — pin the simulated user's opening message instead of letting the AI generate it                      | "Hi, I need help with my bill."                                           |

6. Attach Checks to define how this scenario is scored. You can either pick existing Checks from the Checks tab or create new ones inline:

   | Field                    | Description                                                         | Example                                                                                                                    |
   | ------------------------ | ------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------- |
   | **Type**                 | The Check type                                                      | LLM Judge                                                                                                                  |
   | **Name**                 | Name of the evaluation criterion                                    | "Empathy shown"                                                                                                            |
   | **Type-specific config** | Settings based on the chosen type (see [Check types](#check-types)) | *Evaluation Prompt*: "Did the Agent acknowledge the customer's frustration and express empathy before offering solutions?" |

   Newly created Checks land in the Checks tab and can be reused on other scenarios or Monitor dashboards.

7. (Optional) Add **Tool simulations** to emulate Tool usage without actually calling the underlying Tools. Tool simulations are configured per scenario:

   * Select a Tool to simulate.
   * Provide a prompt describing what the Tool should return (a fake response is generated based on your prompt).
   * In the **Advanced** dropdown, you can select a **Simulation model** to control which model generates the simulated response.

   When evaluating a Workforce, Tool simulations can be configured per node, so each sub-agent uses its own simulated Tool responses.

8. Click **Save test scenario** to save your configuration.

<Tip>
  You can add multiple scenarios to a single test set to evaluate different aspects of your Agent's behavior. Each scenario can have its own description, message cap, run count, attached Checks, and Tool simulations.
</Tip>

### Managing scenarios

Scenarios can be reorganized across test sets as your testing strategy evolves. Each scenario has a dropdown menu (the three-dot icon next to the scenario name) with three operations:

| Operation     | What it does                                            | When to use it                                            |
| ------------- | ------------------------------------------------------- | --------------------------------------------------------- |
| **Move**      | Relocates the scenario to another test set              | Reorganizing test sets or consolidating related scenarios |
| **Copy**      | Creates a duplicate of the scenario in another test set | Reusing a scenario as a baseline in a different test set  |
| **Duplicate** | Creates a copy of the scenario in the same test set     | Quickly creating a variation of an existing scenario      |

### Example scenarios

Here are some example scenarios you might create:

<AccordionGroup>
  <Accordion title="Customer support - empathy test">
    **Scenario name**: Response empathy

    **Description**: You are a long-time customer who was recently charged twice for the same order. You've already contacted support once without resolution and are feeling frustrated but willing to give the Agent a chance to help. Express your concerns clearly and see if the Agent acknowledges your situation before jumping to solutions.

    **Up to**: 10 messages

    **Check**: Empathy shown (LLM Judge)

    * *Evaluation Prompt*: Did the Agent acknowledge the customer's frustration and express empathy before offering solutions? The response should show understanding of the emotional state and validate their concerns.
  </Accordion>

  <Accordion title="Sales - product knowledge test">
    **Scenario name**: Product expertise

    **Description**: You are a procurement manager at a mid-sized company evaluating solutions for your team. You need specific details about enterprise pricing tiers, integration capabilities with existing tools like Salesforce and HubSpot, and data security certifications. Ask clarifying questions and compare features against competitors you're also considering.

    **Up to**: 15 messages

    **Check**: Accurate information (LLM Judge)

    * *Evaluation Prompt*: Did the Agent provide accurate product information without making claims that cannot be verified? Responses should be factual, reference actual product capabilities, and acknowledge when information needs to be confirmed by a sales representative.
  </Accordion>

  <Accordion title="Support - escalation handling">
    **Scenario name**: Escalation request

    **Description**: You are a paying customer who has experienced a service outage affecting your business operations. You've already troubleshooted with the knowledge base articles and need to speak with a senior support engineer or account manager. Be firm but professional in your request, and provide context about the business impact.

    **Up to**: 5 messages

    **Check**: Appropriate escalation (LLM Judge)

    * *Evaluation Prompt*: Did the Agent acknowledge the severity of the situation, validate the customer's need for escalation, and initiate a handoff to a human representative while maintaining a professional and empathetic tone throughout?
  </Accordion>
</AccordionGroup>

***

## Running evaluations

You can run an entire test set or an individual scenario from within a test set by clicking the **Run** button on either.

You can select specific scenarios within a test set to run a subset at once, or run all scenarios in the test set together. Note that you cannot bulk-select and run multiple test sets at the same time.

1. Enter a name for the run (e.g., "Scenario run - Jan 14, 12:14 PM"). A default name with timestamp is provided.
2. Checks already attached to the scenarios are always included. To add Checks from the Checks tab, select the ones you want under **Additional global checks**.
3. Click **Run** to begin. The simulator generates conversations with your Agent based on your scenario prompts and the selected Checks score each conversation.

***

## Understanding results

After running an evaluation, you'll see a detailed results screen:

### Run summary

The top of the results page shows key metrics:

| Metric            | Description                                       |
| ----------------- | ------------------------------------------------- |
| **Average Score** | Overall pass rate across all scenarios and Checks |
| **Tasks**         | How many Agent tasks were evaluated               |
| **Agent Version** | The version of the Agent that was tested          |

### Scenario results

Each scenario displays:

| Column       | Description                                                                |
| ------------ | -------------------------------------------------------------------------- |
| **Status**   | Running, Completed, or Failed                                              |
| **Name**     | The scenario name                                                          |
| **Score**    | Percentage of Checks that passed (shown with progress bar)                 |
| **Checks**   | Pass/fail count (e.g., "1/1 passed")                                       |
| **Credits**  | Credits consumed for this scenario                                         |
| **Actions**  | Actions consumed for this scenario — each Check plus the agent's tool runs |
| **Run time** | How long the scenario took to complete                                     |

### Viewing conversation details

Click **View Conversation** on any scenario to see:

1. **The full conversation** between the simulated user and your Agent.
2. **Check verdicts** from every Check included in the run, with detailed explanations of why each Check passed or failed.

For example, an "Empathy shown" Check might show:

> **Pass**: The Agent demonstrated strong empathy throughout the conversation. Key examples include: acknowledging the customer's frustration with being transferred multiple times ("I completely understand how upsetting it must be to feel like you're not getting the help you need"), validating her experience with the double charge ("I truly understand how frustrating it is to be charged twice"), and directly addressing her skepticism by saying "I completely understand your concerns, especially given your previous experience."

***

## Monitor

The **Monitor** section continuously scores live Agent tasks against Checks from the Checks tab. Unlike Test, which runs simulated conversations, Monitor evaluates the real conversations your Agent is having.

Monitor is organized into **dashboards** — you can create more than one (for example, one focused on tone, another on tool-use accuracy) and configure each independently.

### Creating a Monitor dashboard

1. Go to the **Evaluate** tab and select **Monitor** from the left sidebar.
2. Click **+ New dashboard** and give it a name.
3. Attach one or more Checks from the Checks tab.
4. Set a **Sample rate** — the percentage of incoming tasks to evaluate.
5. (Optional) Set a **Filter tasks** option to control which task statuses trigger evaluations. Leave blank to evaluate all tasks.
6. Save the dashboard.

Once configured, qualifying tasks are automatically scored at the sample rate you've set.

### Viewing dashboard insights

Each Monitor dashboard shows:

| Metric            | Description                                                           |
| ----------------- | --------------------------------------------------------------------- |
| **Overall score** | Aggregate score across all evaluated tasks in the selected date range |
| **Total runs**    | Number of tasks evaluated                                             |
| **Checks**        | Which Checks are attached to the dashboard                            |

You also get:

* **Overall score timeseries** to spot regressions or improvements over time.
* **Per-Check charts** so you can see which criteria are slipping.
* **Version markers** that line up score changes with Agent publishes.
* **A list of evaluation runs** with score, name, and a drill-in to the full conversation.

<Tip>
  To adjust dashboard settings after initial setup, click the **Settings** button in the top right corner of the dashboard.
</Tip>

***

## Publish

The **Publish** section lets you choose which test sets must pass before your Agent can be published. If the results don't meet your minimum pass rate, publishing can be blocked.

You can configure Publish from the **Publish** section in the Evaluate tab.

### Test sets to run

Select which test sets to run before publishing. Click **Add test sets** to choose them — all scenarios in the selected test sets will be evaluated.

### Publish settings

Configure how evaluations affect the publish process:

| Setting                                 | Description                                                                                                                                                                                                   |
| --------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Minimum pass rate (%)**               | The minimum score percentage required for the evaluation to pass (e.g., 100%)                                                                                                                                 |
| **Allow publishing even if eval fails** | When unchecked (the default), the Agent will only be published if the evaluation score meets or exceeds the minimum pass rate. When checked, the Agent publishes regardless of whether the evaluation passes. |

Once configured, click **Save**. When you next publish your Agent, the selected test sets will run automatically and the results will be checked against your minimum pass rate.

***

## Cost and billing

<img alt="Evaluation run results showing the Credits, Actions, and Run time columns for each scenario" />

Evaluations are billed in credits and actions. Each scenario reports what it consumed:

| Column       | What it shows                                                                                                                                                                                                                                                                                     |
| ------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Credits**  | Vendor credits consumed by the scenario — the user-simulation LLM calls, the agent's own run, and any LLM-based Checks. Click the value to open the full breakdown.                                                                                                                               |
| **Actions**  | Actions consumed by the scenario — one for each Check, plus the actions from the agent's own tool runs during the conversation. The evaluation run itself isn't charged a separate action. This column only appears on usage-based plans; on legacy plans the action cost is included in Credits. |
| **Run time** | The wall-clock duration of the scenario's conversation, measured from its first message to its last. A timing metric, not a charge, and shown per scenario only (the Average Score row doesn't total it).                                                                                         |

Clicking a **Credits** or **Actions** value opens a breakdown of where that scenario's cost went. It separates the user simulation, the agent's own run, and the Checks that scored the conversation, so you can see which part drives the cost — usually the agent's own execution. The breakdown is split across three components:

<img alt="Cost breakdown modal splitting a scenario's credits across Scenario Runner, Agent Execution, and Checks" />

| Component           | What it covers                                                                                                                                                                                                                                                                 |
| ------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| **Scenario Runner** | The LLM calls used to simulate the user. Uses credits only.                                                                                                                                                                                                                    |
| **Agent Execution** | The agent's own run during the test. Uses credits, plus the actions from each tool the agent runs. These actions were already billed when the agent ran — they're shown here for attribution, not charged again. Workforces have no action-level data, so only credits appear. |
| **Checks**          | Each Check that scores the conversation, broken down by model. Each Check uses credits and counts as one action.                                                                                                                                                               |

***

## Best practices

<CardGroup>
  <Card title="Start simple" icon="seedling">
    Begin with a few core scenarios that test your Agent's primary use cases. Add complexity as you learn what matters most.
  </Card>

  <Card title="Be specific with Checks" icon="bullseye">
    Write detailed Check prompts. Vague criteria lead to inconsistent scoring. Include specific examples of what passing looks like.
  </Card>

  <Card title="Test edge cases" icon="triangle-exclamation">
    Create scenarios for difficult situations: angry customers, off-topic requests, requests to bypass rules, etc.
  </Card>

  <Card title="Run Monitor on live tasks" icon="chart-line">
    Stand up a Monitor dashboard with your most important Checks so you catch regressions on real conversations, not just simulated ones.
  </Card>

  <Card title="Keep your Checks tab tidy" icon="folder-tree">
    Use the **Unused** filter to clean up Checks that aren't attached anywhere. Group related scenarios into dedicated test sets and reorganize with Move, Copy, and Duplicate as your strategy evolves.
  </Card>

  <Card title="Sample wisely in Monitor" icon="percent">
    Match the sample rate to your task volume. A low-traffic Agent can run at 100%; high-volume Agents can sample lower to keep costs in check without losing the signal.
  </Card>
</CardGroup>

***

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="How many scenarios can I have in a test set?">
    You can add as many scenarios as needed to a single test set. Each scenario is evaluated independently and can have its own attached Checks.
  </Accordion>

  <Accordion title="How many Checks can I add to a scenario?">
    Each scenario supports up to 10 Checks. This applies to scenario-level Checks defined within the scenario itself. Checks added via **Additional global checks** at run time are counted separately.
  </Accordion>

  <Accordion title="How are evaluations billed?">
    Evaluations consume both Actions and Vendor Credits. Each Check costs 1 Action, while Vendor Credits cover the LLM costs of the task run, the user simulator, and any LLM-based Check. Each scenario shows its full breakdown in the results — see [Cost and billing](#cost-and-billing) for details.
  </Accordion>

  <Accordion title="Can I rerun a previous evaluation?">
    Yes, you can run the same scenarios again at any time. Each run is saved in your Runs history, allowing you to compare results across different Agent versions.
  </Accordion>

  <Accordion title="Where do my Checks live?">
    All Checks live in the **Checks** tab under Evaluate. From there you can attach them to scenarios (for evaluation runs), to Monitor dashboards (for live tasks), or to one-off evaluations of completed tasks. The **Scenarios**, **Dashboard**, and **Unused** filters show where each Check is currently attached.
  </Accordion>

  <Accordion title="Can the LLM Judge evaluate long conversations?">
    Yes, with configuration. The LLM Judge Check includes a **Truncate long conversations** toggle in the **Advanced** section when creating a Check. When enabled, conversations that exceed the judge model's context window are trimmed and evaluated. When disabled, those conversations fail with an error rather than producing a partial result.
  </Accordion>

  <Accordion title="What happens when a conversation is truncated?">
    The oldest messages are removed from the start of the conversation until it fits within the judge model's context window. The judge is notified that truncation occurred and evaluates the remaining portion. If your evaluation criteria depend on early context — such as the user's original request or instructions given at the start of the conversation — the result may be less accurate. In those cases, disabling truncation and selecting a model with a larger context window is preferable.
  </Accordion>

  <Accordion title="Can I move scenarios between test sets?">
    Yes. Each scenario has a dropdown menu (three-dot icon) with three options: **Move** relocates the scenario to another test set, **Copy** creates a duplicate in another test set, and **Duplicate** creates a copy in the same test set.
  </Accordion>

  <Accordion title="I don't see the Evals section. How do I get access?">
    Evals is rolling out progressively, starting with Enterprise customers. If you don't see the Evaluate tab in the Agent builder, reach out to your account manager to discuss access.
  </Accordion>
</AccordionGroup>


# Memory
Source: https://relevanceai.com/docs/build/agents/build-your-agent/memory

Extract and recall metadata from your Agent's task output

<img alt="" />

## Overview

Memory in Relevance AI allows your agent to remember and recall information during and across conversations. There are two types of memory that serve different purposes:

<Columns>
  <Card title="Short-Term Memory" icon="list" href="#short-term-memory">
    Also known as **Metadata**

    What your agent remembers during a **single conversation** or agent run. This information is used to maintain context and consistency throughout that specific interaction.
  </Card>

  <Card title="Long-Term Memory" icon="brain" href="#long-term-memory">
    Persistent across all conversations

    What your agent remembers across **all conversations and agent runs** moving forward. This persistent memory allows your agent to build upon previous interactions and provide personalized experiences over time.
  </Card>
</Columns>

These memory systems help make your conversations:

* **Searchable** - Find exactly what you need
* **Filterable** - Sort conversations by specific criteria
* **Consistent** - Ensure the same data is used throughout conversations
* **Intuitive** - Easily recall and update memorized data when needed

## Short-Term Memory

<Info>Short-Term Memory is also known as Metadata.</Info>

Short-Term Memory allows your agent to remember information throughout one conversation. Setup organized memory metadata that your agent can recognize and update as each conversation progresses. This not only can improve the effectiveness of an agent but also can be used to filter, search and analyze tasks.

<Note>
  When using short-term memory in a Workforce with multiple agents, the metadata persists throughout the entire workforce run as long as agents are configured to "continue same task" (rather than "create new task"). This means all agents in the workflow can read and update the same metadata, making it useful for tracking information across your entire multi-agent process.
</Note>

Adding Short-Term Memory allows you to enrich and recall your data with additional context:

1. Navigate to the Build tab located at the top of your screen, adjacent to Tasks
2. Click on "Memory" in the left sidebar navigation menu
3. Choose your extraction method (Agent-decided or Rule-based)
4. Define the appropriate data format and configure your settings
5. Click the "Save" button to apply your changes

<Tip>
  Consider what information would be most valuable for future analysis when deciding which metadata to add. Focus on data points that will help you segment, filter, or understand patterns in your agent interactions.
</Tip>

### Delete Memory

To remove memory that's no longer needed:

1. Find the memory field you want to remove in the Memory page
2. Click the delete (trash) icon next to the field

<Warning>
  Deleting memory is permanent and will remove this information from all associated conversations or tasks. Make sure you no longer need this data before proceeding.
</Warning>

### Edit Memory Field

To modify an existing memory field:

1. Find the memory field you want to edit in the Memory page
2. Click the edit (pencil) icon next to the field
3. Update the field name, data format, or extraction method as needed
4. Save your changes

### Memory Extraction Methods

Relevance AI offers two approaches for extracting memory metadata:

#### Let Agent Decide

With this option, you provide instructions to your agent about how and when to extract fields from memory. You define guidelines for the agent to follow when determining memory values, making it ideal for complex scenarios where context-aware judgment is needed.

#### Rule Based

Improve memory accuracy by defining specific conditions based on tools & triggers. This approach is either triggered to update after tool run or after trigger received, making it perfect for consistent, predictable data points that follow clear patterns.

### Data Formats for Short-Term Memory

You can add the following data formats as metadata:

1. **Text** - Free-form text entries for descriptions, notes, or other unstructured information
2. **Number** - Numerical values for quantitative data, supporting integers and decimal values
3. **True/False** - Binary values for yes/no or true/false conditions
4. **Single Option** - Predefined list where only one value can be selected
5. **Multiple Option** - Predefined list where multiple values can be selected

<Tip>
  Choose the appropriate data format based on how you plan to use the metadata later. For example, if you want to calculate averages or totals, use the number format. If you need to filter by specific categories, single or multiple option formats work best.
</Tip>

### Built-in Tools

When metadata is added to the Agent, the following built-in tools will be automatically added:

* **Add conversation metadata** - Allows you to add conversation metadata during agent interactions
* **Read conversation metadata** - Enables reading existing conversation metadata

This then allows you to add to the Prompt when you want to add conversation metadata, giving you more control over when and how metadata is captured and used.

### Filtering by Short-Term Memory

You can filter your Workforces and Agent conversations/runs by Short-Term Memory metadata. For example, if you've built an agent that uses metadata like an email or user ID for each person, you can filter agent runs by a person's email or ID to see all their interactions.

<img alt="" />

### Best Practices

* **Be consistent**: Use standardized naming conventions and values for your memory fields
* **Start small**: Begin with a few key memory fields and expand as needed
* **Review regularly**: Periodically assess which memory fields are providing value
* **Document your schema**: Keep track of what each memory field represents and how it should be used

By effectively utilizing memory, you can transform raw conversation data into structured, actionable insights that drive better decision-making and agent performance.

### Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Does short-term memory persist across agents in a Workforce?">
    Yes. When agents in a Workforce are configured to 'continue same task,' short-term memory persists across all agents involved in that workforce run. This allows metadata to be shared and updated by multiple agents working on the same task. If agents are configured to 'create new task,' each agent will have its own separate short-term memory.
  </Accordion>
</AccordionGroup>

***

## Long-Term Memory

Long-Term Memory allows your agent to remember information across different conversations. This is great for use cases like storing user preferences, frequently asked questions, and other persistent information that should be available in future interactions.

When Long-Term Memory is enabled, your agent will be able to read Long-Term Memory added to the agent as it goes.

### Built-in Tools

Two built-in tools will be automatically added to the agent:

* **Save to Memory**: Save important information to your Long-Term Memory for future conversations
* **Delete from Memory**: Remove information from your Long-Term Memory

<Info>
  You don't have to use these tools, but they provide a good way to be more granular and control when to save Long-Term Memory based on certain conditions, after specific tools, or following particular workflows.
</Info>

You can view and manage stored memories from the Memory section of your agent.

### Advanced Settings

From the Advanced Settings in Long-Term Memory, you can:

* **Override the system prompt for saving to memory**: Customize how the agent decides what information to save
* **Override the system prompt for deleting from memory**: Control how the agent determines what to remove from memory
* **Disable delete memory tool**: When disabled, the agent cannot delete items from Long-Term Memory. This is useful when you want to ensure your agent can't accidentally lose important context by deleting memories

<Info>
  Long-Term Memory is particularly valuable for maintaining context across multiple conversations, ensuring your agent can build upon previous interactions and provide more personalized experiences.
</Info>


# Prompt
Source: https://relevanceai.com/docs/build/agents/build-your-agent/prompt

Crafting a prompt to instruct your Agent

<img alt="Agent Builder overview screen" />

The Prompt is the core set of instructions that your agent follows every time it works on a task. Think of it like the instructions you'd provide to an employee about how they should work. You might tell them about their responsibilities and expectations, rules and guidelines, common processes and ways to communicate (do's and don'ts).

***

## Agent Profile

<img alt="Agent profile section showing the name and summary fields" />

### Agent Name

Your agent's name serves as its identity throughout the platform. Choose a name that:

* Clearly reflects the agent's purpose or role
* Is memorable and easy to reference
* Distinguishes it from other agents in your workspace

For example, "Sales Qualification Assistant" immediately communicates the agent's function, while "ResponseBot" might be too vague.

<Tip>
  Fun fact: We like to give our Agents names — for example, **Samuel, the Sales Qualification Assistant.**
</Tip>

### Agent Summary

The summary is a descriptive explanation of what your agent does and its area of expertise. This information:

* Guides the agent's behavior and responses
* Sets expectations for users (you) interacting with the agent
* Helps the underlying AI model understand its purpose

***

## Model Selection

Choosing the right model powers how your Agent thinks, responds, and performs tasks. You can either let Relevance AI automatically optimize the model for you, or manually choose a specific provider and model type.

<img alt="Model selection modal for choosing an LLM provider" />

### Pick For Me

Relevance AI automatically selects the best model for your Agent based on performance or cost, no action needed. You're more than welcome to choose the model the suits your Agent's needs.

<CardGroup>
  <Card title="Cost-optimized">
    Prioritizes cheaper models if you want to minimize token costs/credit costs. (e.g., GPT 4o mini - or whichever model is cheaper at the time)
  </Card>

  <Card title="Peformance-optimized">
    Prioritizes the best-performing model available for quality responses (e.g., Claude v3.5 Sonnet - or whichever is more performant at the time).
  </Card>
</CardGroup>

### Choose Your Own Model

You can manually choose from different model families depending on your needs. Each option charges based on your Relevance AI credits per 1,000 tokens — tokens are pieces of text (words/characters) your Agent processes when sending and receiving messages. See each option in the dropdown for current credit/token costs.

1. More credits = more complex, detailed tasks
2. Fewer credits = cheaper, lighter interactions

<AccordionGroup>
  <Accordion title="OpenAI Models">
    Advanced conversational abilities, creative writing, and broad general knowledge.
    Good for: Versatile agents, customer support, brainstorming.
    Cost: Credits charged per 1,000 tokens used for prompts + replies.
  </Accordion>

  <Accordion title="Anthropic's Claude Models">
    Focused on safe, reliable, and ethical AI responses. Often better at reasoning and "thoughtful" tasks.
    Good for: Detailed explanations, structured outputs, sensitive industries.
    Cost: Credits charged per 1,000 tokens processed.
  </Accordion>

  <Accordion title="Azure OpenAI Models">
    Custom deployments hosted on Azure with higher context limits — ideal for large and detailed inputs.
    Good for: Enterprises needing private deployment, security, or very large tasks.
    Cost: Credits charged per 1,000 tokens sent or received.
  </Accordion>

  <Accordion title="Google's Gemini Models">
    Strong coding ability and task complexity handling (especially Google Gemini 2.5 models).
    Good for: Software development agents, complex task execution.
    Cost: Credits charged per 1,000 tokens processed.
  </Accordion>

  <Accordion title="OpenRouter Models">
    Unified access point to a wide variety of open models like Meta's LLaMA family, Google Gemini models, and others.
    Good for: Exploring experimental or emerging models.
    Cost: Credits charged per 1,000 tokens used, depending on selected model. [Create an OpenRouter account here](https://openrouter.ai/)
  </Accordion>
</AccordionGroup>

<Note>
  **Not all models available in the LLM Tool step can be selected as a model for an Agent.** Agents require models with strong tool-calling capabilities to autonomously use Tools and make decisions. Some models (like DeepSeek) excel at reasoning, coding, and text analysis within LLM Tool steps but may not have the robust function-calling abilities needed for Agent operation. If you need to use these models, consider using them in an LLM Tool step that your Agent calls, rather than as the Agent's primary model.
</Note>

### Exploring Available Models

When selecting a model for your Agent, you can use the model selection modal to find the perfect fit for your needs. The interface provides powerful filtering, sorting, and search capabilities to help you compare models at a glance.

<AccordionGroup>
  <Accordion title="Filter Options" icon="filter">
    Narrow down your model choices based on specific criteria:

    * **Credits per 1k tokens** - Filter by cost to stay within budget
    * **Context Window** - Find models that can handle your required input size
    * **Max Output Tokens** - Ensure the model can generate responses of the length you need
    * **Model Provider** - Select from OpenAI, Anthropic, Google, Azure, OpenRouter, and more
    * **Additional Capabilities** - Filter by special features like vision, function calling, or streaming
    * **Supported File Types** - Find models that can process specific file formats
  </Accordion>

  <Accordion title="Sort Options" icon="arrow-down-wide-short">
    Organize models to prioritize what matters most:

    * **Recommended** - Models suggested based on your use case and performance
    * **Context Window** - Sort by maximum input size (largest to smallest)
    * **Credits Cost** - Sort by price per 1,000 tokens (cheapest to most expensive)
    * **Max Output Tokens** - Sort by maximum response length
  </Accordion>

  <Accordion title="View & Search" icon="magnifying-glass">
    Customize how you browse available models:

    * **Grid View** - See model cards with key specs at a glance
    * **List View** - View detailed information in a compact list format
    * **Search** - Quickly find specific models by name or provider
  </Accordion>
</AccordionGroup>

<Tip>
  If you're choosing a specific model, we recommend doing a quick search up the model to qualify the best fit for your use case. [Search model comparisons here](https://artificialanalysis.ai/models/)
</Tip>

***

## Prompt Instructions

<img alt="Prompt Instructions area in the Agent Builder" />

### Invent a Prompt

<img alt="Invent feature generating a prompt from a short description" />

Use the **Invent** feature to automatically generate a smart, structured prompt based on a few words you provide. Perfect for getting started quickly if you're not sure how to write detailed instructions.

### Write Your Own Prompt

You can also manually type your prompt from scratch if you want full control over how your Agent behaves and what tasks it focuses on. Clear, specific prompts produce the best results.

### Reference Tools in your prompt

While writing your instructions, you can reference specific Tools (e.g., "Use the **Extract Text from PDF** tool to pull invoice data."). This tells the Agent exactly *which* Tools to use when completing tasks. You can also swap a referenced Tool directly from the prompt editor without leaving the page.

### Use the Slash ( / ) Menu

<img alt="Slash menu open in the prompt editor with shortcut options" />

Type **/** to open the Slash menu and quickly insert shortcuts like:

* Referencing available Tools
* Adding Tool placeholders
* Adding Markdown styles for your prompt

This speeds up writing structured and action-driven prompts.

### Create or Invent new Tools from the Slash menu

If you insert a **Tool Placeholder** using the Slash menu, you can immediately **Invent** a new tool or **create one manually** — no need to leave the prompt!\
(Note: Currently, Invented Tools will still need to be customized in the separate **Tool Builder** section)

***

## Connect Tools Conveniently

### Add Tools to Your Agent

Tools allow your Agent to perform real-world actions like searching the web, sending emails, extracting data, or connecting with other systems — expanding your Agent's capabilities far beyond simple conversation.

When writing your Prompt, you can reference specific Tools directly to guide how your Agent should complete tasks. You can also insert Tool Placeholders to invent new tools or connect existing ones as needed.

For a full guide on adding, configuring, and best practices for using Agent Tools, check out the [Tools page](/docs/build/agents/build-your-agent/tools).

## View your Prompt in Markdown

<img alt="Prompt Markdown Format Pn" />

You can select if you wish to view your Prompt in Markdown (formatted), or view it in raw formatting, by scrolling down to the bottom of your Prompt and clicking the checkbox named 'Markdown format'.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Do I need coding knowledge to use Invent?">
    No, Invent is specifically designed to make agent creation accessible to users without technical expertise. You only need to describe what you want your agent to do in plain English.
  </Accordion>

  <Accordion title="Can I modify the agent after it's been created with Invent?">
    Yes, the agent blueprint generated by Invent is fully customizable. You can adjust any aspect of the configuration, instructions, or tools to better suit your needs.
  </Accordion>

  <Accordion title="What types of agents can I create with Invent?">
    Invent is flexible and can help you create a wide variety of agents, from customer support assistants to research tools, data analysts, and more. The possibilities are limited only by your imagination and the capabilities of the underlying AI models.
  </Accordion>

  <Accordion title="How does Invent compare to using agent templates?">
    While templates provide pre-configured agents for specific use cases, Invent creates custom agent blueprints based on your unique description. Templates are ideal when your needs align with existing patterns, while Invent offers more flexibility for novel or specialized requirements.
  </Accordion>
</AccordionGroup>

## Follow along on YouTube

If you want help with writing better prompts, we recommend watching this YouTube video by one of our amazing Solution Engineers, Andrew!

<Frame>
  <iframe title="YouTube video player" />
</Frame>


# Give Your Agent Tools
Source: https://relevanceai.com/docs/build/agents/build-your-agent/tools



## Overview

Tools enable your agent to go beyond simple conversation by performing actions such as searching the web, accessing databases, sending emails, or interacting with other systems. When properly configured, tools transform your agent from a conversational interface into a powerful digital assistant that can take meaningful actions on behalf of users.

<img alt="" />

## Adding Tools to Your Agent

In the Agent Builder interface, navigate to the **Tools** section to begin enhancing your agent's capabilities:

1. Select from pre-built tool templates that cover common use cases
2. Add custom tools you've created under the `Add new tool` option
3. Configure each tool's usage settings based on your requirements

## Tool Input Configuration Options

<img alt="" />

When adding Tools to your Agents, you can choose from three different usage modes for your Inputs:

### Let agent decide

This balanced approach gives your agent the flexibility to determine how the Input is entered based on context. The Agent will use its judgement about what Inputs to give your Tools.

**Best for:** A mix of tool types where you want to get things up and running quickly.

### Set manually

This option allows you to explicitly specify the Inputs for each Tool or action. You have full control over which settings are used for your tools.

**Best for**: Scenarios where strict oversight or compliance is needed, and you want to minimize any ambiguity in approval processes.

### Tool output

This option allows you to select the outputs from a previous Tool run by the Agent.

## Guiding Your Agent's Tool Usage

For optimal results, provide your agent with clear guidance about how and when to use its tools:

**Core Instructions**: Include specific directions about tool usage in your agent's core instructions. For example:

* When to prioritize certain tools over others
* How to interpret and present tool results
* What circumstances warrant tool use versus simple conversation

## Best Practices for Tool Implementation

* **Start Simple**: Begin with a few essential tools before adding more complex capabilities
* **Test Thoroughly**: Verify that each tool works as expected in various scenarios
* **Monitor Usage**: Review how your agent is using tools in real conversations to identify opportunities for improvement
* **Refine Gradually**: Adjust tool configurations and instructions based on actual usage patterns
* **Combine Tools Strategically**: Create powerful workflows by combining multiple tools in sequence

## Common Tool Categories

While the specific tools available depend on your implementation, common categories include:

* **Information Retrieval**: Search engines, knowledge bases, document repositories

* **Communication**: Email, messaging, notifications

* **Data Processing**: Analysis, transformation, summarization

* **Integration**: Connecting with external systems and APIs

* **Content Creation**: Generating text, images, or other media

By thoughtfully selecting and configuring the right tools for your agent, you can create a powerful assistant that delivers exceptional value to your users through both conversation and action.

## Tool Approvals

Control whether your agent needs permission before running a tool. You can set the approval mode individually for each tool.

### Auto Run

The tool runs automatically whenever the agent determines it should be used. The agent incorporates tool functionality without interrupting the conversation flow.

**Best for:** Tools that are low-risk, frequently used, or where immediate action is beneficial.

### Approval Required

The agent will ask for explicit permission before using the tool. When the agent determines a tool would be helpful, it will first explain what it wants to do and why, then wait for user confirmation before proceeding.

**Best for:** Tools that perform sensitive actions, have potential costs associated with them, or where user oversight is important.

### Let Agent Decide

This balanced approach gives your agent the flexibility to determine when approval is necessary based on context. The agent will use its judgment about when to ask for permission versus when to proceed automatically.

**Best for:** A mix of tool types where you want to balance convenience with appropriate caution.

## Tool Settings

### Failure policy

Define what your agent should do when a tool errors. Open the **Tool Settings** on the Tools page to configure the failure policy.

#### Always retry errored tool

When on, forces the agent to retry errored tools until success or max retries (at which point the task will fail). When off, the agent decides on its own whether it's worth retrying. For most users, enabling retries is a safe way to handle occasional hiccups without manual intervention.

#### Maximum number of tool retries

The number of times the agent will retry a tool that errors before marking the task as failed. A few retries (like 3) usually strikes a good balance — it gives your agent a second chance without risking long loops or extra costs. Increasing this may raise credit usage.

#### Behavior after max tool retries

Decide what should happen when the maximum number of retries is reached: either **Terminate task** immediately or **Escalate** it to a human for help. Terminating is faster and simpler, but escalation gives you a chance to fix things manually when it matters most.


# Triggers
Source: https://relevanceai.com/docs/build/agents/build-your-agent/triggers



<img alt="Trigger configuration in the agent builder" />

Triggers allow your Agent to start workflows automatically based on specific events. For example, connecting a Gmail trigger will let your Agent process each new email as soon as it arrives in your inbox. Based on your **Prompt** instructions, the Agent can then take action whenever an email "triggers" it.

You can trigger your Agent through three different methods:

* **Scheduled Events**: Set your agent to run at specific times or intervals
* **External Actions**: Configure your agent to respond to specific events or inputs
* **Manual Activation**: Allow users to directly engage with the agent when needed

## Setting up a trigger

1. Navigate to the **Agents** page and select the agent you want to configure
2. In the sidebar on the left, click **Triggers**
3. Click the **Add trigger** button
4. Choose from the available trigger types — integration triggers, scheduled triggers, or build-your-own options like webhooks, SDK, API, or Tools
5. Follow the configuration steps for your chosen trigger type
6. Once configured, your trigger will appear in the Triggers section of your agent

<Note>
  You can add multiple triggers to a single agent. Each trigger operates independently, so your agent can respond to emails, run on a schedule, and accept API requests all at the same time.
</Note>

## Managing your triggers

Once you've set up triggers, you can control them without deleting your configuration.

### Pausing and resuming triggers

You can pause any trigger type — integration triggers (Gmail, Slack, LinkedIn, WhatsApp, etc.), scheduled triggers, webhooks, and all others. Pausing stops the trigger from firing while preserving all your configuration, so you don't need to delete and recreate triggers when you want to temporarily disable them.

#### To pause or resume a single trigger

1. Go to your agent's **Triggers** section in agent settings or the agent builder sidebar
2. Find the trigger you want to pause
3. Click the pause button next to the trigger
4. To resume, click the resume button

#### To pause or resume all triggers at once

Use the **Pause all** or **Resume all** buttons at the top of the Triggers section to quickly control all triggers for your agent.

### Visual status indicators

Active triggers display with a standard appearance, while paused triggers show a visual indicator so you can quickly see which triggers are currently active and which are paused.

### Where to manage triggers

You can pause and resume triggers from two locations:

* **Agent settings**: Navigate to your agent and open the Triggers section
* **Agent builder sidebar**: Access trigger controls directly while building your agent

### Common use cases for pausing triggers

* **Testing**: Pause production triggers while testing changes to your agent
* **Maintenance**: Temporarily disable triggers during system updates or configuration changes
* **Cost control**: Pause high-volume triggers during periods when you don't need them active
* **Debugging**: Isolate specific triggers to troubleshoot issues
* **Seasonal workflows**: Pause triggers that are only relevant during certain times of the year

## Trigger types

### Integration triggers

We have 1,000+ integrations you can connect to and set up to trigger your Agent, including external triggers like Zapier.

<Card title="Integration triggers" href="/build/agents/build-your-agent/agent-triggers/integrations">View all available integration triggers and premium options.</Card>

### Scheduled triggers

Set your Agent to run on a recurring schedule or send messages at specific times.

<Card title="Scheduled triggers" href="/build/agents/build-your-agent/agent-triggers/scheduled-triggers">Set up recurring schedules and scheduled messages.</Card>

### Build your own triggers

Connect to custom webhooks, APIs, and SDKs to trigger your Agent programmatically.

<Columns>
  <Card title="Webhook" href="/build/agents/build-your-agent/agent-triggers/custom-webhook">Trigger your Agent from a custom webhook.</Card>
  <Card title="API Trigger" href="/build/agents/build-your-agent/agent-triggers/api-trigger">Trigger your Agent with direct HTTP requests from any language.</Card>
  <Card title="JavaScript SDK" href="/sdk/introduction">Build a JavaScript or TypeScript app that talks to your Agent.</Card>
  <Card title="Tools as Triggers" href="/build/agents/build-your-agent/agent-triggers/tools-as-triggers">Use a specific Tool to trigger your Agent.</Card>
</Columns>

## Choosing the right trigger

| I want to...                                            | Use this trigger                    |
| ------------------------------------------------------- | ----------------------------------- |
| React to emails, Slack messages, or CRM updates         | **Integration Triggers**            |
| Run a report every morning or weekly                    | **Scheduled Triggers**              |
| Send follow-ups or reminders at a future time           | **Scheduled Messages**              |
| Connect a third-party service like Zapier               | **Integration Triggers** (external) |
| Receive data from my own app or website                 | **Webhook Trigger**                 |
| Trigger my Agent over HTTP from any language            | **API Trigger**                     |
| Build a JavaScript or TypeScript app on top of my Agent | **JavaScript SDK**                  |
| Combine data from multiple sources before triggering    | **Tools as Triggers**               |

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can I add multiple Triggers to one Agent?">
    Yes. Each Trigger operates independently, so your Agent can respond to integration events, run on a schedule, and accept API requests all at the same time.
  </Accordion>

  <Accordion title="What happens when a Trigger fires?">
    The Trigger sends a message to your Agent, which then processes it based on your Prompt instructions and executes its configured Tools.
  </Accordion>

  <Accordion title="Can I pause or disable a Trigger without deleting it?">
    Yes. You can pause any trigger type from your agent's Triggers section — this works for integration triggers (Gmail, Slack, LinkedIn, WhatsApp, etc.), scheduled triggers, webhooks, and all other trigger types.

    Pausing preserves all your configuration, so you don't need to delete and recreate triggers. You can pause individual triggers or use the **Pause all** button to disable all triggers at once. Visual status indicators show which triggers are active and which are paused.

    This is useful for testing, maintenance, cost control, or any time you need to temporarily disable triggers without losing your setup.
  </Accordion>
</AccordionGroup>


# Variables
Source: https://relevanceai.com/docs/build/agents/build-your-agent/variables



<img alt="" />

## Overview

Variables in Relevance AI allow you to store and reuse data across your agents, eliminating the need to repeatedly type the same information. They provide several key benefits:

* **Efficiency**: Define information once and use it multiple times
* **Consistency**: Ensure the same data is used throughout your workflows
* **Flexibility**: Use variables in both tools and prompts

Variables act as dynamic placeholders that can significantly streamline your agent development process and make your workflows more adaptable to changing requirements.

<Tip>Variables are particularly useful for information that might change over time, such as company details, product specifications, or process guidelines.</Tip>

## Configure Variables

The Configure tab allows you to set up and manage your variables:

1. **Name**: Create a descriptive name for your variable
2. **Description**: Add context about what the variable represents and how it should be used
3. **Input**: Enter the actual value of the variable
4. **Variable Name**: Define the reference name used to call this variable in your tools and prompts
5. **Data Type**: Specify what type of data the variable contains

<Note>For more detailed information about data types and how they function, check out our [Inputs page](https://relevanceai.com/docs/build/tools/customize-tool/inputs).</Note>

## Set Values

The Set Values tab provides a streamlined interface for quickly updating your variable values:

* View all your variables in one place
* Modify values without changing other variable properties
* See immediate updates reflected across your agent

This tab offers the same functionality as the Configure tab but focuses specifically on changing variable values, making it ideal for quick updates when your underlying data changes.

<Tip>
  Consider using variables for information like:

  * Company contact details
  * Product pricing
  * Service descriptions
  * API endpoints
  * Frequently used instructions
</Tip>


# Version history
Source: https://relevanceai.com/docs/build/agents/build-your-agent/version-history

Track, restore, and manage versions of your Agents, Tools, and Workforces.

Version history gives you a complete record of every saved state for your Agents, Tools, and Workforces. Every time you save a draft or publish, a new version is created automatically — so you can experiment freely and roll back if something breaks.

<img alt="The Version History panel listing published versions, with the live version marked and Invent badges on versions created by Invent" />

## How version history works

A new version is created in two situations:

* **On every draft save** — whenever you make a change and save without publishing
* **On publish** — when you publish a draft to make it the active version

This means every meaningful edit is captured. You don't need to manually create snapshots or remember to save before experimenting.

When you edit an Agent, Tool, or Workforce, your changes go to a **draft** first — they don't affect the live, published version until you explicitly publish. This means:

* Your published (active) version keeps running while you iterate
* Draft changes are saved automatically as you work
* Publishing is an intentional step — you decide when changes go live

Each saved draft state and each published state becomes a distinct version in the history list.

### Version names and descriptions

Each version has a short name and a longer description, and both are optional. A version saved without a name is listed by its timestamp instead — for example, `Aug 18, 2:30 am` — so naming the versions you might want to return to makes them much easier to find later. Invent sets a name and description when it publishes changes, but neither field is enforced by the platform.

<Tip>
  Use descriptions that capture intent, not just the change. "Added web search tool" is more useful than "Updated version 3" when you're scanning a list of 20 versions.
</Tip>

When you restore a version, the restored state is saved as a new draft with `(restored)` appended to the version name. The description carries over unchanged from the version you restored.

## Accessing and restoring versions

Version history works the same way for Agents, Tools, and Workforces:

1. Open the Agent, Tool, or Workforce in the builder
2. Click the three-dot menu in the top right
3. Select **Versions**

The panel lists every saved version with its name, who saved it, and its description. The active version is marked **Live**, and a version created by Invent or over MCP carries a badge naming the surface that made the change.

<Note>
  The three-dot menu only lists **Versions** in the builder. It's hidden on an Agent's Use tab, when you're not on the Workforce Build tab, and in the Tool builder embedded inside the Agent builder, where you need to open the Tool on its own page instead.
</Note>

To restore a version, click the three-dot menu on that version in the panel, then select **Restore this version**. You can also ask [Invent](/docs/build/invent/invent) in plain language. Either way the restored state lands in draft, so you review it and publish when ready.

### With Invent

Ask Invent to restore a version by describing what you want — for example:

> "Restore the previous version of this Tool"

Invent always asks you to confirm before applying the restore. It can also rename an existing version for you if you'd rather not open the panel yourself.

### Edit a version name or description

You can relabel any version after it's been saved, including the current draft, the active version, and historical versions. In the version history panel, click the three-dot menu on a version, then select **Edit name / description**. This changes only the label — the configuration stored in that version is untouched, as are its author and timestamp.

<Note>
  Renaming an Agent version requires admin permission on that Agent, while Tools and Workforces also accept project editor permission.
</Note>

## Version retention

Versions are kept indefinitely unless your organization has a data retention policy in place. Where one applies, a version is deleted only if both of these hold:

* It is older than the retention window, measured from when the version was created rather than when it was last edited
* It is not the **active version** (the one currently published) or the **draft version** (your current work in progress). Those two are always protected, at any age

For retention windows and how they're configured, see [Data retention](/docs/enterprise/data-retention).

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can I rename a version after it's been saved?">
    Yes. Open the version history panel, click the three-dot menu on the version, then select **Edit name / description**. This works for Agents, Tools, and Workforces, and applies to the current draft, the active version, and historical versions alike. You can also ask Invent to rename a version.
  </Accordion>

  <Accordion title="Does restoring a version publish it automatically?">
    No. A restored version always lands in draft, and you publish it yourself when you're ready. The live version keeps running until you do, so you get a chance to review the restored state before it goes out.
  </Accordion>
</AccordionGroup>


# Create an Agent
Source: https://relevanceai.com/docs/build/agents/create-an-agent

How to create your first Agent in Relevance AI.

This guide walks you through creating an Agent. For an overview of what Agents are and how they work, see [Agents](/docs/get-started/core-concepts/agents).

## Three ways to create

<CardGroup>
  <Card title="Invent an Agent" icon="wand-magic-sparkles" href="#option-1-invent-an-agent">
    Describe what you want and let AI build it for you
  </Card>

  <Card title="Clone from Marketplace" icon="clone" href="#option-2-clone-from-marketplace">
    Grab a pre-built Agent and make it yours
  </Card>

  <Card title="Build from scratch" icon="hammer" href="#option-3-build-from-scratch">
    Full control over every detail
  </Card>
</CardGroup>

## Option 1: Invent an Agent

The fastest way to get started. Describe what you need and we'll generate a working Agent for you.

<img alt="Select Invent to describe what you want and let AI build your Agent." />

1. Click **Agents** in the left sidebar
2. Click **+ Create Agent**
3. Select **Invent**
4. Describe what you want your Agent to do — be specific about the task, the output you expect, and any integrations it should use. For example: *"Create an agent that researches a company before a sales call. It should use web search to find recent news, funding, and key contacts, then output a one-page brief with company overview, decision makers, and talking points."*
5. Review the generated Agent — Inventor will set up the prompt and suggest Tools
6. Connect any integrations your Tools need (e.g. Gmail, HubSpot, LinkedIn) in the Tools section of the Agent builder, or [add integrations](/docs/integrations/add-integrations) from the sidebar

<Tip>
  Best when you know what you want and can describe it clearly. Great for getting a concept off the ground quickly — Inventor handles the setup, you make the tweaks.
</Tip>

## Option 2: Clone from Marketplace

Start with an Agent that's already been built and tested by domain experts.

<img alt="Browse the Marketplace to find pre-built Agents for common use cases." />

1. Go to [Marketplace](https://marketplace.relevanceai.com/) or click **Marketplace** in the left sidebar
2. Browse or search for an Agent that matches your use case
3. Click on the Agent to view its details
4. Click **Clone** to add it to your workspace
5. A setup popup will walk you through connecting any integrations the Agent needs — follow the prompts to link your accounts
6. Review the prompt, Tools, and settings — adjust anything you need

<Tip>
  Great when you have a common use case — like lead qualification, meeting prep, or customer support — and want a proven starting point. Clone it, tweak it, ship it.
</Tip>

## Option 3: Build from scratch

Full control from the ground up. Best if you have a specific workflow in mind.

<img alt="The Agent builder — configure your prompt, Tools, and Knowledge from scratch." />

1. Click **Agents** in the left sidebar
2. Click **+ Create Agent**
3. Select **Start from scratch**
4. Write a [prompt](/docs/build/agents/build-your-agent/prompt) — tell the Agent who it is, what it does, and how it should behave
5. Add [Tools](/docs/build/agents/build-your-agent/tools) — give it the actions it needs (search the web, send emails, update your CRM, etc.)
6. Optionally add [Knowledge](/docs/build/knowledge/create-knowledge) — upload docs or connect data sources for context

## What's next?

Once you've created an Agent, explore the rest of the build guides to refine it:

* [Prompt](/docs/build/agents/build-your-agent/prompt) — write instructions that define your Agent's behavior
* [Tools](/docs/build/agents/build-your-agent/tools) — give your Agent the actions it needs
* [Triggers](/docs/build/agents/build-your-agent/triggers) — run your Agent automatically on a schedule, from a webhook, or from an integration
* [Alerts](/docs/build/agents/build-your-agent/alerts) — get notified when your Agent needs attention, and define when it should loop in a human
* [Memory](/docs/build/agents/build-your-agent/memory) — configure how your Agent retains context across conversations
* [Version history](/docs/build/agents/build-your-agent/version-history) — every save creates a snapshot you can restore, so you can experiment without losing working versions


# Delete Your Agent
Source: https://relevanceai.com/docs/build/agents/delete-agent

Remove your Agents from Relevance AI

Deleting Agents is a permanent action and cannot be undone. Once an Agent is deleted, it will no longer be available in your Relevance AI project.

## How to delete an Agent

<div>
  <iframe title="How to delete an Agent" />
</div>

1. Click 'Agents' in the sidebar of Relevance AI
2. Click on the three dots button next to the Agent you want to delete, then select 'Delete'
3. On the pop-up that appears, click 'Delete' to confirm the permanent deletion

<Warning>
  Deleting an Agent is a permanent action and cannot be undone. Once an Agent is deleted, it will no longer be available in your Relevance AI project.
</Warning>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What happens if my Agent that I've deleted is clonable?">If your deleted Agent is clonable, it will no longer be available for cloning, but people who have already cloned it will still be able to use it.</Accordion>
  <Accordion title="I accidentally deleted an Agent, can you recover it?">Unfortunately, we cannot recover deleted Agents. Once an Agent is deleted, it will no longer be available in your Relevance AI project. Our support team also cannot recover deleted Agents.</Accordion>
</AccordionGroup>


# Bulk scheduling
Source: https://relevanceai.com/docs/build/agents/give-your-agent-tasks/bulk-schedule

Learn how to schedule batch tasks to run automatically for your agent

<Warning>
  **Team plan or above required**: Bulk scheduling is only available to users on the Team plan or above.
</Warning>

Bulk Scheduling allows you to automate large-scale task execution for your agent by processing data from Excel files or knowledge tables. Instead of manually creating individual tasks, you can import a spreadsheet or knowledge table where each row becomes a separate task that your agent will process automatically.

This feature is ideal for scenarios where you need to:

* Process many similar tasks automatically
* Personalize agent interactions using data from spreadsheets
* Schedule tasks to run during specific working hours
* Distribute task execution over time to manage workload

With bulk scheduling, you can set up message templates that dynamically pull data from your source files, configure working hours to align with your team's schedule, and control how tasks are distributed over time.

## How to set up bulk scheduling

<div>
  <iframe title="How to set up an MCP Server connection for a tool" />
</div>

To create a bulk schedule for your agent:

1. Navigate to the agent builder and click on the `Run` tab.
2. Look for the dropdown menu next to the `+ New Task` button and select `Schedule in bulk`.
3. Select a spreadsheet or knowledge table that contains the data you want your agent to process.
4. Enter a message template for your agent to use. This message will be sent to your agent for every single task run automatically. You can reference variables from your data using double curly braces (e.g., **\{\{variable\_name}}**). Read more about how to create message templates [below](#using-the-agent-message-template).
5. Configure working hours for the batch schedule to ensure that tasks are being run at the appropriate time.
6. Click `Schedule` to create your bulk schedule.

<Note>
  We recommend using descriptive column names in your spreadsheet or knowledge table, and testing with a small dataset first before scheduling large batches of tasks.
</Note>

## Using the agent message template

The message template you create will be sent to your agent automatically for every single task run. This allows you to personalize each task with data from your spreadsheet or knowledge table.

For example, if you want each task to reference a different email address or name, you can use variables in your message template. Variables are referenced using double curly braces with the column name from your data source: **\{\{column\_name}}**.

**How it works:**

* Each row in your spreadsheet or knowledge table becomes a separate task
* The message template is sent to your agent for each task
* Variables in double curly braces (like **\{\{email}}** or **\{\{name}}**) are replaced with the actual values from that row
* You can use any column from your selected data source as a variable

**Example:**

If your spreadsheet has columns for "company" and "email", you could create a template like:

"Please research **\{\{company}}** and send the findings to **\{\{email}}**."

When the tasks run, each one will automatically use the specific company and email from that row's data.

## Using the working hours

A key feature of bulk scheduling is the ability to queue working hours, which allows you to fine-tune how and when tasks are distributed. This ensures that your automated workflows align perfectly with your team's schedule and work patterns.

**Key components:**

* Timezone: Select your preferred timezone to ensure tasks are scheduled according to your local time, for example: Australia - Sydney.

* Work Days: Specify which days of the week tasks should be scheduled.

* Work Hours: Set the time range during which tasks should be executed. You also have the option to select "All day" for 24/7 task execution.

* Task Distribution: Define the number of tasks to be sent per interval. For example, set a specific number of tasks per day. The actual number of tasks your agent can execute may vary due to factors like conflicting contract conditions and rate limiting.

* Override Mode: This give you the option to run all created tasks in override mode, which can be useful for testing or when you need to maintain specific inputs/outputs across all tasks.

<Note>
  We recommend aligning your work days and hours with your team's actual working schedule to ensure timely handling of task outputs. Also, start with a conservative number of tasks per interval and gradually increase as you assess your team's capacity and the agent's performance.
</Note>

## Monitoring your bulk schedule

To view and manage your bulk schedule, navigate to the agent builder, click on the `Run` tab, and then click on `Agent queue` in the top left on the sidebar. From there, you can click on the CSV or knowledge table that you just chose to see your tasks organized by status:

* **Upcoming**: Tasks that are scheduled but haven't started yet
* **In progress**: Tasks that are currently being processed
* **Failed**: Tasks that encountered errors during execution
* **Processed**: Tasks that have completed successfully
* **Cancelled**: Tasks that have been cancelled

Note that you can't cancel specific tasks individually. You can only cancel all tasks within the upcoming view.


# Interact with your Agent
Source: https://relevanceai.com/docs/build/agents/give-your-agent-tasks/interact-with-your-agent

What are tasks and how do they differ from prompt?

## **Agent Chat**

### What is Agent Chat?

Agent Chat gives your Relevance AI agents a friendly, modern chat interface that your users will love. It's like giving your AI a professional "storefront" where people can easily interact with it.

Think of it as the difference between talking to someone through a clunky intercom versus having a face-to-face conversation at a well-designed help desk.

### Why You'll Love It

* **Looks Professional**: Creates a polished, branded experience that builds trust
* **Works Anywhere**: Easily embeds into your website, app, or other digital spaces
* **Fully Customizable**: Matches your brand colors, style, and messaging
* **User-Friendly**: Provides an intuitive interface people already know how to use

### Key Features

### Conversational Interface

Agent Chat provides a clean, familiar messaging experience that feels natural to users. No technical knowledge required - if they've ever used a messaging app, they'll feel right at home.

The interface supports flowing conversations that make even complex interactions feel simple and accessible.

### Upload Files

Users can easily share documents, images, and other files directly in the chat, making it simple to provide context or get help with specific materials.

### Talk to Your Agent

Enable voice interactions for situations where typing isn't convenient or accessible.

### Embed Anywhere

Place your AI assistant exactly where users need help:

* Add it to your website with a simple embed code
* Deploy as a chat widget that appears on any web page
* Maintain consistent branding across all touchpoints

This flexibility means you can put AI assistance precisely where it delivers the most value.

### Customization Options

Make the chat experience truly yours:

* Match your brand colors and icons
* Customize the welcome message and input text
* Configure the chat widget's appearance and behavior
* Create a seamless extension of your brand experience

### How to Set It Up

1. **Configure**: In your Relevance AI dashboard, set up your Agent Chat preferences
2. **Customize**: Adjust colors, messages, and behavior to match your needs
3. **Test**: Try different scenarios to ensure everything works as expected
4. **Deploy**: Generate and implement the embed code on your site

### Important Considerations

* Chat interactions use credits from your Relevance AI account
* Think strategically about where the chat will provide the most value
* Decide if your chat should proactively engage users or wait for them to initiate
* Plan when and how to escalate complex issues to human support

### Perfect For

* **Customer Support**: Provide immediate answers to common questions
* **Product Guidance**: Help users find the right products or features
* **Technical Troubleshooting**: Walk users through solutions step-by-step
* **Information Delivery**: Make knowledge easily accessible through conversation

### Related Features

[**Agent Templates**](https://relevanceai.com/docs/build/agents/templates) - Explore pre-built agent configurations that can be customized for your specific needs and deployed through Agent Chat.

[**Customize an Agent**](https://relevanceai.com/docs/build/agents/customise-agent/agent-profile) - Learn how to tailor your agent's personality, knowledge, and capabilities to create the perfect conversational partner for your users.

[**Share Your Agent**](https://relevanceai.com/docs/build/agents/customise-agent/share-agent) - Discover options for sharing your agents with team members, embedding them in websites, or making them available as templates for others.


# Tasks
Source: https://relevanceai.com/docs/build/agents/give-your-agent-tasks/task-overview

Learn what a Task is, where to view your Tasks, and how to read one in the Run tab

A Task is a single run of work carried out by an Agent or a Workforce: a conversation with an Agent, or a workflow a Workforce runs. It's made up of the messages exchanged and the Tool runs along the way. You can view Tasks in a variety of places, but the two main ones are the Run tab of the Agent and Workforce builders and the Tasks page.

## Where you can view Tasks

<CardGroup>
  <Card title="Run tab" icon="play">
    The Agent and Workforce builders each have a Run tab showing that Agent or Workforce's Tasks, where you can also start new ones.
  </Card>

  <Card title="Tasks page" icon="list-check" href="/build/agents/give-your-agent-tasks/tasks-page">
    Every Task across all your Agents and Workforces in one list, for monitoring and acting in bulk.
  </Card>
</CardGroup>

## What's inside a Task

A Task is made up of the messages exchanged, the steps the Agent took, and the details of the run. The most common place to see all of it is the Run tab, where you interact with an Agent inside the Agent builder. The same Task can also open through an embed or in [Relevance Chat](/docs/get-started/chat/introduction).

<img alt="A Task open in the Run tab, with the Task list on the left, the Timeline in the center, and the details on the right" />

### Task list

The left side lists every conversation for the Agent, grouped by recency, with **All** and **To Review** tabs, search and filters, a **Tabular view**, and the **Agent queue**. Start a fresh one with **New task**.

### Timeline

The center shows the Task thread (the messages exchanged and the Tool runs), with a comment box at the bottom to message the Task directly.

### Details

The right side has **Pause Task** and **Mark as complete**, plus the Task's **Created** by, **Status**, **Date created**, **Credits used**, **Run time**, and **Linked tools**.

## Filtering Tasks

The Task list has a filter builder for narrowing down to exactly the Tasks you need. You can filter by:

* **Status** — Task state like Running, Completed, Paused, or Timed out
* **Agent** — Tasks from a specific Agent
* **Tool usage** — which Tools the Task used
* **Errors** — failure type: **Agent has errored** (the Agent itself stopped on an error) or **Has active tool failures** (a Tool failed mid-run, even if the Agent kept going)
* **Metadata, date, and more** — custom fields, time range, who triggered the Task, and other conditions

Combine the two error filters to catch every kind of failure, including Tool failures that never surfaced as an obvious error.

## Viewing Tasks across all your Agents

The views above each focus on a single Agent or Workforce. To monitor and troubleshoot every Task across all your Agents and Workforces from one place, head to the [Tasks page](/docs/build/agents/give-your-agent-tasks/tasks-page). It's where you triage escalations, errors, and approvals, and act in bulk when a whole group needs the same fix. Selecting any Task there opens the same conversation (its messages and Tool runs) in a side panel.


# Task Queue
Source: https://relevanceai.com/docs/build/agents/give-your-agent-tasks/task-queue

Learn how to access and manage your agent task queue, including how to cancel tasks and view task statuses

## Understanding the Task Queue

When you set up triggers or create bulk schedules for your agent, tasks are organized into queues. The task queue gives you visibility and control over all tasks that are being processed, queued, or have been completed by your agent.

## How to Access the Task Queue

To access your agent's task queue:

1. Navigate to your agent in the agent builder
2. Click on the **Run** tab
3. Look at the **left sidebar** (the task list area)
4. Click on the **Agent queue** button

<div>
  <iframe title="How to access the agent queue" />
</div>

The agent queue button will open a view showing all your task queues organized by their source (triggers and bulk schedules).

## Task Statuses

When you open agent queue, you'll see different sections for each operation that creates tasks (per trigger or per bulk schedule). Within each section, tasks are organized by their current status. You can view tasks in the following categories:

* **In Progress**: Tasks that are currently being processed by your agent
* **Upcoming Queue**: Tasks that are scheduled but haven't started yet
* **Failed**: Tasks that encountered errors during execution
* **Processed**: Tasks that have completed successfully
* **Cancelled**: Tasks that have been cancelled

<img alt="Screenshot showing the agent queue view with task statuses: In Progress, Upcoming Queue, Failed, Processed, Cancelled" />

This organization helps you quickly understand what's happening with your agent's workload and identify any issues that need attention.

## Cancelling Tasks

From the agent queue, you can cancel tasks in bulk. You can cancel all tasks within a specific queue section (like all upcoming tasks in a bulk schedule)

To cancel tasks from the agent queue:

1. Navigate to the queue section you want to manage (e.g., a specific bulk schedule)
2. Go to the **Upcoming Queue** view
3. Click the **Cancel All** button to cancel all tasks in that queue

<img alt="Screenshot showing the Cancel All button in the Upcoming Queue view" />

<Note>
  You can cancel, complete, or pause individual tasks (including tasks in progress) when viewing them in the normal task tab. However, from the agent queue view, you can only cancel tasks in bulk, and only tasks in the "Upcoming Queue" status can be cancelled from the queue.
</Note>

## Viewing Queue Insights

The task queue provides valuable insights about what's being imported and processed:

* **Import status**: See how many tasks have been imported from your CSV or knowledge table
* **Processing progress**: Track how many tasks have been processed versus how many are still queued
* **Trigger activity**: Monitor which triggers are generating tasks and how frequently

These insights help you understand your agent's workload and make informed decisions about scaling or adjusting your task schedules.

## Best Practices

* **Monitor regularly**: Check your task queue regularly to ensure tasks are processing as expected
* **Review failed tasks**: Pay attention to failed tasks to identify patterns or issues that need addressing
* **Plan cancellations**: If you need to cancel tasks, do so from the "Upcoming Queue" view before they start processing
* **Use insights**: Leverage the queue insights to optimize your bulk schedules and trigger configurations

<Note>
  If you notice a high number of failed tasks, review your agent's configuration, tool integrations, or the data format in your bulk schedule to identify and fix the root cause.
</Note>


# Task Operations
Source: https://relevanceai.com/docs/build/agents/give-your-agent-tasks/tasks-page

Monitor, troubleshoot, and resolve Agent and Workforce Tasks in bulk from one place

<img alt="The Tasks page showing the Approvals, Escalated, Errors, and All tasks tabs above a list of Agent Tasks" />

The Tasks page is where you monitor everything your Agents and Workforces are doing: reviewing outputs, approving pending actions, and rerunning failed runs. You can act on a single Task or apply the same action across every matching Task at once.

Tasks from both Agents and Workforces appear here, and the same triage and bulk action flows apply to both.

<Note>
  Each Task is a single run of work (a conversation or workflow) carried out by an Agent or Workforce. For what a Task is and how to read one, see [Tasks](/docs/build/agents/give-your-agent-tasks/task-overview).
</Note>

## Tabs

The page is split into tabs:

* **Approvals** — Tasks waiting on your approval before they can continue.
* **Escalated** — Tasks the Agent paused on to ask you a question.
* **Errors** — Tasks that hit a Tool failure or runtime error.
* **All tasks** — every conversation across your Agents and Workforces.

## Bulk actions

When you open a Task, the page checks for other Tasks from the same Agent in the same state and surfaces a bottom sheet so you can act on the whole group at once.

Available bulk actions:

* **Approve** or **Reject** pending approvals.
* **Rerun** errored Tasks from the failing step.

A confirmation modal shows how many Tasks will be affected before anything is sent.

<Note>
  A Task with more than one pending approval isn't bulk-actionable and stays one-at-a-time.
</Note>


# Share Your Agent
Source: https://relevanceai.com/docs/build/agents/share-your-agent

Learn how to share your agent with others using different UI options

## Overview

Easily share your AI agent with others by choosing how they can interact with it. Whether you want to send a live chat link, embed a support widget, showcase tasks in an iframe, or allow full template cloning, you have flexible options to fit any use case.

<Warning>
  Enabling any embed option (Chat UI, Chat Widget, or Task UI) or the Clonable Template option makes your Agent publicly accessible. The resulting link is a public URL that search engines can index and that anyone can share, so the Agent may be discovered without you sending the link yourself — anyone with it can run your Agent on your credits or clone it into their own project. Disable the option in the Agent's share settings to revoke access.
</Warning>

## Clonable Template

Let others clone your Agent - complete with its instructions and setup - into their own workspace. This makes it easy to share best practices, pre-built use cases, or workflows without needing to rebuild from scratch. Cloned agents use the recipient's credits, not yours, making this perfect for scalable distribution.

<div>
  <iframe title="Relevanceai Demo" />
</div>

## Copy to Project

Want to create a copy of your agent for yourself? Use the "Copy to Project" feature to duplicate agents quickly between different projects. This preserves all agent configurations, settings, and instructions.

<div>
  <iframe title="Relevanceai Demo" />
</div>

## Chat UI

Let others interact with your Agent through a clean, conversation-style interface — just like a chatbot. You can share it via a link or embed it as an iframe. This is perfect for letting people experience your Agent without modifying anything. All usage is billed to your account, and others can use the Agent without needing to set it up themselves.

<div>
  <iframe title="Relevanceai Demo" />
</div>

#### Developer Notes

<Note>
  The `iframe` inherits `100%` of its parent container by default. Configure your container's dimensions carefully to ensure the iframe displays correctly within your page layout without causing layout issues.
</Note>

## Chat Widget

Turn your Agent into a pop-up chat interface you can embed on any website with a simple script tag. It behaves like a support chat window and is ideal for customer-facing use cases like answering questions, handling requests, or collecting feedback directly on your site.

<div>
  <iframe title="Relevanceai Demo" />
</div>

<Warning>
  **User Level Authentication is not supported in embedded chat widgets.** If your agent uses [User Level Authentication](/docs/enterprise/user-level-authentication) (an Enterprise feature that allows each user to authenticate with their own credentials), this will not work in embedded chat widgets. For embedded agents, you'll need to configure your agent to use a shared account for integrations instead.
</Warning>

#### Developer Notes

<Note>
  Target the `.relevanceai-chat-bubble-app` class to style the chat bubble. Adjust its `z-index` property to position the bubble above or below other UI elements such as cookie popups and toast messages.
</Note>

| URL parameter              | Type      | Values                                                                                                                                                                                                   | Description                                         | Examples                                               |
| -------------------------- | --------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | :-------------------------------------------------- | ------------------------------------------------------ |
| `primary_color`            | `string`  |                                                                                                                                                                                                          | A valid CSS color for your form.                    | `#685fff`                                              |
| `bubble_style`             | `enum`    | `agent, icon`                                                                                                                                                                                            | Use an agents avatar or the `bubble_icon`.          | `agent`                                                |
| `bubble_icon`              | `enum`    | `pd/ai-agent, notebook/deploy, pencil, star, sparkle, sparkles, pd/lightning-filled, pd/chat, pd/send-diagonal, stack, duration, pd/check, pd/play, arrow-r, pd/search, share, plus, pd/question-filled` | The icon to use if `bubble_style` is set to `icon`. |                                                        |
| `input_placeholder_text`   | `string`  |                                                                                                                                                                                                          | Set the message input placeholder text.             | `Type your question here…`                             |
| `hide_file_uploads`        | `boolean` |                                                                                                                                                                                                          | Toggle file uploads.                                | `true`                                                 |
| `hide_conversation_list`   | `boolean` |                                                                                                                                                                                                          | Toggle multiple conversations.                      | `true`                                                 |
| `hide_tool_steps`          | `boolean` |                                                                                                                                                                                                          | Toggle tool steps.                                  | `true`                                                 |
| `starting_message_prompts` | `string`  |                                                                                                                                                                                                          | Add starting prompts.                               | `["I need help with my billing", "Write me an email"]` |
| `hide_logo`                | `boolean` |                                                                                                                                                                                                          | Toggle Relevance AI branding.                       | `false`                                                |
| `hide_description`         | `boolean` |                                                                                                                                                                                                          | Toggle agent description.                           | `true`                                                 |

## Task UI

This gives users access to a task-focused version of your Agent via an iframe. Users can view past task runs, pick from existing tasks, and even interact with outputs — all within a self-contained UI. It's great for operational dashboards, workflow demos, or giving others visibility into what your Agent is doing.

<div>
  <iframe title="Relevanceai Demo" />
</div>

## SDK

For developers who want full control over the user experience, you can build custom agent integrations directly into your website or application using our JavaScript/TypeScript SDK.

The SDK gives you complete flexibility to:

* Design custom chat interfaces that perfectly match your brand
* Build unique user experiences beyond what the standard embeds offer
* Integrate agent capabilities into existing applications
* Handle real-time messaging and task management programmatically

The SDK works seamlessly with modern frameworks and runtimes including React, Vue, Node.js, Deno, Bun, and Cloudflare Workers. You can use AI-powered coding tools like Cursor to help you build your integration quickly.

Visit our SDK repository to get started: [https://github.com/RelevanceAI/relevance-js-sdk](https://github.com/RelevanceAI/relevance-js-sdk)

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Why can't I clone an agent that I originally cloned from the Marketplace?">
    Some Relevance Builders have chosen to restrict cloning of agents that originated from their Marketplace listings. This prevents users from redistributing paid listings to avoid payment and protects the builder's intellectual property.

    If you need another copy of the agent, you can return to the [Marketplace](/docs/get-started/marketplace/introduction) and clone the original listing again or add it to other projects directly from there.
  </Accordion>

  <Accordion title="Why can't I use Copy to Project on an agent I cloned from the Marketplace?">
    Some Relevance Builders have chosen to restrict copying of agents that originated from their Marketplace listings. The Copy to Project feature is disabled for these agents to prevent users from redistributing paid listings without proper authorization and to protect the builder's intellectual property.

    To add the agent to additional projects, visit the [Marketplace](/docs/get-started/marketplace/introduction) and clone the original listing directly to each project where you need it.
  </Accordion>
</AccordionGroup>


# Meeting Auto-Scheduling Agent
Source: https://relevanceai.com/docs/build/agents/use-cases/calendar-scheduling

Automate meeting coordination with AI-powered scheduling

Meeting Auto-Scheduling Agents are digital teammates that actively negotiate optimal meeting times by understanding context, preferences, and priorities. They process thousands of scheduling permutations in seconds while coordinating across teams, time zones, and calendars. No more back-and-forth emails.

## Key features

<CardGroup>
  <Card title="Natural language" icon="comments">
    Understands scheduling requests in plain English
  </Card>

  <Card title="Multi-timezone coordination" icon="globe">
    Finds times during working hours for all participants across time zones
  </Card>

  <Card title="Conflict resolution" icon="circle-check">
    Handles hard conflicts, soft conflicts, and priority-based scheduling
  </Card>

  <Card title="Automatic rescheduling" icon="arrows-rotate">
    Proposes alternatives when meetings need to move
  </Card>

  <Card title="Learning preferences" icon="brain">
    Adapts to individual and team scheduling patterns over time
  </Card>

  <Card title="Buffer time management" icon="clock">
    Adds breaks between consecutive meetings
  </Card>

  <Card title="Intelligent prioritization" icon="ranking-star">
    Respects organizational hierarchy and meeting importance
  </Card>

  <Card title="Multi-platform integration" icon="plug">
    Works with Google Calendar, Outlook, and major meeting platforms
  </Card>
</CardGroup>

## How to set up an agent for meeting auto-scheduling

### Triggers for meeting scheduling agents

<iframe title="Meeting Scheduler Triggers Setup" />

To enable your scheduling agent, you'll need a Google Calendar integration with read/write access to check availability and create events, plus a Gmail trigger to monitor your inbox and kick off the agent when scheduling requests arrive.

### Add tools to your scheduling agent

To build a calendar scheduling agent, you'll need these basic tools to get started:

<CardGroup>
  <Card title="Check Google calendar availability" icon="calendar-check">
    Find open slots in your sales team's calendars
  </Card>

  <Card title="Send email (Gmail)" icon="envelope">
    Reply with available time options to prospects
  </Card>

  <Card title="Create Meet Event in Google Calendar" icon="video">
    Automatically generate video links for each demo
  </Card>

  <Card title="Get events from Google Calendar" icon="list">
    Avoid double-booking or scheduling too many demos in one day
  </Card>
</CardGroup>

#### How to add these tools to your agent

1. Go to **Tools** in your Agent Builder
2. Click **+ Add Tool**
3. Click **New Tool**
4. Click **Build from Scratch**
5. Click **Add Step**
6. Search for your calendar provider (e.g., "google calendar" or "outlook")
7. Select the tools you need from the results

<Tip>
  Search for your specific calendar provider to discover available built-in tools.
</Tip>

<Frame>
  <img alt="Available calendar tools for scheduling agents" />
</Frame>

## Use case example: Automating demo requests for sales

<Accordion title="Calendar Bookings Scheduler Agent - clone from marketplace">
  **Scenario**

  Your sales team receives numerous demo requests via email. Instead of manually checking calendars and coordinating schedules, an AI agent can handle the entire process automatically.

  ***

  **How it works:**

  <Steps>
    <Step title="Monitors inbox">
      Automatically detects demo requests from incoming emails
    </Step>

    <Step title="Checks availability">
      Reviews your calendar in real-time to find open slots
    </Step>

    <Step title="Coordinates timezones">
      Handles timezone differences automatically to suggest appropriate meeting times
    </Step>

    <Step title="Proposes time slots">
      Suggests 30-minute meeting slots across two different days (avoiding weekends)
    </Step>

    <Step title="Books & confirms">
      * Books immediately if a specific time is requested and available
      * Proposes 2-3 times per day when flexibility is needed
      * Sends professional confirmation emails
    </Step>
  </Steps>

  ***

  **Ready to use this agent?**

  [Clone from Relevance AI Marketplace →](https://marketplace.relevanceai.com/)
</Accordion>


# Meeting Agents
Source: https://relevanceai.com/docs/build/agents/use-cases/meeting-agents-overview

Stop taking notes and coordinating schedules. Let AI handle your meetings.

Meeting Agents coordinate schedules, take notes, transcribe conversations, and handle follow-ups so you can focus on what matters.

## Key features

<CardGroup>
  <Card title="Real-time transcription" icon="waveform">
    Automatic transcription of every conversation
  </Card>

  <Card title="Smart scheduling" icon="calendar-check">
    Find optimal times across timezones
  </Card>

  <Card title="Action item extraction" icon="list-check">
    Automatically identify tasks and owners
  </Card>

  <Card title="Multi-platform support" icon="plug">
    Google Meet, Zoom, Microsoft Teams
  </Card>

  <Card title="Custom summaries" icon="file-lines">
    Configure format and distribution
  </Card>

  <Card title="Live control" icon="bolt">
    Use On-Call AI Commands during meetings
  </Card>
</CardGroup>

## What can meeting agents do?

### Meeting Auto-Scheduling Agent

Coordinate meeting times across participants, handle timezone conflicts, and send calendar invitations. No more back-and-forth emails.

<Accordion title="Example use case: Interview scheduling">
  When scheduling candidate interviews across multiple team members in different timezones, your agent finds times that work for everyone, sends calendar invites, and handles rescheduling if conflicts arise. What would take hours of email chains happens automatically in minutes.
</Accordion>

[Meeting Auto-Scheduling Agent →](/docs/build/agents/use-cases/calendar-scheduling)

### Meeting Notetaker Agent

Automatically join meetings, record conversations, transcribe audio, and generate summaries with action items. Works with Google Meet, Zoom, and Microsoft Teams.

<Accordion title="Example use case: Interview note-taking">
  When interviewing a candidate, let your agent handle the note-taking. You can focus on building rapport, reading body language, and asking thoughtful follow-up questions instead of scrambling to write everything down. After the interview, review a complete transcript with key discussion points and next steps.
</Accordion>

[Meeting Notetaker Agent →](/docs/build/agents/use-cases/meeting-notetaker)


# Meeting Notetaker Agent
Source: https://relevanceai.com/docs/build/agents/use-cases/meeting-notetaker



A Meeting Notetaker Agent joins your meetings, records conversations, and generates summaries with action items.

* **Joins automatically**: The agent attends meetings on your behalf
* **Records everything**: Get video recordings and full transcripts without lifting a finger
* **No device recording needed**: The agent handles it in the background
* **Customizable**: Extract insights tailored to your needs, whether product feedback, sales objections, or technical requirements

## Key features

<CardGroup>
  <Card title="Automatic meeting attendance" icon="calendar-check">
    Agent joins meetings directly from your calendar with no manual intervention required
  </Card>

  <Card title="Audio & video recording" icon="video">
    Captures comprehensive recordings without using your device
  </Card>

  <Card title="Speaker-identified transcription" icon="microphone">
    Creates accurate transcripts with speaker labels and timestamps
  </Card>

  <Card title="Intelligent summarization" icon="list-check">
    Extracts key moments, decisions, and action items automatically
  </Card>

  <Card title="Cross-platform support" icon="display">
    Works with Zoom, Microsoft Teams, and Google Meet
  </Card>

  <Card title="Flexible distribution" icon="share-nodes">
    Send summaries to Slack, email, Google Docs, Notion, or your CRM
  </Card>

  <Card title="On-call AI commands" icon="message">
    Control the agent during live meetings with chat commands
  </Card>

  <Card title="Customizable insights" icon="sliders">
    Tailor outputs to your domain including product feedback, sales objections, and technical requirements
  </Card>
</CardGroup>

## How it works

<Steps>
  <Step title="Pre-meeting" icon="calendar">
    Agent triggers from your calendar and sends a meeting bot to join.
  </Step>

  <Step title="During meeting" icon="video">
    While you focus on the conversation, the agent works silently:

    * Capturing comprehensive audio and video recordings
    * Creating accurate speaker-identified transcription
    * Detecting key moments, decisions, and actionables
    * Identifying discussion themes and topic clusters
  </Step>

  <Step title="Post-meeting" icon="file-lines">
    Processes the complete transcript by speaker and timestamp, then extracts domain-specific insights tailored to your needs.
  </Step>
</Steps>

## How to setup a Meeting Notetaker Agent

### Step 1: Create your agent

[Create a new agent](/docs/build/agents/build-your-agent) or clone Nate, a production-ready notetaker from the [Relevance AI Marketplace](https://relevanceai.com/agent-templates-meetings/). If you want to build your own, you can take inspiration from Nate's configuration.

### Step 2: Add critical tools

Your Meeting Notetaker Agent requires these essential tools:

| Tool                               | Description                                               |
| ---------------------------------- | --------------------------------------------------------- |
| **Send meeting bot**               | Dispatches the meeting bot to join and record a call      |
| **Send meeting bot message**       | Sends a chat message from the meeting bot to participants |
| **Pause meeting bot recording**    | Pauses the recording while staying in the meeting         |
| **Resume meeting bot recording**   | Resumes recording after it has been paused                |
| **Turn transcript into text file** | Converts the transcript into a downloadable text file     |
| **Google Search**                  | Searches the web for keywords during live meetings        |

These are pre-built tools available in your agent's tool library.

<Tip>
  Start with **Send meeting bot** and **Turn transcript into text file** for basic functionality. Add the other tools as you expand your agent's capabilities.
</Tip>

### Step 3: Choose how to run your agent

#### Option A: Automatic (add triggers)

<div>
  <iframe title="Setting up Meeting Notetaker triggers" />
</div>

Set up triggers if you want your agent to automatically join meetings from your calendar. Connect two triggers:

* **Calendar** (Google Calendar or Outlook): Monitors your schedule and sends the bot to join meetings
* **Relevance Meeting Bot**: Receives the transcript after the meeting ends

<Note>
  You'll need to admit the bot as a guest into the meeting through your video platform.
</Note>

#### Option B: manual

If you prefer to run your agent manually, simply send the meeting link directly to your agent:

```
Join this meeting: https://meet.google.com/abc-def-ghi
```

## Use cases

Your meeting notes don't have to stay as notes. Connect your Meeting Notetaker to other tools to turn conversations into real deliverables:

<AccordionGroup>
  <Accordion title="Case studies & testimonials">
    Had a great customer call? Turn it into a case study or testimonial you can share with prospects.
  </Accordion>

  <Accordion title="Product roadmaps">
    Capture feedback from user interviews and automatically organize it into a product roadmap.
  </Accordion>

  <Accordion title="Sales enablement">
    Use insights from sales calls to improve your pitch. Add a pitch deck tool to automatically create slides from meeting content.
  </Accordion>

  <Accordion title="Social content">
    Record an expert interview and let the agent pull out quotes and highlights for social media posts.
  </Accordion>

  <Accordion title="Onboarding docs">
    Turn training sessions into step-by-step guides for new team members.
  </Accordion>
</AccordionGroup>

## Advanced: On-call AI commands

You can control your agent in real-time by typing commands in the meeting chat:

* **Pause recording**: Stop recording during sensitive discussions
* **Search**: Get instant research while you're still on the call
* **Add notes**: Flag important items to highlight in the summary

[Learn more →](/docs/build/agents/advanced-features/on-call-ai-commands)


# Build with Relevance AI
Source: https://relevanceai.com/docs/build/introduction

An overview of building and managing agents with Invent or configuring agents, tools, knowledge, and workforces directly.

Relevance AI has two main experiences — the builder platform for creating, and Chat for using.

The **[builder platform](https://app.relevanceai.com)** is where you create and configure everything — agents, tools, workforces, and knowledge. You'll also find the [Marketplace](/docs/get-started/marketplace/introduction) here, where you can clone pre-built agents and tools instead of starting from scratch. This section of the docs covers the builder platform and all of its features.

**[Relevance Chat](/docs/get-started/chat/introduction)** is a conversational interface where you and your team interact with your agents directly. You can @ mention agents, use multiple agents in a single conversation, and give your team access to your agents — all through a familiar chat experience.

This section focuses on building. To learn more about Chat, head to the [Relevance Chat](/docs/get-started/chat/introduction) page in Getting Started.

<img alt="The builder platform — where you create and configure Agents, Tools, Knowledge, and Workforces." />

## Build with Invent

Invent is one of the main ways to build in Relevance AI. Describe a business process in plain language and work with Invent to define its inputs, systems, decisions, exceptions, and measures of success.

Invent creates and connects the Agents, Tools, Triggers, Knowledge, and Workforces needed to run the process. After launch, it can review Evals and live performance, diagnose quality problems, and propose tested improvements for your approval.

<Card title="Explore Invent" icon="wand-magic-sparkles" href="/build/invent/invent">
  Learn how Invent helps you understand a process, build the Agent team, and maintain its quality after launch.
</Card>

## How it all fits together

Invent can assemble and coordinate these components for you, or you can configure them directly:

<Steps>
  <Step title="Agents — the workers" icon="robot">
    You create [agents](/docs/build/agents/create-an-agent) that use your tools and knowledge to autonomously complete tasks, following the instructions and guardrails you set.
  </Step>

  <Step title="Tools — the actions" icon="screwdriver-wrench">
    You build [tools](/docs/build/tools/create-a-tool) that define what your agents can do — call an API, send an email, search your CRM, run custom code, or chain steps together into a workflow.
  </Step>

  <Step title="Workforces — the team" icon="users-gear">
    You connect multiple agents into a [workforce](/docs/build/workforces/create-a-workforce) on a visual canvas — they collaborate and hand off work to each other.
  </Step>

  <Step title="Knowledge — the context" icon="brain">
    You connect [knowledge](/docs/build/knowledge/create-knowledge) so agents have access to your data — documents, spreadsheets, databases, websites — and can respond with accuracy.
  </Step>
</Steps>

## Choose how to build

You can work with Invent or configure each component directly:

<CardGroup>
  <Card title="Build with Invent" icon="wand-magic-sparkles" href="/build/invent/invent">
    Describe a business process in plain language. Invent builds and connects the Agents, Tools, Knowledge, Triggers, and Workforces needed to run it, then helps monitor and improve performance after launch.
  </Card>

  <Card title="Build directly" icon="hammer" href="/build/agents/create-an-agent">
    Full step-by-step control over every setting and configuration. Ideal when you need precision, have a specific workflow in mind, or want to fine-tune behavior.
  </Card>
</CardGroup>

## What's next?

Start with [Invent](/docs/build/invent/invent) to build and manage a process with natural language. To configure each component directly, begin with [building an agent's prompt](/docs/build/agents/build-your-agent/prompt), then explore how to create [custom tools](/docs/build/tools/customize-tool/tool-style-guide), connect [knowledge sources](/docs/build/knowledge/create-knowledge), and wire it all together in a [workforce](/docs/build/workforces/build-an-ai-workforce/add-triggers).


# Build with Invent
Source: https://relevanceai.com/docs/build/invent/invent

Invent is a forward deployed agent that builds, manages, and improves the specialist Agents that run your business processes.

<img alt="The Invent panel in Relevance AI" />

## What is Invent

Invent is a forward deployed agent built into Relevance AI. Give it a business process and it works with you to understand how that process runs, including the systems involved and the decisions that require judgment.

It then creates and coordinates the Agents, Tools, Triggers, and Workforces needed to carry out the process, with performance requirements in place before real work begins.

Its job continues after launch. Invent reviews live performance, detects quality problems, and proposes and tests improvements. You bring the process knowledge; Invent builds and manages the Agent team underneath it.

## What can Invent do?

<CardGroup>
  <Card title="Understand your process" icon="magnifying-glass">
    Invent works through your process with you, including its inputs, systems, judgment calls, exceptions, and definition of success. It can read the process back to you, identify missing information, and help determine which parts should become Tools, Agents, or Workforces.
  </Card>

  <Card title="Build and connect the Agent team" icon="diagram-project">
    Invent creates the Agents needed for the process, equips them with the required Tools and context, connects the systems where the work happens, and wires the Triggers and Workforces that coordinate the process from end to end.
  </Card>

  <Card title="Set and maintain the quality bar" icon="gauge">
    Invent defines the checks each Agent must pass before launch and samples live runs after deployment. When quality drops, it diagnoses the problem and proposes changes to prompts, Tools, examples, models, or orchestration. It can test those changes against the same requirements before you approve them.
  </Card>

  <Card title="Review performance and find opportunities" icon="chart-line">
    Invent reviews Agent activity, errors, Evals, and performance dashboards to diagnose problems and recommend improvements. Describe a recurring process, and it can help identify where Agents, Tools, or Workforces could automate it.
  </Card>
</CardGroup>

## Where to find Invent

<img alt="The Invent entry points on the Relevance AI home page" />

* **Left sidebar:** Click Invent under Build.
* **Home page:** Click Build with Invent.
* **Floating button:** Click the floating button to open the Invent panel from any page in your project.

## Access controls

### Availability

Invent is available to organization admins by default. An organization admin can turn on Invent for everyone in an individual project from that project's settings.

### Approval controls

<Tabs>
  <Tab title="Session approvals">
    <img alt="The Ask first and Auto-accept options in the Invent message composer" />

    Use the approval control beside the message composer to set how Invent handles tool approvals for the current session:

    * **Ask first** pauses before tools set to **Ask** or **Always ask** run. Tools set to **Auto** still run without approval.
    * **Auto-accept** lets tools set to **Ask** run without approval. Tools set to **Always ask** still require your approval.
  </Tab>

  <Tab title="Tool permissions">
    <img alt="Invent Permissions settings showing approval controls for individual tools" />

    Tool permissions save how Invent should handle each action for you in the current project. Set an entire category at once or choose a different setting for an individual tool.

    | Setting        | What it means                                                   |
    | -------------- | --------------------------------------------------------------- |
    | **Auto**       | Runs without asking.                                            |
    | **Ask**        | Requests approval unless **Auto-accept** is on for the session. |
    | **Always ask** | Requests approval every time, even when **Auto-accept** is on.  |

    Run actions default to **Auto**, create and edit actions to **Ask**, and destructive actions such as deleting or publishing to **Always ask**. Setting a destructive action to **Auto** allows Invent to perform it without another tool-approval prompt.

    <Steps>
      <Step title="Open Invent settings">
        In the Invent panel, click the three-dot menu, then select **Invent settings**.
      </Step>

      <Step title="Open Permissions">
        Select **Permissions**, configure categories or individual tools, then click **Save**.
      </Step>
    </Steps>

    <Note>
      These settings control approval prompts, not access. Invent still uses your existing project, asset, and integration permissions. Each user manages their own preferences in each project; organization and project admins cannot use this screen to grant access or set approval rules for other users.
    </Note>
  </Tab>
</Tabs>

### Enterprise security controls

<AccordionGroup>
  <Accordion title="Project isolation">
    Invent's access to Relevance AI is scoped to the active project at the API-key and fine-grained authorization layers, not only in the interface. Each internal platform request uses the active project ID and the caller's API key, then passes through the standard authorization checks. Enabling Invent in a sandbox project does not grant it access to assets in a production project.
  </Accordion>

  <Accordion title="Integration reach">
    Within the active project, Invent can use connected integrations such as Salesforce or Gmail with credentials available to that project. The project is the access boundary, so connect only sandbox credentials if the project must remain isolated from production data.
  </Accordion>

  <Accordion title="Data egress">
    Conversation content is sent to the selected LLM provider to generate responses. Web search is off by default and used only when enabled for a message. Voice is feature-gated and off by default. Memory is off unless the user enables it. Audit data and operational metrics stay within Relevance AI unless the organization has configured enterprise event streaming to a customer-controlled destination.
  </Accordion>
</AccordionGroup>

## How to use Invent

<div>
  <iframe title="How to use Invent" />
</div>

<Warning>
  Invent is expensive to run and consumes significant credits. Use it deliberately and keep an eye on your usage.
</Warning>

## Invent features

<Tabs>
  <Tab title="Set up and use Invent">
    <AccordionGroup>
      <Accordion title="Start a task" icon="wand-magic-sparkles">
        <img alt="The Invent start menu with suggested prompts and the message composer" />

        Describe a process you know end to end, including the outcome, systems, judgment calls, and measure of success. Invent works through it with you before building the Agent team.

        You can also choose a task suggested from your project setup, such as creating evals if none exist.
      </Accordion>

      <Accordion title="About you" icon="user">
        <img alt="The About You setting in Invent" />

        About You tells Invent who you are so it tailors every response to your role and work from the first message — no re-explaining your context each session. It shapes how Invent talks to you and what it assumes about your work, but it doesn't change Invent's role or hardcode any Tool configuration.

        It adjusts the level of technical detail, makes domain-aware assumptions, and lets Invent address you by name at natural moments.

        To set it up:

        <Steps>
          <Step title="Open Invent settings">
            In the Invent panel, click the three-dot menu, then select **Invent settings**.
          </Step>

          <Step title="Open General">
            Select **General**, enter your details under **About You**, then click **Save**.
          </Step>
        </Steps>
      </Accordion>

      <Accordion title="Model selection and credit usage" icon="brain">
        <img alt="The Invent model list showing each model's cost, performance profile, and support for images and files" />

        #### Model selection

        The model dropdown sets which large language model (LLM) powers your Invent session.

        The dollar signs show each model's relative credit cost. Labels such as **Fast**, **Balanced**, **Powerful**, and **Thorough** summarize the trade-off between speed, cost, and capability. The icons beside each model show whether it can work with images and file attachments.

        Your plan determines the default model. You can switch models at any time from the dropdown in the Invent panel.

        #### Credit usage

        The counter beside the model selector shows the credits used in your current session and updates with each interaction.
      </Accordion>

      <Accordion title="Task status and history" icon="list-check">
        <img alt="Invent task history showing approve, idle, and other statuses" />

        Past sessions in your conversation history show each task with a status: approved (the action was applied), idle (waiting for your input), or completed. You can review what Invent did and track credit spend across sessions from the history view.
      </Accordion>

      <Accordion title="Memories" icon="brain">
        <img alt="The Memory settings in Invent" />

        When Memory is enabled, Invent can save and recall details you want it to remember across conversations, such as naming conventions, preferred models, workflows, and recurring constraints.

        To view, search, edit, or delete your memories:

        <Steps>
          <Step title="Open Invent settings">
            In the Invent panel, click the three-dot menu, then select **Invent settings**.
          </Step>

          <Step title="Open Memory">
            Select **Memory**, manage your memory settings, then click **Save**.
          </Step>
        </Steps>

        <Warning>
          Deleting a memory removes it permanently. Invent will re-learn that information if it comes up again in future conversations.
        </Warning>
      </Accordion>

      <Accordion title="Bug reporting" icon="triangle-exclamation">
        When Invent hits a platform issue it can't fix, it logs a bug to the Relevance AI engineering team on your behalf. These reports go directly to engineering rather than support, so you won't receive progress updates.
      </Accordion>
    </AccordionGroup>
  </Tab>

  <Tab title="Manage performance">
    <AccordionGroup>
      <Accordion title="Scheduled check-ins" icon="calendar-clock">
        <img alt="A scheduled check-in in Invent" />

        <Note>
          Before setting up a scheduled check-in, add an email or Slack notification channel in [Alerts](/docs/build/invent/invent#alerts) so Invent can notify you about the result.
        </Note>

        #### Schedule a check-in

        Tell Invent what to review and when. It can suggest an interval based on the Agent's activity.

        #### What happens when a check-in runs

        Invent reviews evals, errors, and recent activity. If it finds an issue, it sends a summary and suggested fix. If everything looks healthy, it stays quiet.
      </Accordion>

      <Accordion title="Alerts" icon="envelope">
        <img alt="The email and Slack alert settings in Invent" />

        Add at least one email address or Slack channel so Invent can notify you when a scheduled check-in finds something.

        <Steps>
          <Step title="Open Invent settings">
            In the Invent panel, click the three-dot menu, then select **Invent settings**.
          </Step>

          <Step title="Open Alerts">
            Select **Alerts**, add an email address or Slack channel, then click **Save**.
          </Step>
        </Steps>
      </Accordion>

      <Accordion title="Restore a version" icon="clock-rotate-left">
        Ask Invent to roll an asset back to an earlier version — for example, "Restore the previous version of this tool". Invent copies that version into the current draft, leaving the live published version unchanged until you publish. This restores an earlier version of an existing asset; it doesn't recover one that's been deleted.

        Invent always asks you to confirm before restoring, and appends `(restored)` to the version name so the restored state is easy to identify later. For a full guide to version history — including how to restore and rename versions from the builder UI, and how retention works — see [Version history](/docs/build/agents/build-your-agent/version-history).
      </Accordion>
    </AccordionGroup>
  </Tab>
</Tabs>

## Manage Evals with Invent

Invent can build an initial Evals suite, review results, investigate failed Checks, and diagnose performance alarms for Agents and Workforces.

<Card title="Learn more about Evals" icon="chart-line" href="/build/agents/build-your-agent/evals">
  Learn how to create scenarios and reusable Checks, run evaluations, and monitor live performance.
</Card>

<Tabs>
  <Tab title="Create Evals">
    <img alt="Creating evals with Invent" />

    Invent can turn an Agent or Workforce's real usage into a starting Evals suite, including a test set, scenarios, and reusable Checks.

    <Steps>
      <Step title="Start from your Agent or Workforce">
        On the **Run** page, select **Create with Invent** from the testing banner. If a recent Agent task is available, you can instead select **Automate testing** beside it so Invent uses that run as the example.
      </Step>

      <Step title="Describe what success looks like">
        Add any outcomes, edge cases, or requirements that the Evals should cover. A representative task gives Invent useful context about the inputs, expected behavior, and tools involved.
      </Step>

      <Step title="Review and run the Evals">
        Invent creates the Evals setup and runs the test. Open **Evaluate → Runs** to review the score and see which Checks passed or failed.
      </Step>
    </Steps>

    <Tip>
      Start with a task that represents normal production use, then ask Invent to add edge cases once the baseline Evals are working.
    </Tip>
  </Tab>

  <Tab title="Review performance">
    <img alt="Reviewing performance results with Invent" />

    Use **Review with Invent** for an on-demand analysis of a completed Evals run or Performance dashboard. Selecting it starts a fresh conversation with the relevant page as context, including the dashboard name when applicable. Invent summarizes the results and recommends improvements, and you can ask it to compare Checks, explain score changes, identify recurring weak points, or prioritize the highest-impact recommendations.

    <Note>
      A review is requested manually; it does not watch performance continuously. Configure an alarm if you want to be notified when a score crosses a threshold.
    </Note>
  </Tab>

  <Tab title="Fix failing Evals">
    <img alt="Invent proposing fixes for failing evals" />

    Target a specific failed or errored Check when you already know which result needs attention. Invent receives the Check name and run page, investigates why it did not pass, and works with you on a fix.

    <Steps>
      <Step title="Open the failed result">
        Go to **Evaluate → Runs**, open a completed Evals run, then expand a failed test to see its Check results and failure reasons.
      </Step>

      <Step title="Select Fix">
        The **Fix** button appears beside each Check that failed or returned an error. Selecting it starts a new Invent conversation focused on that Check.
      </Step>

      <Step title="Review and verify the fix">
        Invent identifies the likely cause and proposes changes to the Agent, Workforce, or Evals setup. Review the proposed actions before applying them, then rerun the Evals to confirm the Check passes without breaking other behavior.
      </Step>
    </Steps>

    <Tip>
      Use **Review with Invent** for a summary of the entire run; use **Fix** when you want Invent to investigate one failed Check.
    </Tip>
  </Tab>

  <Tab title="Configure an alarm">
    <img alt="The alarm configuration in Invent" />

    #### Turn on auto-diagnosis

    Turn on **Auto-diagnosis by Invent** to investigate the first breach and send a diagnosis and proposed fix to the alarm creator's [Invent alert channels](/docs/build/invent/invent#alerts). Invent still shares its findings if the cause is unclear and never applies changes without your approval.

    <Note>
      Alarms are configured in an Agent or Workforce's **Evaluate → Monitor** area, not in Invent settings.
    </Note>

    Alarms evaluate the scores collected by a Performance dashboard and notify you when an overall score or individual Check crosses the condition you set.
  </Tab>
</Tabs>

## MCP vs Invent

Invent and the Relevance AI MCP are often compared, but they solve different problems. Invent builds things *inside* Relevance AI from natural language. The MCP connects Relevance AI *to the outside world* — letting external clients call your Tools, and letting your Agents reach into external services.

<CardGroup>
  <Card title="Invent" icon="wand-magic-sparkles">
    A forward deployed agent inside Relevance AI. Describe the process you want to run and it builds or edits Agents, Tools, and Workforces for you, landing changes in draft for review.

    Reach for it when you want to build, edit, or debug something inside Relevance AI by describing the outcome.
  </Card>

  <Card title="Relevance AI MCP" icon="plug">
    A bridge to other tools. Call your existing Relevance AI tools from external clients like Cursor or Claude with the [MCP Server](/docs/integrations/mcp/mcp-server), or let your Agents call services like Notion and Linear with the [MCP Client](/docs/integrations/mcp/mcp-client).

    Reach for it when the Tool already exists and you want to run it from Cursor or Claude, or let an Agent call an outside service.
  </Card>
</CardGroup>

<Tip>
  The two are complementary, not competing. Invent uses the same MCP tool surface under the hood — so a common pattern is to build your Agents and Tools with Invent, then call them from Cursor or Claude through the MCP in your day-to-day work.
</Tip>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="How do I get started with Invent?">
    Choose a process you know well and can explain from beginning to end. Tell Invent the outcome you need, the systems involved, the decisions that require judgment, and how you measure success. Invent will work through the process with you before it builds the Agent team.
  </Accordion>

  <Accordion title="Who can use Invent?">
    Invent is available to organization admins by default. An organization admin can turn on Invent for everyone in an individual project from that project's settings.
  </Accordion>

  <Accordion title="How does project isolation work for Enterprise customers?">
    Invent's access to Relevance AI is scoped to the active project at the API-key and fine-grained authorization layers, not only in the interface. Each internal platform request uses the active project ID and the caller's API key, then passes through the standard authorization checks. Enabling Invent in a sandbox project does not grant it access to assets in a production project.
  </Accordion>

  <Accordion title="Which integrations can Invent access in an Enterprise project?">
    Within the active project, Invent can use connected integrations such as Salesforce or Gmail with credentials available to that project. The project is the access boundary, so connect only sandbox credentials if the project must remain isolated from production data.
  </Accordion>

  <Accordion title="Where can Enterprise Invent data be sent?">
    Conversation content is sent to the selected LLM provider to generate responses. Web search is off by default and used only when enabled for a message. Voice is feature-gated and off by default. Memory is off unless the user enables it. Audit data and operational metrics stay within Relevance AI unless the organization has configured enterprise event streaming to a customer-controlled destination.
  </Accordion>

  <Accordion title="Does Invent publish changes automatically?">
    By default, Invent makes changes in draft so you can review them before they go live. Auto-accept lets Invent apply changes without pausing for each approval, but restoring an earlier version always requires confirmation.
  </Accordion>

  <Accordion title="Can I resume a previous Invent conversation?">
    Yes. Invent saves conversations as sessions automatically. Open the **Chats** dropdown in the Invent panel to browse or search previous sessions and continue where you left off.
  </Accordion>

  <Accordion title="How does Invent track credit usage?">
    The Invent panel displays a real-time credit counter that updates with each interaction. The counter tracks the credits used in your current session.
  </Accordion>
</AccordionGroup>


# Knowledge
Source: https://relevanceai.com/docs/build/knowledge/create-knowledge

Our Retrieval Augmented Generation (RAG) solution

# Create and Organize Knowledge

This guide walks you through creating knowledge bases, organizing them, and connecting them to your agents. For an overview of what Knowledge is and how it works, see [Knowledge](/docs/get-started/core-concepts/knowledge).

## Creating Knowledge Sources

Relevance AI offers multiple flexible approaches to create knowledge, accommodating various information types and sources:

### Manual Input

When you have specific information that isn't contained in existing documents, you can manually create and structure your knowledge:

1. Navigate to the Knowledge section in your Relevance AI dashboard
2. Click "Create Knowledge"
3. Select "Empty Table" as your source
4. Name your knowledge base
5. Add columns to structure your information (e.g., Question, Answer, Category)
6. Enter your information directly into the table
7. Click "Create" to save your knowledge base

This approach gives you complete control over the structure and content of your knowledge base, allowing you to create precisely tailored information resources.

### Upload Documents

When your information already exists in document form, you can quickly import it:

1. Navigate to the Knowledge section
2. Click "Create Knowledge"
3. Select "Upload Files" as your source
4. Choose the file(s) you want to upload (supported formats include CSV, PDF, Excel, JSON, and audio files)
5. Configure parsing options if needed
6. Name your knowledge base
7. Click "Create" to process and import your documents

Relevance AI automatically processes these documents, extracting the relevant information and making it searchable for your agents.

### Extract Website Content

To incorporate information from websites:

1. Navigate to the Knowledge section
2. Click "Create Knowledge"
3. Select "Website" as your source
4. Enter the URL(s) you want to extract content from
5. Configure crawling options (depth, specific pages, etc.)
6. Name your knowledge base
7. Click "Create" to fetch and process the website content

This method is perfect for incorporating information from your company website, documentation sites, or other online resources without manually copying information.

### Connect Third-Party Integrations

You can use third-party integrations as integrated Knowledge sources to ensure your Agents have access to the most current information across your entire technology stack. You can learn more about each of our supported integrated Knowledge sources below.

<Card title="Google Drive" href="/build/knowledge/integrated-knowledge-sources/google-drive" icon="google-drive">Use Google Drive as a source of Knowledge for your Agents</Card>

<Card title="Sharepoint" href="/build/knowledge/integrated-knowledge-sources/sharepoint" icon="microsoft">Connect to Sharepoint as a source of Knowledge for your Agents</Card>

<Card title="Notion" href="/build/knowledge/integrated-knowledge-sources/notion" icon="notion">Use Notion as a source of Knowledge for your Agents</Card>

## Organizing Your Knowledge

Effective knowledge organization is crucial for ensuring your agents can quickly find and use the right information. Here are strategies for organizing your knowledge in Relevance AI:

### Structured Tables

For FAQ-style knowledge or structured data:

1. Create clear column headers that describe the information (e.g., "Question," "Answer," "Category")
2. Use consistent formatting within each column
3. Include metadata columns to help with filtering and organization
4. Consider adding tags or categories to group related information

### Document Chunking

When importing documents:

1. Configure appropriate chunking settings to break down large documents into manageable pieces
2. Balance chunk size - too small may lose context, too large may dilute relevance
3. Consider overlapping chunks to maintain context across sections
4. Use metadata to maintain relationships between chunks from the same source

### Knowledge Collections

Group related knowledge bases into collections:

1. Create separate knowledge bases for different domains or purposes
2. Use consistent naming conventions for your knowledge bases
3. Consider the scope of each knowledge base - more focused bases often yield more precise results
4. Document the purpose and contents of each knowledge base for team reference

## Enhancing Knowledge Quality

To maximize the effectiveness of your knowledge bases:

### Metadata Enrichment

Add contextual metadata to your knowledge entries:

1. Include creation and update dates to track freshness
2. Add author information for accountability
3. Tag entries with relevant categories or topics
4. Include confidence scores or verification status where applicable

### Knowledge Verification

Implement processes to ensure accuracy:

1. Regularly review and update your knowledge bases
2. Set up workflows for subject matter experts to verify information
3. Track usage patterns to identify frequently accessed or potentially problematic entries
4. Implement version control for important knowledge bases

### Semantic Organization

Enhance retrievability with semantic approaches:

1. Use consistent terminology throughout your knowledge bases
2. Consider adding synonyms or alternative phrasings for important concepts
3. Structure information in a question-answer format where appropriate
4. Include examples and use cases to illustrate concepts

## Connecting Knowledge to Agents

Once your knowledge is created and organized, connect it to your agents:

1. Navigate to your agent's configuration page
2. Select "Add existing knowledge"
3. Select the knowledge set you would like to add
4. Select how the agent should use this knowledge
   1. Add all to prompt: this option adds the entire knowledge set directly into the prompt every time the agent runs.
      * *Best for small or simple data sets*
      * *Easier to use with no extra configuration needed*
   2. Allow agent to search: this uses Retrieval-Augmented Generation (RAG), where the agent searches the knowledge set using a tool.
      * *Best for large or complex data sets*
      * *More flexible and scalable*
5. Depending on the option you chose, you can now see the name of your knowledge set as a  tool or knowledge set variable in your agent prompt.

## Advanced Knowledge Configuration

Fine-tune how your agents interact with knowledge:

### Retrieval Settings

Adjust how your agent searches for information:

1. Configure the number of results returned for each query
2. Set relevance thresholds to ensure only highly relevant information is used
3. Adjust search algorithms between keyword-based and semantic approaches
4. Configure whether the agent should use knowledge for all queries or only specific types

### Knowledge Prioritization

When using multiple knowledge bases:

1. Set priority levels for different knowledge sources
2. Configure fallback behavior when information isn't found in primary sources
3. Set up rules for resolving conflicts between different knowledge sources
4. Consider creating specialized knowledge bases for specific query types

### Query Transformation

Enhance retrieval effectiveness:

1. Configure query preprocessing to improve match quality
2. Set up query expansion to include related terms
3. Implement query classification to route different types of questions to appropriate knowledge bases
4. Use feedback loops to improve retrieval quality over time

## Best Practices for Knowledge Management

To maintain effective knowledge bases over time:

1. **Regular Updates**: Schedule periodic reviews to keep information current
2. **Version Control**: Track changes to your knowledge bases, especially for critical information
3. **Usage Analytics**: Monitor how your agents use knowledge to identify gaps or improvement opportunities
4. **Feedback Integration**: Create processes to incorporate user feedback into your knowledge bases
5. **Documentation**: Maintain clear documentation about your knowledge structure and organization
6. **Quality Control**: Implement validation processes for new knowledge entries
7. **Scalable Structure**: Design your knowledge organization to accommodate growth

## Related Features

* [Access Knowledge](https://relevanceai.com/docs/build/knowledge/access-knowledge): Learn how to configure how your agents and tools access and use your knowledge bases.
* [Enrich with Tool](https://relevanceai.com/docs/build/knowledge/enrich-with-tool): Discover how to enhance your knowledge bases with additional context using tools.
* [Advanced Knowledge Retrieval](https://relevanceai.com/docs/build/knowledge/advanced-knowledge-retrieval): Explore advanced techniques for optimizing how your agents retrieve and use knowledge.


# Enrich with Tool
Source: https://relevanceai.com/docs/build/knowledge/enrich-with-tool

Run a Tool across rows in your Knowledge table to add AI-generated columns

Enrich with Tool lets you run any Tool across your Knowledge table and write the results back as new columns. It's useful any time you want to process, analyze, or generate data at scale — for example, summarizing a column of text, classifying records, or extracting structured fields from unstructured content.

## Enriching your data with a Tool

There are two ways to start an enrichment on a Knowledge table.

### From the Knowledge table

<div>
  <iframe />
</div>

1. Navigate to **Knowledge** and open the table you want to enrich.
2. Click the **Enrich** button in the table toolbar.
3. Select a Tool from the library.

### From the Tool's Use tab

On the **Use** tab of any Tool, select the **Run in bulk** tab.

<img alt="Run in bulk tab on the Tool's Use page" />

Click **Run on my data** to go directly to your Knowledge tables and start the enrichment from there.

## Configuring the enrichment

Once you've selected a Tool, configure how it should run on your data. The modal shows **Map tool inputs to columns** with a dropdown for each input.

<img alt="Input mapping — map to a table column or manually type data" />

* **Map to table column** — click the dropdown to see your table's column headers and select the one to use. The value from that column will be used for each row.
* **Manually type data** — click **+ Manually type data** to switch to manual entry and use the same fixed value for every row.

Once your inputs are mapped, click **Confirm** to finish the setup.

## Running the enrichment

After confirming, a new column is added to your Knowledge table showing the Tool name. From there you can choose to run on individual rows or across the entire dataset.

<img alt="Enrichment column with run buttons" />

* The **play button in the column header** runs the Tool across the whole table.
* The **play button on each row** runs the Tool on that individual row.

<Tip>
  Always test on a few individual rows before running across the whole table to confirm the output is correct and check the credit cost.
</Tip>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can I use a Tool I built myself, or only Tools from the library?">
    You can use any Tool — including ones you've built yourself. The library shows all Tools available in your project alongside marketplace Tools.
  </Accordion>

  <Accordion title="How do I know what a bulk run will cost before I commit?">
    The configuration modal shows an estimated credit cost before you click **Confirm**. Use the single-row run to get a more precise sense of cost per row before scaling up.
  </Accordion>

  <Accordion title="Can I have more than one enrichment column on the same table?">
    Yes. Each enrichment is independent — click **Enrich** again to set up another Tool on the same table. Each creates its own column.
  </Accordion>

  <Accordion title="What happens to rows I add to the table after the enrichment is set up?">
    New rows are not automatically enriched. You'll need to run the Tool on them manually using the play button on each row, or trigger another bulk run.
  </Accordion>

  <Accordion title="Can I run the enrichment on a specific subset of rows?">
    Not as a selection — but you can run the play button on any individual row without affecting others. For targeted bulk runs on a subset, consider filtering your data into a separate table first.
  </Accordion>
</AccordionGroup>


# Find and Use Knowledge
Source: https://relevanceai.com/docs/build/knowledge/find-and-use-knowledge

Knowledge is the foundation of intelligent AI agents in Relevance AI.

## Understanding Knowledge in Relevance AI

Knowledge in Relevance AI refers to any information that your agents can access and reference when responding to queries or performing tasks. This information can come in various forms:

* **Factual information** stored in knowledge bases
* **Contextual examples** that demonstrate desired outputs
* **External data sources** connected through integrations
* **Website content** extracted from specific URLs

Think of knowledge as the reference material your agents consult before formulating responses. Just as a human expert might refer to documentation, databases, or examples before answering a question, your AI agents can leverage knowledge to ground their responses in factual information.

## Finding Knowledge

There are several ways to find and access knowledge within Relevance AI:

### Navigating to the Knowledge Section

1. Log into your Relevance AI dashboard
2. Click on "Knowledge" in the main navigation menu
3. Browse existing knowledge bases or create new ones

### Searching for Specific Knowledge

Within the Knowledge section, you can:

* Use the search bar to find specific knowledge bases by name or content
* Filter knowledge bases by type, creation date, or other attributes
* Sort knowledge bases to quickly locate what you need

### Exploring Knowledge Categories

Knowledge bases in Relevance AI are typically organized into categories based on their purpose:

* **Reference Knowledge**: Comprehensive information repositories like FAQs, product details, or company policies

* **Example Knowledge**: Samples of desired outputs that demonstrate "what good looks like"

* **Procedural Knowledge**: Step-by-step instructions for completing specific tasks

* **External Knowledge**: Information retrieved from integrated tools and services

## Using Knowledge with Your Agents

Once you've found the knowledge you need, there are several ways to use it with your agents:

### Connecting Knowledge Bases to Agents

1. Navigate to the agent you want to enhance
2. Select "Knowledge" from the agent configuration options
3. Choose the knowledge base(s) you want to connect
4. Configure retrieval settings to determine how the agent accesses the knowledge

### Providing Contextual Examples

For tasks where you want to guide your agent's output style or format:

1. Create a knowledge base with exemplary outputs
2. Connect this knowledge base to your agent
3. Include instructions that direct the agent to follow these examples

For instance, if you want your agent to write case studies in a specific format, provide examples of well-crafted case studies in a knowledge base.

### Leveraging Knowledge in Agent Prompts

When designing agent prompts, you can explicitly instruct the agent to reference specific knowledge:

> Before responding, please check the knowledge base for information about \[specific topic].

This ensures your agent actively seeks relevant information rather than relying solely on its pre-trained knowledge.

## Knowledge Retrieval Methods

Relevance AI offers several methods for retrieving knowledge:

### Semantic Search

Semantic search finds information based on meaning rather than exact keyword matches. This allows agents to retrieve relevant information even when queries use different terminology than what's in the knowledge base.

To optimize for semantic search:

* Structure your knowledge with clear, descriptive content

* Include synonyms and alternative phrasings for important concepts

* Organize related information together for better contextual understanding

### Hybrid Search

Hybrid search combines semantic search with keyword matching for more precise results. This is particularly useful when your knowledge base contains specific terminology or unique identifiers.

### Filtered Retrieval

You can set up filters to narrow down knowledge retrieval based on specific attributes:

1. Navigate to the knowledge base settings
2. Configure retrieval filters based on metadata fields
3. Apply these filters to ensure agents only access relevant information

## Knowledge Types and Their Uses

Different types of knowledge serve different purposes:

### Snippet Knowledge

Small pieces of information that provide specific guidance or examples:

* Use for: Style guides, tone examples, formatting templates
* Best for: Shaping how agents respond rather than what they respond with
* Example: "This is what a professional email response should look like..."

### Database Knowledge

Comprehensive collections of information organized in a structured format:

* Use for: FAQs, product catalogs, support documentation
* Best for: Providing detailed, factual information on a wide range of topics
* Example: A table of product specifications and features

### Integration Knowledge

Information retrieved from external tools and services:

* Use for: Real-time data, website content, third-party databases
* Best for: Ensuring agents have access to the most current information
* Example: Documentation pulled from your company website

## Best Practices for Knowledge Management

To get the most out of knowledge in Relevance AI:

### Organize Knowledge Strategically

* Create separate knowledge bases for different topics or departments
* Use clear, descriptive names for easy identification
* Add metadata to help with filtering and organization

### Update Knowledge Regularly

* Schedule regular reviews of your knowledge bases

* Update information as products, policies, or procedures change

* Remove outdated information to prevent confusion

### Monitor Knowledge Usage

* Review which knowledge bases are being accessed most frequently
* Identify gaps where agents might need additional information
* Refine knowledge based on actual usage patterns

### Balance Comprehensiveness with Relevance

* Include enough detail for accurate responses
* Avoid overwhelming knowledge bases with unnecessary information
* Focus on quality over quantity

## Troubleshooting Knowledge Retrieval

If your agents aren't effectively using knowledge:

### Check Knowledge Connection

Ensure the knowledge base is properly connected to the agent and that retrieval settings are configured correctly.

### Review Knowledge Content

Examine whether the knowledge base contains the information the agent needs in a format it can effectively use.

### Adjust Retrieval Settings

Modify search parameters, result limits, or relevance thresholds to improve retrieval accuracy.

### Test with Sample Queries

Use the testing interface to see how the agent retrieves and uses knowledge for specific questions.

## Frequently asked questions (FAQs)

**Q: How much knowledge can I add to my agents?**

A: Relevance AI supports extensive knowledge bases, but for optimal performance, focus on providing relevant, well-structured information rather than simply maximizing volume.

**Q: Can I use multiple knowledge bases with a single agent?**

A: Yes, you can connect multiple knowledge bases to one agent. The agent will search across all connected knowledge bases when responding to queries.

**Q: How do I know if my agent is using the knowledge effectively?**

A: Use the agent testing interface to see which knowledge sources are being retrieved for specific queries. You can also review conversation logs to see how knowledge is being incorporated into responses.

**Q: Can I prioritize certain knowledge over others?**

A: Yes, you can configure retrieval settings to prioritize specific knowledge bases or types of information based on relevance scores or metadata attributes.

**Q: What's the difference between knowledge and tools?**

A: Knowledge provides information for agents to reference, while tools enable agents to perform actions or access external systems. Knowledge informs what agents know, while tools expand what agents can do.


# Inactive table deletion
Source: https://relevanceai.com/docs/build/knowledge/inactive-table-deletion

One-time deletion of Knowledge tables on inactive accounts on May 4, 2026

<Info>
  This is a one-time cleanup event scheduled for May 4, 2026 — not an ongoing or recurring policy. If this becomes a recurring policy in the future, this page will be completely rewritten to reflect that.
</Info>

On May 4, 2026, Relevance AI will perform a one-time deletion of Knowledge tables on accounts that have been inactive for an extended period. This page explains what qualifies as inactive, how you are notified, and how to prevent your Knowledge tables from being deleted.

<Warning>
  Deletion of inactive Knowledge tables is permanent and cannot be undone. All data stored in those tables will be lost.
</Warning>

## What counts as inactive

Your account is considered inactive when **both** of the following conditions have been true for the last 182 days:

* No tool or agent has been run
* No login to the Relevance AI platform

Both conditions must apply simultaneously. If either condition is not met — for example, if you have logged in within the last 182 days — your account is considered active and your Knowledge tables will not be deleted.

## Warning email

You will receive an email warning 4 days before your Knowledge tables are scheduled for deletion on May 4, 2026. This email comes from `support@relevanceai.com` and is a legitimate notification from Relevance AI — not spam or phishing.

## What gets deleted

All data stored in your Knowledge tables will be permanently deleted, including all rows, columns, and any documents or files used to populate those tables. Your account itself will remain active — only the Knowledge stored in those tables is removed.

## How to prevent deletion

Log in to Relevance AI and run any Agent or Tool. Either action marks your account as active, and your Knowledge tables will not be deleted in this one-time cleanup. You do not need to re-upload data or reconfigure anything — simply using the platform is enough.

If you receive a warning email, log in and run an Agent or Tool before the 4-day window closes to prevent deletion.

## Received the email in error?

If you believe you received a deletion warning in error — for example, if you have been actively using the platform — reply directly to the warning email to open a support ticket. You can also [contact support](/docs/get-started/support) directly.

***

## Frequently asked questions

<AccordionGroup>
  <Accordion title="Does this apply to Enterprise customers?">
    No. This one-time cleanup event does not apply to Enterprise accounts. Only non-Enterprise accounts are subject to this deletion.
  </Accordion>

  <Accordion title="Does this apply to all accounts?">
    No, not all accounts. This one-time cleanup applies to non-Enterprise Relevance AI accounts that have been inactive for 182 or more days. Enterprise accounts are not affected.
  </Accordion>

  <Accordion title="Can I recover deleted Knowledge tables?">
    No. Once Knowledge tables are deleted, the data cannot be recovered. If you receive a warning email, log in and run an Agent or Tool before the 4-day window closes.
  </Accordion>

  <Accordion title="What counts as 'running' a tool or agent?">
    Any execution of a Tool or Agent through the Relevance AI platform counts as activity. This includes manual runs as well as automated or scheduled runs.
  </Accordion>

  <Accordion title="How is inactivity measured?">
    Inactivity is measured from two signals: the last tool or agent run, and the last login to the platform. Both must be absent for 182 consecutive days for your Knowledge tables to qualify for deletion in this one-time event. A single login or run within that window is sufficient to prevent it.
  </Accordion>

  <Accordion title="Is this related to the Enterprise Data Retention feature?">
    No. These are two separate things. This one-time cleanup targets Knowledge tables on non-Enterprise accounts with 182 days of inactivity. The [Data Retention](/docs/enterprise/data-retention) feature is an Enterprise-only configuration for managing the lifecycle of tool runs, agent conversations, and other activity data — it does not control Knowledge tables.
  </Accordion>

  <Accordion title="Will my account be deleted?">
    No. Your account will remain active. Only the Knowledge data stored in inactive tables is deleted in this one-time event. You can continue using Relevance AI normally after May 4, 2026.
  </Accordion>
</AccordionGroup>


# Google Drive as a Knowledge Source
Source: https://relevanceai.com/docs/build/knowledge/integrated-knowledge-sources/google-drive

Use Google Drive as a source of Knowledge for your Agents

<img alt="Marketing image of Google Drive as a source of Knowledge" />

Google Drive can be added as an integrated Knowledge source via the Agent Builder for your Agents to read files and use them as Knowledge.

You can select specific Google Drive files for your Agent to read as Knowledge sources.

<Note>You can also trigger your Agents automatically when files in Google Drive are created, modified, or deleted. [Learn more about Google Drive triggers](/docs/integrations/popular-integrations/google-drive).</Note>

## Set up Google Drive as a Knowledge source

<div>
  <iframe title="How to set up Google Drive as a source of Knowledge" />
</div>

To set up Google Drive as a Knowledge source, first, open an Agent up and head to the Build screen. Then...

1. Click 'Knowledge'
2. Click 'Add Google Drive'
3. Select or add the Google Drive account you want to access
4. Select the Google Drive files you wish to have the Agent read
5. Once you've selected the files you want to add as Knowledge, click 'Create'
6. Choose how you want the Agent to use this Knowledge on the next screen

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Is there a file size limit for Google Drive files added to Knowledge?">
    Yes, there is a limit of 50MB. If you try to select a file that exceeds the 50mb limit, you'll receive an error message indicating that the file is too large. To resolve this, you'll need to either compress the file to meet the size requirement or select a different file that is within the limit.
  </Accordion>

  <Accordion title="How often do agents fetch data from Google Drive?">
    Agents automatically fetch data from the selected Google Drive files every 15 minutes. This ensures that your agents have access to the most up-to-date information available, helping them make informed decisions in real-time.
  </Accordion>

  <Accordion title="What should I do if my agent fails to fetch data from Google Drive?">
    If your Agent fails to fetch data, you'll see an error notification indicating the failed fetch. You can check the specific file that caused the issue and ensure it's accessible and within the size limit. If the problem persists, try reconnecting your Google Drive or [reach out to support](/docs/get-started/support).
  </Accordion>
</AccordionGroup>


# Notion as a Knowledge Source
Source: https://relevanceai.com/docs/build/knowledge/integrated-knowledge-sources/notion

Use Notion as a source of Knowledge for your Agents

Notion can be added as an integrated Knowledge source via the Agent Builder for your Agents to read pages and use them as Knowledge.

You can select specific Notion pages or databases for your Agent to read as Knowledge sources.

## Set up Notion as a Knowledge source

<div>
  <iframe title="How to set up Notion as a source of Knowledge" />
</div>

To set up Notion as a Knowledge source, first, open an Agent up and head to the Build screen. Then...

1. Click 'Knowledge'
2. Click 'Add Notion'
3. Select or connect the Notion account you want to access
4. Select the Notion files you wish to have the Agent read
5. Once you've selected the files you want to add as Knowledge, click 'Create'
6. Choose how you want the Agent to use this Knowledge on the next screen

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Are Notion databases supported?">
    Yes! You can select both Notion pages and databases as Knowledge.
  </Accordion>

  <Accordion title="How often do agents fetch data from Notion?">
    Agents automatically fetch data from the selected Notion pages every 5 minutes. This ensures that your agents have access to the most up-to-date information available, helping them make informed decisions in real-time.
  </Accordion>
</AccordionGroup>


# Sharepoint as a Knowledge Source
Source: https://relevanceai.com/docs/build/knowledge/integrated-knowledge-sources/sharepoint

Use Sharepoint as a source of Knowledge for your Agents

<img alt="Sharepoint Knowledge Pn" />

Sharepoint can be added as an integrated Knowledge source via the Agent Builder for your Agents to read files and use them as Knowledge.

You can select specific Sharepoint folders or files for your Agent to read as Knowledge sources.

## Set up Sharepoint as a Knowledge source

<div>
  <iframe title="How to set up Sharepoint as a source of Knowledge" />
</div>

To set up Sharepoint as a Knowledge source, first, open an Agent up and head to the Build screen. Then...

1. Click 'Knowledge'
2. Click 'Add Sharepoint'
3. Select or add the Sharepoint account and drive you want to access
4. Select the Sharepoint files you wish to have the Agent read
5. Once you've selected the files you want to add as Knowledge, click 'Create'
6. Choose how you want the Agent to use this Knowledge on the next screen

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Will any changes I make to my Sharepoint files be reflected in Relevance AI?">
    Yes, our automatic content syncing feature ensures that any documents you upload initially, as well as any changes made to those files in SharePoint, are automatically updated in your agent's knowledge store. This means you won't have to worry about manually syncing files every time there's an update; it all happens seamlessly in the background.
  </Accordion>
</AccordionGroup>


# Knowledge Tool steps
Source: https://relevanceai.com/docs/build/knowledge/knowledge-tool-steps

Use Tool steps to interact with your Knowledge tables

You can use Tool steps to interact with your Knowledge tables.

<Snippet />


# Use Snippets for Quick Access
Source: https://relevanceai.com/docs/build/knowledge/use-snippets-for-quick-access

Snippets transform how you manage shared text across your Relevance AI workspace.

## Overview

This powerful feature provides a centralized way to store and reference reusable text variables that can be used throughout your tools and agents. By creating a single source of truth for commonly used text elements, Snippets eliminate redundant updates and streamline your workflow.

Whether you're managing company names, standard greetings, or complex message templates, Snippets make your text management more efficient and consistent across your entire AI ecosystem.

## Key Benefits

Snippets deliver several important advantages that enhance your workflow:

* **Centralized Text Management** – Store common strings (company names, greetings, standard messages) in one accessible location.
* **Universal Availability** – Use snippets anywhere a variable is accepted, including agents and tool inputs.
* **Single-Source Updates** – Edit once and update everywhere, eliminating the need to manually change the same text across multiple tools.
* **Simplified Architecture** – Eliminate 'prop-drilling' by making shared text easily accessible throughout your workspace.

## How It Works

Snippets integrate seamlessly into your existing workflow:

1. **Define a Snippet**: Create a named text variable in the Snippets section.
2. **Reference in Tools and Agents**: Use the `snippets.` prefix followed by your snippet name to reference it anywhere variables are accepted.
3. **Update Centrally**: When you need to change the text, update it once in the Snippets section, and the change propagates everywhere it's used.

For example, you can reference a snippet using this syntax: `{{ snippets.company_name }}`

## Getting Started

Accessing and using Snippets is straightforward:

1. Navigate to the Snippets section in the left sidebar of your Relevance AI workspace
2. Click "Create New Snippet"
3. Provide a descriptive name and the text content for your snippet
4. Save your snippet
5. Reference your snippet in tools and agents using the `{{ snippets.your_snippet_name }}` syntax

## Best Practices

To get the most out of Snippets, consider these recommended practices:

* **Use Descriptive Names**: Choose clear, intuitive names that indicate the snippet's purpose (e.g., `company_name`, `support_email`, `legal_disclaimer`).
* **Categorize Related Snippets**: Use naming conventions to group related snippets (e.g., `email_header`, `email_signature`).
* **Document Usage**: Add comments within your tools and agents to indicate where snippets are being used.
* **Audit Regularly**: Periodically review your snippets to ensure they contain up-to-date information.
* **Consider Scope**: Create snippets for text that appears in multiple places or might need updating across tools.

## Use Cases

Snippets are particularly valuable in these scenarios:

* **Brand Consistency**: Maintain consistent company names, taglines, and messaging across all AI interactions.
* **Contact Information**: Store email addresses, phone numbers, and URLs that might change over time.
* **Standard Messages**: Create reusable templates for common responses, greetings, or disclaimers.
* **Compliance Text**: Manage legal disclaimers, privacy statements, or terms of service that must be consistent.
* **Personalization Elements**: Store reusable personalization templates that can be combined with dynamic data.

## Implementation Examples

### Example 1: Company Information

```
// Define snippets
snippets.company_name = "Acme Corporation"
snippets.support_email = "support@acmecorp.com"
snippets.company_tagline = "Innovating for tomorrow, today."

// Use in agent prompt
"Thank you for contacting {{ snippets.company_name }}.
Our mission is {{ snippets.company_tagline }}
For additional support, please email {{ snippets.support_email }}."

```

### Example 2: Email Templates

```
// Define snippets
snippets.email_greeting = "Hello and thank you for reaching out,"
snippets.email_signature = "Best regards,\\nThe Customer Success Team\\nAcme Corporation"

// Use in email tool
"{{ snippets.email_greeting }}

[Personalized response goes here]

{{ snippets.email_signature }}"

```

## Related Features

[Agent Settings](https://relevanceai.com/docs/agent-settings) - Centralize your agent configurations for easier management, complementing snippets by providing a single source of truth for agent details.

[Tool Version Control](https://relevanceai.com/docs/tool-version-control) - Save and track changes to your tools, working alongside snippets to create more maintainable and professional tool experiences.

## Frequently asked questions (FAQs)

**Q: Can snippets contain formatting like bold or italics?**

A: Yes, snippets can contain Markdown formatting that will be rendered appropriately when used.

**Q: Is there a size limit for snippets?**

A: While there's no strict character limit, snippets are designed for reusable text elements rather than entire documents.

**Q: Can I use variables within snippets?**

A: Snippets are primarily for static text, but you can combine them with other variables in your tools and agents.

**Q: How many snippets can I create?**

A: There's no practical limit to the number of snippets you can create in your workspace.

**Q: Can I organize snippets into folders or categories?**

A: Currently, snippets are organized in a flat structure, but you can use naming conventions to create logical groupings.


# Create a Tool
Source: https://relevanceai.com/docs/build/tools/create-a-tool

How to create your first Tool in Relevance AI.

This guide walks you through creating a Tool. For an overview of what Tools are and how they work, see [Tools](/docs/get-started/core-concepts/tools).

## Two ways to create

<CardGroup>
  <Card title="Invent" icon="wand-magic-sparkles" href="#option-1-invent-a-tool">
    Describe what you want in plain language and the platform generates a working Tool for you. Great for getting started fast — you make the tweaks.
  </Card>

  <Card title="Build from scratch" icon="hammer" href="#option-2-build-from-scratch">
    Full step-by-step control over inputs, steps, and outputs. Ideal when you have a specific workflow in mind or need precision.
  </Card>
</CardGroup>

## Option 1: Invent a Tool

1. Click **Tools** in the left sidebar
2. Click **+ New Tool**
3. Select **Invent**
4. Describe what the Tool should do — be specific about the input it receives, the action it performs, and the output you expect
5. Review the generated Tool — Inventor will set up the inputs, steps, and outputs for you
6. Test and tweak as needed

## Option 2: Build from scratch

1. Click **Tools** in the left sidebar
2. Click **+ New Tool**
3. Select **Start from scratch**
4. Give your Tool a name and description, then click **Edit agent prompt** to describe how an Agent should use the Tool
5. Add your **inputs** — what data does the Tool need to receive?
6. Add **steps** — what should the Tool do with that data? Steps include LLM prompts, API calls, integrations, code execution, and more
7. Define your **outputs** — what should the Tool return when it's done?
8. Run individual steps to test as you build
9. Give the Tool to an Agent or [share it](/docs/build/tools/share-your-tool) as a standalone form

<div>
  <iframe title="Build an AI agent from scratch in Relevance AI" />
</div>

## The Tool Builder

<img alt="The Tool Builder — switch between Use, Build, and Logs tabs." />

The Tool Builder has three tabs:

* **Use** — run and test your Tool as a shareable form
* **Build** — create your Tool by defining inputs and chaining steps together
* **Logs** — review past executions and debug issues

## Editing shared tools

When you open a tool that is used by two or more agents, a warning banner appears at the top of the editor. The banner shows how many agents use the tool and warns that your changes will affect all of them.

To edit without affecting other agents, click **Clone this tool** in the banner. This creates an independent copy you can modify and assign to specific agents.

Edit the tool directly when the change should apply across all agents using it — for example, fixing a bug or updating a shared integration. Clone the tool when you need a variation for a specific agent, or want to test changes before applying them broadly.

The banner can be minimized. Your preference is saved per tool, so minimized banners stay minimized on future visits.

## Follow along on YouTube

<Frame>
  <iframe title="Build Tools for Multiple Data Sources" />
</Frame>

## What's next?

Once you've created a Tool, explore the rest of the build guides to refine it:

* [Style guide](/docs/build/tools/customize-tool/tool-style-guide) — naming conventions and best practices for making Tools work well with Agents
* [Inputs](/docs/build/tools/customize-tool/inputs) — configure what data your Tool receives
* [Steps](/docs/build/tools/customize-tool/steps) — chain actions together into a workflow
* [Outputs](/docs/build/tools/customize-tool/outputs) — define what your Tool returns
* [Version history](/docs/build/agents/build-your-agent/version-history) — every save creates a snapshot you can restore, so you can experiment without losing working versions


# Branching
Source: https://relevanceai.com/docs/build/tools/customize-tool/branching

Add conditional branches to Tool steps in your Tools

Conditional branching allows users to create different paths in their tool steps based on specific conditions. When you set up a branch, you can define criteria using rules, code, AI prompts, or fallback options. Each branch is assessed independently, meaning multiple branches can run if their conditions are met. This feature enhances the flexibility of your tools, allowing for more complex workflows.

## How to set up Branching in a Tool

<div>
  <iframe title="How to set up Branching in a Tool" />
</div>

1. Add a Branch by clicking the button to create a new Tool step, then navigate to the Flow Controls tab to select a Branch
2. Add new branches by clicking '+ Add branch'
3. For each branch, choose the Assessment mode for your branch

* **Rules**: allows you to make a logical assessment between inputs, such as contains, greater than, etc.
* **Code expression**: will run based on whether the code returns true or false
* **Let AI decide**: lets you write a plain text prompt, which returns true or false to run the branch based on your instructions. You can include variables using the variable selector menu.

4. Include a branch with the Assessment mode set to 'Fallback', this means the step will run if no other steps return true.
5. Add Tool steps to each of your branches, then run the Tool to test

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What happens if there's an error in one of the branches?">
    If there's an error in any node within a branch, the system will throw an error just like it does with standard Tool steps. It's important to ensure that each node is functioning correctly to avoid disruptions in your workflow.
  </Accordion>

  <Accordion title="How many branches can I have at once?">
    You can set a maximum of four branches.
  </Accordion>

  <Accordion title="Can multiple branches run for one input?">
    All branches are assessed individually, so multiple branches can run if the assessment returns true.
  </Accordion>

  <Accordion title="Can branches access nodes from other branches?">
    No, nodes within a branch cannot access nodes from different branches. Each branch operates independently, which helps maintain clarity and organisation in your tool's logic.
  </Accordion>
</AccordionGroup>


# Inputs
Source: https://relevanceai.com/docs/build/tools/customize-tool/inputs

Inputs for tools

## What Are Inputs?

Inputs are the essential data your tool needs to do its job. Think of them as the information your tool asks for before it can perform its magic - whether that information comes from a human user, an AI agent, or another tool.

## Basic Input Types

### Text input

A simple single-line text field for short answers like names, IDs, or brief responses

### Long text input

A multi-line text area perfect for detailed descriptions, paragraphs, or longer content

### Numeric input

Accepts only numbers, ideal for quantities, ages, or any numerical data

### Checkbox

A simple yes/no toggle for binary choices (true/false, on/off)

***

## Selection Input Types

### Options dropdown

A menu with predefined choices for users to select from

### Text list

Allows entering multiple text items, typically one per line

### Table

Structured data organized in rows and columns, similar to a spreadsheet

***

## Advanced Input Types

### JSON

When you specify a JSON input, you can optionally provide a [JSON Schema](https://json-schema.org/) in the advanced options to describe the format of the JSON input. This schema describes the expected structure of the JSON input, and is used to help guide the agent to provide data in the correct format. Providing a schema helps to improve the accuracy of calling the tool.

<Tip>The structure outlined in `schema.json` corresponds the data format used in `example.json`.</Tip>

<CodeGroup>
  ```json example.json theme={null}
  {
    "timestamp": 1948832903,
    "response": {
      "name": "Pierre Vinken",
      "age": 30,
      "bio": "Pierre Vinken, 61 years old, joined the board as a nonexecutive director Nov. 29th"
    },
    "is_prospect": true,
    "status": "inactive"
  }
  ```

  ```json schema.json theme={null}
  {
    "properties": {
      "timestamp": {
        "type": "number"
      },
      "response": {
        "type": "object",
        "properties": {
          "name": {
             "type": "string"
          },
          "age": {
             "type": "number"
          },
          "bio": {
             "type": "string"
          }
        }
      },
      "is_prospect": {
        "type": "boolean"
      },
      "status": {
        "type": "string",
        "enum": ["inactive", "active", "archived"]
      }
    }
  }
  ```
</CodeGroup>

The schema describes the overall structure of the data and the types that each item can be. Use the documentation at [json-schema.org](https://json-schema.org/) for more options.

<Warning> Note that the only supported keywords are `properties` - all other schema keywords (e.g. `additionalProperties`, `examples`, etc.) will be ignored.</Warning>

### List of JSONs

Handles multiple JSON objects in a list structure

### Knowledge table

For inputting or selecting from predefined knowledge bases

## File Handling

### File to text

Uploads a file and converts its contents to text

### File to URL

Uploads a file and provides a URL to access it

### Multiple files to URLs

Same as above but for multiple files at once

***

## Authentication Types

### API key

Securely input API keys for external services

### OAuth account

Handles secure connection to third-party services with proper authorization. OAuth account inputs allow users to connect their accounts from third-party services (like HubSpot, Slack, Google Sheets, etc.) to your tool.

OAuth account inputs can be used in tools that support OAuth authentication, or in Python code steps. For details on using OAuth accounts in Python code, see the [Python Tool step documentation](/docs/build/tools/tool-steps/python-code/code-python#oauth-authentication).

***

## Creating an Input

To add an input to your tool:

1. Select the appropriate input type from the options
2. Give your input a descriptive name
3. Add a helpful description so users know what to enter
4. Assign a variable name (this is how your tool will reference this input)

<div>
  <iframe title="Build an AI agent from scratch in Relevance AI" />
</div>

***

## Using Default Values as Constants

Default values are pre-filled inputs that make your tools more efficient and easier to use.

### Why Use Default Values?

* **Create constants** - Set values that stay the same across multiple runs (API endpoints, base URLs)
* **Provide examples** - Show realistic sample data to demonstrate how the tool works
* **Save time** - Pre-fill common inputs to reduce repetitive typing and potential errors

### How to Set Default Values

1. Find the input field you want to set a default for
2. Enter your desired default value
3. Click "Set value as default"

Once set, these values will automatically appear whenever the tool runs, unless specifically changed by the user.

<div>
  <iframe title="Build an AI agent from scratch in Relevance AI" />
</div>


# Looping
Source: https://relevanceai.com/docs/build/tools/customize-tool/looping

Loop through Tool steps in your Tools

Looping allows you to Loop through a list of Tool steps in your Tools, based on a list outputted from previous steps, or an input from the Tool.

## How to set up Looping in a Tool

<div>
  <iframe title="How to set up Looping in a Tool" />
</div>

1. Add a Loop by clicking the button to create a new Tool step, then navigate to the Flow Controls tab to select a Loop
2. Select the list you want to Loop over in **List to loop**
3. Choose how to process list items, either at the same time or one by one
4. Add Tool steps inside the Loop
5. Once you've done that, run the Tool to test the Loop

## Accessing Loop items within steps

Inside a Loop, each step runs once per item in the list. You can reference the current item and its position using template variables.

### `{{foreach.item}}`

Use `{{foreach.item}}` to access the current element being processed. For example, if your list contains movie names, `{{foreach.item}}` resolves to the name of the current movie on each iteration.

### `{{foreach.index}}`

Use `{{foreach.index}}` to access the zero-based position of the current item in the list. The first item has index `0`, the second has index `1`, and so on. This is useful when you need to track position or produce numbered output.

### Referencing step outputs

To Loop over the output of a previous step, enable variable mode on the **List to loop** field and use `{{step_name.output_field}}` syntax. For example, if an earlier step named `python` returns a list in a field called `transformed`, you would enter `{{python.transformed}}` as the list source.

<Accordion title="Example: run an LLM on each item in a list">
  Suppose you have a Python step named `python` that returns a list of movie names in a field called `transformed`. You want to run an LLM step on each movie with the prompt "Write me a summary of ...".

  1. Add a Loop step and set **List to loop** to `{{python.transformed}}` (enable variable mode first).
  2. Inside the Loop, add an LLM step.
  3. In the LLM prompt, write: `Write me a summary of {{foreach.item}}`.
  4. Run the Tool. The Loop produces one row per item in the list, with each row containing the LLM's response for that movie.

  You can also reference the item's position — for example, `Item {{foreach.index}}: {{foreach.item}}` — to include a numbered label in each output.
</Accordion>

## Output format

When a Loop runs, the output is returned in a cell-based format where each row represents one run of the Loop — one per item in the list. If your list has five items, you will see five rows, each containing the output for that iteration. This lets you inspect the result for each individual item in a table structure rather than a single aggregated output.

If an item errors during processing, its row appears under the **Errors** tab of the Loop step results by default (see error handling settings below).

## Advanced settings

### How should errors be handled?

Specifies how the Loop should handle errors. Defaults to continue (skip item), but you can also choose to throw an error.

### Index of item to test Loop on

You can also choose a specific index to test with. This is helpful for debugging and also for testing your overall flow with a single item to reduce credit costs.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can I run the Loop steps in parallel, and how does that work?">
    Yes, you can choose to process items in your list either in parallel or sequentially. If you select parallel processing, all items will be processed at the same time, which can speed up execution. If you prefer to handle them one at a time, you can opt for sequential processing.
  </Accordion>

  <Accordion title="How do I handle errors when using the Loop step?">
    By default, if an item in the Loop errors, it will be skipped, and you can view the errored items in the errors tab of the Loop step results. If you prefer to be notified of errors instead, you can adjust the error handling settings to throw a tool error. This way, you can address issues as they arise.
  </Accordion>
</AccordionGroup>


# Outputs
Source: https://relevanceai.com/docs/build/tools/customize-tool/outputs

Outputs for tools

## Outputs

Outputs are the data that your tool returns. They're the result of your tool's actions.

By default, what you see will be the output of the last step in the tool, however you can also configure this.

A common reason to configure tool outputs is to control your agent's context, as everything your tool outputs will enter the context of your agent - it's a good idea to manually configure tool to only output what matters.

For example, if your tool returns the results of a Google search step, you probably don't want to dilute your agent's context with suggested searches and 'people also ask', so you should probably configure a manual output to return the results output only.

<div>
  <iframe />
</div>


# Steps
Source: https://relevanceai.com/docs/build/tools/customize-tool/steps

Steps for tools

## Steps

Steps are the individual actions that your tool performs to generate outputs. They're the building blocks of your tool's functionality.

### Adding steps

Click the `Add Step` button to insert a new action in your tool's workflow.

<div>
  <iframe />
</div>

### Naming steps

Give each step a descriptive name for easy reference.

<div>
  <iframe />
</div>

### Referencing steps

Use variables to reference outputs from previous steps in subsequent actions.

<div>
  <iframe />
</div>

### Running steps

Test individual steps to ensure they're working correctly.

<div>
  <iframe />
</div>

### Moving steps up or down

Rearrange the order of steps by moving them within the chain.

<div>
  <iframe />
</div>

### Duplicating steps

Copy existing steps to create similar actions quickly.

<div>
  <iframe />
</div>


# Style Guide
Source: https://relevanceai.com/docs/build/tools/customize-tool/tool-style-guide

Make tools perform better with agents

## Workflow considerations

Single responsibility - each tool should perform one task exceptionally well.

## Tool naming conventions

The names you give your tools are crucial, as they are added to the LLM prompt that powers the agent using your tool. Names significantly impact performance, as agents consider these strongly when deciding how and when to use your tools.

### Best practices

* Use clear, concise and descriptive names.
* Focus on the intent of the tool, not its implementation or how it performs its action.
* Be specific about the tool's purpose.
* Include an action and object, and system if using an integration.

#### Good examples

* 'Get LinkedIn company posts'
* 'Summarize Glassdoor interviews'

#### Bad examples

* 'Search LinkedIn' (too vague)
* 'LinkedIn scraper' (implementation-focused)

### Tool description guidelines

Tool descriptions should let users know what a tool will do.

**Purpose statement**: Clearly describe the tool's objective using active voice. Explain what the tool does, not how it does it. Include the primary use case or purpose of your tool.

#### Good examples

* "Get a set of recent LinkedIn posts by a company or person."
* "Scrape and summarize company interview reviews from a provided Glassdoor interview review URL."

### Using LLMs to optimize tool names and descriptions

If you're unsure how to create the optimal name and description for your tool, we recommend that you use ChatGPT-4o to help you write.

Enter what you currently have for a tool name and description and provide some information about the agent that will use your tool.

Example prompt: "You are being provided a tool name and description that might be used by an AI agent to perform research on various prospects.

Tool name: Google search

Description: Search the internet

When given the tool name and description, provide a better name and description that will allow the agent to use this tool correctly. The tool searches Google for given a query.

The tool name needs to be more action based and focused."

## Agent prompt

The agent prompt for your tool serves as a crucial guide to help AI agents use it effectively, as agents are powered by LLMs.

It is vitally important you include an agent prompt when setting up your tool.

### Best practices for agent prompts

* **Explain usage**: Clearly outline when and how the agent should use the tool.
* **Provide context**: Include any necessary background information or assumptions the agent might need to make informed decisions.

#### Good example of an agent prompt

* Tool name: Get LinkedIn company posts
* Tool description: Get a set of recent LinkedIn posts by a company or person.
* Prompt for agent to use: "Use this tool when the user requests information about recent LinkedIn activity for a specific company or person. Provide accurate results by ensuring you input a valid LinkedIn profile URL and only return posts from the past three months."

## Variables

There are three key places you will use variables in Tool Builder:

* Inputs
* Steps
* Referencing outputs

### Naming conventions

* **Descriptive and concise**: Names should describe the variable's purpose or content. Avoid vague or overly long names.
* **Use snake\_case**: Use a consistent format across all tools (e.g., recent\_posts).
* **Avoid generic terms**: Replace terms like data, info, or results with more specific identifiers.

#### Good examples

* **company\_name** – Identifies the target company for a query.
* **blog\_post\_titles** – A collection of titles extracted from a blog, indicating the specific content being stored.

#### Bad examples:

* **name** – Too vague.
* **link** – Does not specify the type of link.

### Input parameters

There are three things to consider when creating inputs:

* Input names
* Input descriptions
* Input variables

#### Input names

Use clear, actionable, and descriptive input names. The input name should be self-explanatory, enabling users or agents to understand what the input is for.

#### Input descriptions

When writing input descriptions, make sure they provide clear, concise and actionable guidance for the user or agent. Good input descriptions should reflect:

* **Context**: Explain how the input is relevant to the tool's goal.
  * **Example**: "Provide the URL of the blog page to extract content from."
* **Requirements**: Specify what format or type of input is needed.
  * **Example**: "Enter a full URL, including https\://, to ensure proper parsing."
* **Scope**: Define the boundaries or constraints for the input.
  * **Example**: "Include only URLs from the company’s official domain."
* **Purpose**: Indicate why the input is essential and what it will achieve.
  * **Example**: "This helps identify the blog content relevant to your analysis."

#### Input variables

Input variables should be written in snake\_case, matching the input name closely to ensure consistency. The variable name should be:

* **Descriptive**: A clear, concise representation of the input it corresponds to.
* **Example**: If the input name is Company name , the variable would be company\_name.
* **Readable and Scalable**: It should be easy to read, maintain, and scale, especially when you have multiple inputs.

<Accordion title="Example">
  You want to scrape the latest blog post content, titles and publication dates from a tech company’s blog page to track their content strategy trends.

  #### Tool setup

  * **Tool name**: Scrape website content
  * **Description**: Extract website content as markdown including title and publication date given a specific website url.
  * Input parameters:
    * **Input name**: Blog URL
      * **Description**: The official blog URL of the company you’re analyzing.
      * **Variable**: blog\_url
    * **Input name**: Objective of scraping
      * **Description**: Extract content, titles and publication dates of all blog posts published in the last six months. Ignore promotional or sponsored articles.
      * **Variable**: objective\_of\_scraping
</Accordion>

## Steps

Step names act as function identifiers. You should prioritize clarity and simplicity. Like tool names, include an action and object, and system if using an integration.

### Good examples

* Send email via Gmail
* Create new Notion page

### Benefits of standardised step names

* **Clarity and readability**: Immediately understand the workflow purpose.
* **Consistency**: Create a cohesive experience.
* **Easy debugging and documentation**: Troubleshooting and self-documentation become simpler.

### Step variable naming

Step variables should be descriptive and use snake\_case.

#### Good example:

* **Step name**: Send email via Gmail
* **Variable**: send\_email


# Delete a tool or tool step
Source: https://relevanceai.com/docs/build/tools/delete-tool

How to delete tools and individual tool steps

## Differences between deleting a tool and deleting a tool step

**Tool deletion vs. tool step deletion:**

* **Delete tool** (Tools section): Permanently removes the tool from your project and all agents
* **Delete a tool step** (Agent Builder → Tools): Removes a single action within an existing tool's workflow

# Delete a tool

When you no longer need a tool, you can permanently delete it from your project.

<Warning>
  Deleting a tool is permanent and cannot be undone. If an agent or workflow is using this tool, removing it will cause those agents to lose access and may result in errors during task execution. Consider checking which agents use this tool before deleting.
</Warning>

## Before you delete

Before deleting a tool, check if any agents are using it:

1. Navigate to **Agents** in the left sidebar
2. Open each agent that might use this tool
3. Go to the **Tools** section in the Agent Builder
4. Check if your tool appears in the agent's tool list

If agents are using the tool, you'll need to remove it from those agents first or assign alternative tools to avoid possible execution errors within the agent.

## Steps to delete a Tool

<div>
  <iframe title="Invite a user to a project in Relevance AI" />
</div>

1. Navigate to **Tools** in the left sidebar (not the Tools section within an agent)
2. Find the tool you want to delete
3. Click on the three-dot menu (⋮) in the top right corner
4. Select **Delete**
5. Confirm the deletion when prompted

## What happens when you delete a tool

When you delete a tool:

* The tool is permanently removed from your project
* Any agents using this tool will no longer have access to it
* Bulk runs in progress using this tool may fail
* Shared links to the tool will stop working
* The tool cannot be recovered

# Delete individual tool steps

You can remove specific steps from a tool without deleting the entire tool. This is useful when refining your tool's workflow or removing unnecessary steps.

## Steps to delete a tool step

<div>
  <iframe title="Invite a user to a project in Relevance AI" />
</div>

1. Open your tool in the **Build** tab
2. Find the step you want to remove
3. Click on the three-dot menu (⋮) in the top right corner of the step
4. Select **Delete step**
5. The step will be removed immediately

<Tip>
  Before deleting a step, check if other steps reference its output. If they do, you'll need to update those references to avoid errors.
</Tip>

## What happens when you delete a tool step

When you delete an individual tool step:

* The step is immediately removed from the tool's workflow
* Any subsequent steps that reference this step's output will show an error
* You'll need to update or remove references to the deleted step
* The deletion takes effect immediately upon saving the tool


# Share your Tool
Source: https://relevanceai.com/docs/build/tools/share-your-tool

Learn how to share your Tool with others using different UI options

When building a tool, as the owner of the tool, you have full control over how you want it to be shared with your colleagues or teammates.

By default, your tool is only accessible by you and your agents. Click on `Share` to:

1. Create a sharable link for your tool
2. Get an embeddable iframe for your tool
3. Allow others to clone a copy of your tool

<Warning>
  Enabling a sharable link, embed, or cloneable template option makes your Tool publicly accessible. The resulting link is a public URL that search engines can index and that anyone can share, so the Tool may be discovered without you sending the link yourself — anyone with it can run your Tool on your credits or clone it into their own project. Disable the option in the Tool's share settings to revoke access.
</Warning>

<div>
  <iframe title="Build an AI agent from scratch in Relevance AI" />
</div>

## Sharable link

If you share your Tool as a sharable link, people you share your Tool with will be able to run your Tool from your account.

<Warning>You will be charged for any Tool runs that those with access to your Tool perform.</Warning>

When sharing your Tool as a sharable link, you can select:

* A primary color for your form
* A guidance tip
* A Call to Action (CTA) button
* Relevance AI branding

## Embed your Tool

You can also embed your Tool onto a website so that users can run your Tool from your site / app.

<Warning>You will be charged for any Tool runs from those who run your Tool through your embedded iframe.</Warning>

When sharing your Tool as a sharable link, you can select:

* A primary color for your form
* A guidance tip
* A Call to Action (CTA) button
* Relevance AI branding

## Cloneable template

You can share a link to allow anyone with the link to clone your Tool to their Relevance AI project. This is the best option to use if you want to share your Tool with someone but do not want them using your project / credits.

Once someone clones your tool to their project, their cloned tool will not be updated with any changes you make to it.


# Tool steps
Source: https://relevanceai.com/docs/build/tools/tool-steps

Find info on our most popular Tool steps in Relevance AI

<Info>
  This section of our knowledge base is under construction. We have 2000+ integrations and we're adding to this page gradually with more info on each of our most popular Tool steps. If you're running into issues with a certain Tool step, or would like to see a doc on a certain Tool step, please [reach out to our support team](https://relevanceai.com/docs/get-started/support)!
</Info>

## LLM Tool steps

<Columns>
  <Card title="LLM" href="/build/tools/tool-steps/llms/llm-tool-step">
    Use a LLM as a step in your Tools
  </Card>

  <Card title="More to come!">
    More help docs to come!
  </Card>
</Columns>

## Popular Tool steps

<Columns>
  <Card title="Firecrawl" href="/build/tools/tool-steps/firecrawl">
    Use Firecrawl to turn websites into LLM-ready data
  </Card>

  <Card title="Extract Website Content" href="/build/tools/tool-steps/extract-website-content">
    Turn websites into text or HTML to use in your Tools and Agents
  </Card>
</Columns>

<Columns>
  <Card title="API" href="/build/tools/tool-steps/api">
    Use a Tool step to call any API
  </Card>

  <Card title="Google Search" href="/build/tools/tool-steps/google-search">
    Search Google for any query to analyse in your Tools and Agents
  </Card>
</Columns>

<Columns>
  <Card title="Python" href="/build/tools/tool-steps/python-code/code-python">
    Use a Tool step to code using Python
  </Card>

  <Card title="Delay" href="/build/tools/tool-steps/delay">
    Delay your Tools and Agents using a Tool step
  </Card>
</Columns>

## Conversion Tool steps

<Columns>
  <Card title="Convert spreadsheet to JSON" href="/build/tools/tool-steps/convert-spreadsheet-to-json">
    Use a Tool step to convert CSVs and XLSXs into JSON
  </Card>

  <Card title="Convert a file to text" href="/build/tools/tool-steps/file-to-text">
    Use a Tool step to convert supported files into text
  </Card>
</Columns>

<Columns>
  <Card title="Convert PDF to text" href="/build/tools/tool-steps/convert-pdf-to-text">
    Use a Tool step to convert PDFs into text for LLMs to read
  </Card>

  <Card title="Convert audio/video to text" href="/build/tools/tool-steps/convert-audio-video-to-text">
    Use a Tool step to convert audio and video files to text using three different models
  </Card>
</Columns>

## Knowledge Tool steps

<Snippet />

## Integration Tool Steps

### Confluence

<Snippet />

### Google Calendar

<Snippet />

### Google Docs

<Snippet />

### Google Sheets

<Snippet />

### Gmail

<Warning>This section is under construction.</Warning>

<Snippet />

### LinkedIn

<Snippet />

### Slack

<Snippet />

### Trello

<Snippet />


# Airtop - Page Information & Monitoring
Source: https://relevanceai.com/docs/build/tools/tool-steps/airtop-browser-automation/page-information

Query page content, capture screenshots, and monitor page conditions with Airtop browser automation

<Note>
  Before using Airtop browser automation tool steps, you need to configure the Airtop integration. See the [Airtop integration setup guide](/docs/integrations/popular-integrations/airtop) for instructions on connecting your Airtop account.
</Note>

The Airtop Page Information & Monitoring tool steps allow you to extract information from web pages, capture visual representations, and monitor pages for specific conditions.

## Add the 'Airtop - Query Page' Tool step

<div>
  <iframe title="Add the Confluence - Create Post Tool step to your Tool" />
</div>

You can add the 'Airtop - Query Page' Tool step to your Tool by:

1. Creating a new Tool, then clicking `+ Add Step`
2. Searching for and adding the 'Airtop - Query Page' Tool step
3. Clicking 'Expand' to see the full Tool step
4. **Window ID** (Required): Enter the identifier of the browser window containing the page. Type `{{` to select a variable from a previous step.
5. **Session ID** (Required): Enter the identifier of the browser session. Type `{{` to select a variable.
6. **Prompt to query the page** (Required): Enter a natural language prompt describing what information you want to extract from the page (e.g., "extract all product names and prices"). Type `{{` to select a variable.
7. **Use paginated extraction** (Optional): Check this box if the page contains paginated content and you want to extract data from multiple pages automatically.
8. **Output JSON schema** (Optional): Provide a JSON schema (as a string) describing the structure you want returned. When set, Airtop returns structured JSON matching the schema instead of free-form text. Airtop requires draft-07 — include `"$schema": "http://json-schema.org/draft-07/schema#"` at the top of your schema, or the request will fail with HTTP 422.
9. Clicking `Run step` to test out the Tool

The step extracts information from the page based on your prompt. The extracted data is returned and can be used in subsequent steps. When 'Output JSON schema' is set, the output is a structured JSON object matching the schema, which is useful when downstream steps expect a consistent shape.

## Add the 'Airtop - Screenshot Page' Tool step

<div>
  <iframe title="Add the Confluence - Create Post Tool step to your Tool" />
</div>

You can add the 'Airtop - Screenshot Page' Tool step to your Tool by:

1. Creating a new Tool, then clicking `+ Add Step`
2. Searching for and adding the 'Airtop - Screenshot Page' Tool step
3. Clicking 'Expand' to see the full Tool step
4. **Window ID** (Required): Enter the identifier of the browser window containing the page. Type `{{` to select a variable from a previous step.
5. **Session ID** (Required): Enter the identifier of the browser session. Type `{{` to select a variable.
6. **Capture scope** (Optional): Choose what to capture. `viewport` captures only what's currently visible, `page` captures the entire scrollable page in one shot, `scan` performs a scroll-based capture for sites that don't render correctly with `page`, and `auto` lets Airtop choose. Defaults to `auto`.
7. **Response format from Airtop** (Optional): How Airtop returns the raw screenshot before Relevance re-hosts it. `base64` returns inline image data; `url` returns a signed download URL from Airtop's storage. Either way, the step returns a Relevance-hosted URL. Defaults to `base64` for viewport scope and `url` for full-page scopes.
8. **Max scan scrolls** (Optional): Scan scope only. Maximum number of scrolls performed while capturing. Defaults to `50`.
9. **Scan scroll delay (ms)** (Optional): Scan scope only. Delay in milliseconds between scrolls. Defaults to `1000`.
10. **Overlap percentage** (Optional): Percentage of overlap between captured chunks, between `0` and `100`. Defaults to `30`.
11. **Partition direction** (Optional): How the screenshot is split into chunks: `vertical` (default, recommended for most pages), `horizontal` (for very wide pages), or `bidirectional`.
12. Clicking `Run step` to test out the Tool

The step returns `screenshot_url` (the first chunk, kept for backward compatibility) and `screenshot_urls` (an array of every chunk, all re-hosted on Relevance storage). For `scope: viewport` there is one chunk; for `scope: page` and `scope: scan` there may be several depending on page dimensions and partition direction.

## Add the 'Airtop - Monitor Page for Condition' Tool step

<div>
  <iframe title="Add the Confluence - Create Post Tool step to your Tool" />
</div>

You can add the 'Airtop - Monitor Page for Condition' Tool step to your Tool by:

1. Creating a new Tool, then clicking `+ Add Step`
2. Searching for and adding the 'Airtop - Monitor Page for Condition' Tool step
3. Clicking 'Expand' to see the full Tool step
4. **Window ID** (Required): Enter the identifier of the browser window containing the page. Type `{{` to select a variable from a previous step.
5. **Session ID** (Required): Enter the identifier of the browser session. Type `{{` to select a variable.
6. **Condition Description** (Required): Enter a description of the condition you want to monitor (e.g., "when the submit button becomes visible" or "when the text 'Success' appears on the page"). Type `{{` to select a variable.
7. **Timeout Seconds** (Optional): Enter the maximum time in seconds to wait for the condition to be met. If the condition is not met within this time, the step will fail.
8. **Check Interval Seconds** (Optional): Enter how frequently (in seconds) to check for the condition. Smaller intervals check more frequently but may use more resources.
9. Clicking `Run step` to test out the Tool

The step continuously monitors the page for the specified condition. Once the condition is met (or the timeout is reached), the step completes. This is useful for waiting for dynamic content to load or for specific page states to occur.

## Common use cases

* Extracting data from web pages for processing or storage
* Capturing a full-page screenshot as visual proof of automation actions
* Waiting for dynamic content to load before proceeding with automation
* Monitoring pages for specific states or changes

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can I use screenshots in subsequent steps?">
    Yes. Every chunk is re-hosted on Relevance storage and returned as a URL in `screenshot_urls`. Pass an entry to any step that accepts an image URL.
  </Accordion>

  <Accordion title="When should I use viewport, page, or scan capture scope?">
    Use `viewport` for the visible area only — fastest and always one chunk. Use `page` for the entire scrollable content in one Airtop call. Use `scan` for sites that don't render correctly with `page` (typically infinite-scroll or lazy-loaded pages); scan performs a scroll-based capture and returns one chunk per scroll position.
  </Accordion>

  <Accordion title="Why do I get multiple URLs in 'screenshot_urls'?">
    `scope: page` and `scope: scan` can return more than one chunk depending on page dimensions and the partition direction. `screenshot_url` always points to the first chunk for backward compatibility; iterate over `screenshot_urls` to access them all.
  </Accordion>

  <Accordion title="What conditions can I monitor for?">
    Common conditions include element visibility, text content, and element count.
  </Accordion>
</AccordionGroup>


# Airtop - Page Interaction
Source: https://relevanceai.com/docs/build/tools/tool-steps/airtop-browser-automation/page-interaction

Interact with web pages using Airtop - click elements, type text, and scroll pages for browser automation

<Note>
  Before using Airtop browser automation tool steps, you need to configure the Airtop integration. See the [Airtop integration setup guide](/docs/integrations/popular-integrations/airtop) for instructions on connecting your Airtop account.
</Note>

The Airtop Page Interaction tool steps allow you to interact with web page elements, including clicking buttons, typing text, and scrolling through pages.

## Add the 'Airtop - Click Element' Tool step

<div>
  <iframe title="Add the Confluence - Create Post Tool step to your Tool" />
</div>

You can add the 'Airtop - Click Element' Tool step to your Tool by:

1. Creating a new Tool, then clicking `+ Add Step`
2. Searching for and adding the 'Airtop - Click Element' Tool step
3. Clicking 'Expand' to see the full Tool step
4. **Window ID** (Required): Enter the identifier of the browser window containing the page. Type `{{` to select a variable from a previous step.
5. **Session ID** (Required): Enter the identifier of the browser session. Type `{{` to select a variable.
6. **Element Description** (Required): Enter a description of the element you want to click. This can be a natural language description (e.g., "the submit button") or a CSS selector/XPath. Type `{{` to select a variable.
7. **Wait Until Timeout Seconds** (Optional): Enter the maximum time in seconds to wait for the element to become clickable.
8. **Wait For Navigation** (Optional): Check this box if you want the step to wait for page navigation after clicking the element (useful when clicking links or buttons that cause page loads).
9. **Wait For Navigation Condition** (Optional): When 'Wait For Navigation' is checked, choose which event to wait for: `load`, `domcontentloaded`, `networkidle0`, or `networkidle2`. `networkidle0` and `networkidle2` are the most reliable for detecting that a post-submit page has fully settled.
10. Clicking `Run step` to test out the Tool

The step clicks the specified element on the page. If the element is not found or not clickable within the timeout period, the step will fail.

## Add the 'Airtop - Type Text' Tool step

<div>
  <iframe title="Add the Confluence - Create Post Tool step to your Tool" />
</div>

You can add the 'Airtop - Type Text' Tool step to your Tool by:

1. Creating a new Tool, then clicking `+ Add Step`
2. Searching for and adding the 'Airtop - Type Text' Tool step
3. Clicking 'Expand' to see the full Tool step
4. **Window ID** (Required): Enter the identifier of the browser window containing the page. Type `{{` to select a variable from a previous step.
5. **Session ID** (Required): Enter the identifier of the browser session. Type `{{` to select a variable.
6. **Element Description** (Required): Enter a description of the input field where you want to type. This can be a natural language description (e.g., "the email input field") or a CSS selector/XPath. Type `{{` to select a variable.
7. **Text to Type** (Required): Enter the text you want to type into the element. Type `{{` to select a variable, or enter the text directly.
8. **Press Enter Key** (Optional): Check this box to press the Enter key after typing the text (useful for submitting forms or triggering search).
9. **Press Tab Key** (Optional): Check this box to press the Tab key after typing the text. If both Enter and Tab are enabled, Enter is pressed first. Useful for advancing focus to the next field in a form.
10. **Clear Input Field** (Optional): Check this box to clear the input field's existing value before typing. Use this when filling fields that may already contain a value.
11. **Wait Until Timeout Seconds** (Required): Enter the maximum time in seconds to wait for the element to become available.
12. **Wait For Navigation** (Optional): Check this box if you want the step to wait for page navigation after typing (useful when typing triggers a page load).
13. **Wait For Navigation Condition** (Optional): When 'Wait For Navigation' is checked, choose which event to wait for: `load`, `domcontentloaded`, `networkidle0`, or `networkidle2`.
14. Clicking `Run step` to test out the Tool

The step types the specified text into the target element. If the element is not found within the timeout period, the step will fail.

## Add the 'Airtop - Scroll Page' Tool step

<div>
  <iframe title="Add the Confluence - Create Post Tool step to your Tool" />
</div>

You can add the 'Airtop - Scroll Page' Tool step to your Tool by:

1. Creating a new Tool, then clicking `+ Add Step`
2. Searching for and adding the 'Airtop - Scroll Page' Tool step
3. Clicking 'Expand' to see the full Tool step
4. **Window ID** (Required): Enter the identifier of the browser window containing the page. Type `{{` to select a variable from a previous step.
5. **Session ID** (Required): Enter the identifier of the browser session. Type `{{` to select a variable.
6. **X Scroll Amount in pixels** (Optional): Number of pixels to scroll horizontally. Positive scrolls right, negative scrolls left. Ignored when 'Scroll To Element' is set.
7. **Y Scroll Amount in pixels** (Optional): Number of pixels to scroll vertically. Positive scrolls down, negative scrolls up. Ignored when 'Scroll To Element' is set.
8. **Scroll To Element** (Optional): Natural-language description of what to scroll to (e.g., "the first name field", "the pricing section"). Takes precedence over the pixel-based X/Y values.
9. **Scroll Within** (Optional): Natural-language description of the scrollable container to scroll inside (e.g., "the modal dialog"). If left blank, the entire page is scrolled.
10. Clicking `Run step` to test out the Tool

The step scrolls the page (or a named container) by the specified pixel amounts, or scrolls to a described element. Use pixel-based scrolling for precise control such as paging through an infinite scroll list, and 'Scroll To Element' when you want to reach a named section without knowing its exact position.

## Common use cases

* Filling out forms automatically, including advancing fields with Tab and clearing pre-filled values before typing
* Clicking buttons and links to navigate through a website
* Scrolling to load dynamic content, either by pixel amount or by natural-language description of the target
* Automating repetitive user interactions

## Common errors

<AccordionGroup>
  <Accordion title="Element not clickable">
    This error occurs when trying to click an element that is not currently visible or clickable. You may need to wait for the element to load or scroll to make it visible.
  </Accordion>
</AccordionGroup>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can I wait for an element to appear before clicking?">
    For dynamic content that loads asynchronously, consider using the 'Airtop - Monitor Page for Condition' step to wait for elements before interacting with them.
  </Accordion>

  <Accordion title="How do I advance focus to the next field without identifying it?">
    Enable 'Press Tab Key' on the 'Airtop - Type Text' step after typing into a field. The next focusable element receives focus, which is useful for filling multi-field forms without separately describing each input.
  </Accordion>

  <Accordion title="How do I overwrite a pre-filled input field?">
    Enable 'Clear Input Field' on the 'Airtop - Type Text' step. The existing value is cleared before the new text is typed, which is more reliable than simulating Ctrl+A then Delete because it works regardless of the current cursor position or selection state.
  </Accordion>

  <Accordion title="When should I use 'Scroll To Element' vs pixel-based scrolling?">
    Use 'Scroll To Element' when you want to reach a named section or element and don't know its exact pixel position. Use the X/Y pixel inputs when you need precise control over the scroll distance, such as paging through an infinite scroll list a fixed amount at a time.
  </Accordion>

  <Accordion title="What does 'Wait For Navigation Condition' do?">
    When 'Wait For Navigation' is enabled on the click or type step, the condition controls which page-load event the step waits for. `load` and `domcontentloaded` complete earlier, while `networkidle0` and `networkidle2` wait for network activity to settle and are more reliable for detecting that a post-submit page has fully rendered.
  </Accordion>
</AccordionGroup>


# Airtop - Window Management
Source: https://relevanceai.com/docs/build/tools/tool-steps/airtop-browser-automation/window-management

Manage browser windows with Airtop - create windows, load URLs, and close windows for browser automation

<Note>
  Before using Airtop browser automation tool steps, you need to configure the Airtop integration. See the [Airtop integration setup guide](/docs/integrations/popular-integrations/airtop) for instructions on connecting your Airtop account.
</Note>

The Airtop Window Management tool steps allow you to control browser windows for automation workflows. These steps include creating new browser windows, loading URLs, and closing windows.

## Add the 'Airtop - Create Window' Tool step

<div>
  <iframe title="Add the Confluence - Create Post Tool step to your Tool" />
</div>

You can add the 'Airtop - Create Window' Tool step to your Tool by:

1. Creating a new Tool, then clicking `+ Add Step`
2. Searching for and adding the 'Airtop - Create Window' Tool step
3. Clicking 'Expand' to see the full Tool step
4. **Initial URL** (Optional): Enter a URL to load immediately when the window is created. Type `{{` to select a variable, or leave blank to create an empty window.
5. Clicking `Run step` to test out the Tool

The step creates a new browser window that can be referenced using the returned window ID, session ID, and live view URL in subsequent steps.

## Add the 'Airtop - Load URL' Tool step

<div>
  <iframe title="Add the Confluence - Create Post Tool step to your Tool" />
</div>

You can add the 'Airtop - Load URL' Tool step to your Tool by:

1. Creating a new Tool, then clicking `+ Add Step`
2. Searching for and adding the 'Airtop - Load URL' Tool step
3. Clicking 'Expand' to see the full Tool step
4. **Window ID** (Required): Enter the identifier of the browser window where you want to load the URL. Type `{{` to select a variable from a previous step (e.g., the window ID returned from 'Create Window').
5. **Session ID** (Required): Enter the identifier of the browser session. Type `{{` to select a variable.
6. **URL** (Required): Enter the website URL you want to navigate to. Type `{{` to select a variable, or enter the URL directly (e.g., `https://example.com`).
7. Clicking `Run step` to test out the Tool

The step navigates the specified browser window to the provided URL.

## Add the 'Airtop - Close Window' Tool step

<div>
  <iframe title="Add the Confluence - Create Post Tool step to your Tool" />
</div>

You can add the 'Airtop - Close Window' Tool step to your Tool by:

1. Creating a new Tool, then clicking `+ Add Step`
2. Searching for and adding the 'Airtop - Close Window' Tool step
3. Clicking 'Expand' to see the full Tool step
4. **Window ID** (Required): Enter the identifier of the browser window you want to close. Type `{{` to select a variable from a previous step.
5. **Session ID** (Required): Enter the identifier of the browser session. Type `{{` to select a variable.
6. Clicking `Run step` to test out the Tool

The step closes the specified browser window. Make sure the window ID exists before attempting to close it.

## Common use cases

* Opening multiple browser windows for parallel automation tasks
* Navigating to specific pages before performing actions
* Cleaning up browser windows after completing automation workflows

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can I manage multiple windows simultaneously?">
    Yes, you can create multiple windows by using different Window IDs and managing them independently throughout your automation workflow.
  </Accordion>

  <Accordion title="Do I need to create a window before loading a URL?">
    Yes, you need to create a window first using 'Airtop - Create Window' before attempting to load a URL in that window.
  </Accordion>
</AccordionGroup>


# API Tool step
Source: https://relevanceai.com/docs/build/tools/tool-steps/api

Use the API Tool step to run a custom API call to another product / service

The API Tool step allows you to run a custom API call to any another product / service. These are best used to integrate with products / services that Relevance AI doesn't have dedicated integrations for, but can be used for any product / service.

## What is an API?

Think of an API like a waiter in a restaurant. You (the user) look at the menu (a list of what's possible), place your order with the waiter (the API), and then the kitchen (the system or server) prepares the food and the waiter brings it back to you.

In the tech world, an API is a set of rules that allows one piece of software (like an app) to talk to another (like a server).

<Warning>In order to use this Tool step, we would recommend having a beginner level of understanding of what an API is, and how to use them. You can also use the Invent feature to help you use this Tool step to integrate with different products and services.</Warning>

## Add the API Tool step to your Tool

<div>
  <iframe title="Add the API Tool step to your Tool" />
</div>

You can add the API Tool step to your Tool by:

1. Creating a new Tool, then searching for the 'API' Tool step
2. Click 'Expand' to see the full Tool step
3. Select the method you wish to use for your API call
4. Enter the URL / pathway / endpoint you want to call
5. Add headers, body and URL parameters as needed for your API call
6. Click 'Run step' to run the Tool step (see [Configuring the Request Body](#configuring-the-request-body) for details on body formatting)

### Configuring the Request Body

The Body field accepts JSON data for POST, PUT, PATCH, and other HTTP methods that support request bodies. Both JSON objects and JSON arrays are supported as top-level request bodies.

**JSON Objects** (most common):

```json theme={null}
{
  "name": "John Doe",
  "email": "john@example.com",
  "role": "admin"
}
```

**JSON Arrays** (for bulk operations or APIs that require array format):

```json theme={null}
[
  {
    "id": 1,
    "status": "active"
  },
  {
    "id": 2,
    "status": "inactive"
  }
]
```

<Info>The request body is automatically serialized to JSON when sent to the API. You don't need to manually stringify the data.</Info>

**When to use objects vs. arrays:**

* Use **objects** for single-item operations (creating one user, updating one record)
* Use **arrays** for bulk operations (creating multiple items at once) or when the API specifically requires an array format (e.g., some Confluence endpoints)

## Example API Call Configurations

### Object Body Example

Most APIs accept JSON objects for single-item operations. Here's an example of creating a user:

**Method**: POST\
**URL**: `https://api.example.com/users`\
**Headers**:

```json theme={null}
{
  "Content-Type": "application/json",
  "Authorization": "Bearer YOUR_TOKEN"
}
```

**Body**:

```json theme={null}
{
  "username": "johndoe",
  "email": "john@example.com",
  "firstName": "John",
  "lastName": "Doe"
}
```

### Array Body Example

Some APIs require array bodies for bulk operations or specific endpoints. Here's an example using the Confluence API to update page properties:

**Method**: PUT\
**URL**: `https://your-domain.atlassian.net/wiki/api/v2/pages/{pageId}/properties`\
**Headers**:

```json theme={null}
{
  "Content-Type": "application/json",
  "Accept": "application/json"
}
```

**Body**:

```json theme={null}
[
  {
    "key": "property-key-1",
    "value": "property-value-1"
  },
  {
    "key": "property-key-2",
    "value": "property-value-2"
  }
]
```

<Tip>Always check the API documentation for your specific endpoint to determine whether it expects an object or array body format.</Tip>

## Advanced Settings

### Response format

Allows you to select the format of the response. You can choose from:

* string
* json
* arrayBuffer
* blob

### Cookies

Allows you to add cookies to send with the request.

### Throw error on 4xx/5xx response

When checked, throws an error when the API response contains a 4xx/5xx response code.

## Common errors

<AccordionGroup>
  <Accordion title="Missing scopes">
    The "missing scopes" API error code generally indicates that the access token or credentials used in the API request do not include the necessary permissions (scopes) required to perform the requested action. This usually indicates your API key / token is missing, or you may not have the access to perform the call you're attempting.
  </Accordion>

  <Accordion title="Invalid values">
    An "invalid values" API error generally occurs when one or more of the parameters or fields in your API request contain data that doesn't conform to the expected format, type, or allowed range defined by the API. This can also occur when your request hasn't filled out a mandatory / required field for the API call. To fix this, make sure all required fields have been added to your call, and your parameters match the expected format (e.g. strings, booleans).
  </Accordion>
</AccordionGroup>


# 'Confluence - Create Post' Tool step
Source: https://relevanceai.com/docs/build/tools/tool-steps/confluence/create-post

Create a blog post in one of your Confluence spaces

The 'Confluence - Create Post' Tool step allows you to add a new blog post to one of your Confluence spaces.

<Note>You can learn more about the Confluence integration [here](/docs/integrations/popular-integrations/confluence).</Note>

## Add the 'Confluence - Create Post' Tool step to your Tool

<div>
  <iframe title="Add the Confluence - Create Post Tool step to your Tool" />
</div>

You can add the 'Confluence - Create Post' Tool step to your Tool by:

1. Creating a new Tool, then searching for the 'Confluence - Create Post' Tool step
2. Click 'Expand' to see the full Tool step
3. Select or add your Confluence account
4. Select the Confluence space you want to create the blog post in under 'Space ID'
5. Add the title of your blog post
6. Add the content you want to add to the blog post under 'Body'
7. Use the 'Status' field to choose if you want to create the blog post published ('current') or as a draft.
8. Click 'Run Step' to test out your Tool step


# 'Confluence - Search Content' Tool step
Source: https://relevanceai.com/docs/build/tools/tool-steps/confluence/search-content

Search Confluence using CQL in Relevance AI

The 'Confluence - Search Content' Tool step allows you to search through Confluence using CQL in Relevance AI.

<Note>
  You can learn more about the Confluence integration [here](/docs/integrations/popular-integrations/confluence).
</Note>

## What is CQL?

CQL, or **Confluence Query Language**, is a powerful search language used in Confluence to retrieve content based on structured queries. It allows users to define precise criteria — such as content type, space, title, labels, or last modified date —when searching for pages, blog posts, or attachments. CQL is especially useful for advanced searches and automations, including filtering dynamic content in macros, building reports, or integrating with the Confluence Cloud REST API. Its syntax is similar to SQL, making it intuitive for users familiar with query languages. For example, `type=page AND label=marketing` would return all pages tagged with the "marketing" label.

## Add the 'Confluence - Create Post' Tool step to your Tool

<div>
  <iframe title="Add the Confluence - Create Post Tool step to your Tool" />
</div>

You can add the 'Confluence - Create Post' Tool step to your Tool by:

1. Creating a new Tool, then searching for the 'Confluence - Create Post' Tool step
2. Click 'Expand' to see the full Tool step
3. Select or add your Confluence account
4. Enter your CQL query in the 'CQL' field
5. Use the 'Max Results' to enter the max number of results you want to receive for your query
6. Click 'Run step' to test out your Tool step


# 'Convert audio/video to text' Tool steps
Source: https://relevanceai.com/docs/build/tools/tool-steps/convert-audio-video-to-text

Use three different Tool steps to convert audio/video to text

Relevance AI supports three different models for converting audio to text, and two for converting video to text. You can use one of these Tool steps which use different models:

* Deepgram (audio and video)
* AssemblyAI (audio and video)
* OpenAI (only audio)

By using these steps, you can convert your audio and video into readable text for other Tools to use.

## Add a 'Convert audio/video to text' Tool step to your Tool

<div>
  <iframe title="Add the 'Convert audio/video to text' Tool step to your Tool" />
</div>

You can add the 'Convert audio/video to text' Tool step to your Tool by:

1. Creating a new Tool, then searching for one of the 'Convert audio/video to text' Tool step
2. Click 'Expand' to see the full Tool step
3. Upload the file you want to convert
4. Click 'Run step' to see the Tool step's output!

### Deepgram steps

If you use the Deepgram Tool step, you can select an option to recognise speaker changes.

### OpenAI steps

<Warning>Using this tool step requires an OpenAI API Key.</Warning>

If you use the OpenAI Tool step, you can select a Response Type. There are two options, `Transcript only` and `Transcript + advanced metadata`.

## Common errors

<AccordionGroup>
  <Accordion title="Unsupported file format">
    An error similar to the one noted below indicates that the provided input file is not of the supported formats.

    ```
    "err_code":"Unsupported Media Type","err_msg":"..
    ```

    Use an audio/video editor to export your file to common audio/video formats and try again.
  </Accordion>

  <Accordion title="File size">
    This error only occurs for too large files (60+ MB). If you are working with large video files, use video editing tools such
    as Quicktime to extract the audio.

    ```
    Request body larger than maxBodyLength limit
    ```
  </Accordion>
</AccordionGroup>


# 'Convert PDF to text' Tool step
Source: https://relevanceai.com/docs/build/tools/tool-steps/convert-pdf-to-text

Use a Tool step to convert PDFs into text for your Agents and Tools to read

The 'Convert PDF to text' Tool step allows you to convert a PDF into text for your tools and Agents to use.

## Add the 'Convert PDF to text' Tool step to your Tool

<div>
  <iframe title="Add the Convert a Spreadsheet to JSON Tool step to your Tool" />
</div>

You can add the 'Convert PDF to text' Tool step to your Tool by:

1. Creating a new Tool, then searching for the 'Convert PDF to text' Tool step
2. Click 'Expand' to see the full Tool step
3. Add the PDF you want to convert to text
4. Click 'Run step' to run the Tool step

## Advanced Settings

### Use OCR

OCR (Optical character recognition or optical character reader) is needed for image PDFs (e.g. scanned data). This option uses more credits and therefore should only be activated for image PDFs.

<Warning>This will not return images in the PDFs you upload, but will convert text in images to text for you to use in your Tools.</Warning>

### OCR Type

This allows you to choose the model used if 'Use OCR' is selected. The options available are:

* Fast model (\~95% accuracy)
* Quality model (\~99% accuracy but slow)

### Return PDF Text as Documents (List of Texts)

This can only be selected if 'Use OCR is selected'.

Toggling this will either return the PDF text as an aggregated text or as a list of texts.

## Common errors

<Accordion title="Unsupported file type">
  The error message similar to the one below indicates that a non-PDF file was uploaded to the Tool step. Only PDF files are supported.
  `Unsupported file type: xlsx`
</Accordion>


# 'Convert a spreadsheet to JSON' Tool step
Source: https://relevanceai.com/docs/build/tools/tool-steps/convert-spreadsheet-to-json

Use a Tool step to convert CSVs and XLSXs into JSON

The 'Convert a spreadsheet to JSON' Tool step allows you to convert a CSV or XLSX file into JSON for your Tools to use.

## Add the 'Convert a spreadsheet to JSON' Tool step to your Tool

<div>
  <iframe title="Add the Convert a Spreadsheet to JSON Tool step to your Tool" />
</div>

You can add the 'Convert a spreadsheet to JSON' Tool step to your Tool by:

1. Creating a new Tool, then searching for the 'Convert a spreadsheet to JSON' Tool step
2. Click 'Expand' to see the full Tool step
3. Add the CSV or XLSX file you want to convert to JSON
4. Click 'Run step' to run the Tool step
5. Your output will be formatted - click 'Formatted' to see the JSON output if needed

## Using the Tool output

The output is a dictionary with a key `rows`.

* `spreadsheet_to_json[rows]` is a list containing as many rows as included in the uploaded file.
* Each row is a dictionary representing one row of data (i.e. keys are column headers in the uploaded spreadsheet).

Below you can see samples where the default name assigned to the step `spreadsheet_to_json` is used.

```
spreadsheet_to_json.rows
spreadsheet_to_json.rows[0]
```

## Common errors

<Accordion title="Unsupported file type">
  The error message similar to the one below indicates that a non-spreadsheet file was uploaded to the Tool step. Only CSV and XLSX files are supported.
  `Unsupported file type: pdf`
</Accordion>


# Delay Tool step
Source: https://relevanceai.com/docs/build/tools/tool-steps/delay

Delay your Agents and Tools by using the Delay Tool step

The Delay Tool step will allow you to delay your Agents and Tools by a specified number of seconds.

## Add the Delay Tool step to your Tool

<div>
  <iframe title="Add the Delay Tool step to your Tool" />
</div>

You can add the Delay Tool step to your Tool by:

1. Creating a new Tool, then searching for the 'Delay' Tool step
2. Click 'Expand' to see the full Tool step
3. Enter the number of seconds you want the Tool step to 'delay' / make your Agent / Tool wait for in the 'Delay' field
4. If you click 'Run step', the Tool will run for the number of seconds you've specified!


# Extract Website Content Tool step
Source: https://relevanceai.com/docs/build/tools/tool-steps/extract-website-content

Use the Extract Website Content Tool step to turn websites into text and HTML for your Agents and Tools

The Extract Website Content Tool step will allow you to scrape and convert a website into text or HTML files for your Agents and Tools.

## Add the Extract Website Content Tool step to your Tool

<div>
  <iframe title="Add the Firecrawl Tool step to your Tool" />
</div>

You can add the Extract Website Content Tool step to your Tool by:

1. Creating a new Tool, then searching for the 'Extract Website Content' Tool step
2. Click 'Expand' to see the full Tool step
3. Enter the URL of the website you want to scrape and extract content from in 'Website URL'
4. Under 'Method', select 'Text' or 'HTML' based on what format you want to have your website content be extracted as
5. Click 'Run step' to test out your Tool step with your inputs!

## Advanced Settings

### Model

You can choose between two models to use for this Tool step:

* Apify
* Browserless

The following Advanced Settings will then vary based on which Model you choose.

### Apify Advanced Settings

#### Scrape Type

You can select between two scrape types:

* Simple HTML (cheaper)
* Full Web Page (expensive)

#### Use proxies

You can use proxies to scrape the website - this is more expensive.

#### Max depth

The maximum number of links starting from the start URL that the crawler will recursively follow. The start URLs have a depth of 0. Capped at 10.

#### Max pages

The maximum number of pages to crawl. Capped at 100.

### Browserless Advanced Settings

These Advanced Settings can only be used if you select 'Text' as your 'Method' in the Tool step.

#### Element selector

You can specify which element from the HTML components to scrape. By default, it is set to `body`. Note that using '+ New item', you can specify a list of elements to be scrapped.

#### Extra headers

If you need to provide special information to be able to scrape a website, provide the data as a JSON object. The below object shows an example where an authentication token called `auth-token` and a `user-id` are required.

```
{
    "auth-token":"AUTHENTICATION-TOKEN",
    "user-id":"USER-ID"
}
```

## Common errors

<AccordionGroup>
  <Accordion title="Over max depth">
    The error message below indicates that the 'Max depth' value has been set to under 0 or over 10.

    ```
    Studio transformation browserless_scrape input validation error: must be <= 10 {"comparison":"<=","limit":10} /model/max_depth
    ```
  </Accordion>

  <Accordion title="Over max pages">
    The error message below indicates that the 'Max pages' value has been set to over 100.

    ```
    Studio transformation browserless_scrape input validation error: must be <= 100 {"comparison":"<=","limit":100} /model/max_pages

    ```
  </Accordion>
</AccordionGroup>

## What's the difference between this Tool step and Firecrawl?

[Firecrawl](/docs/build/tools/tool-steps/firecrawl) is another Tool step you can use to scrape and extract website content. Unlike this Tool step, it requires you to sign up for [Firecrawl](https://firecrawl.dev) and bring your Firecrawl API Key - however, it comes with more options for scraping outputs.

Both Tool steps can be used, and we recommend trying both to see which one suits your needs more for your Agents and Tools.


# 'Convert a file to text' Tool step
Source: https://relevanceai.com/docs/build/tools/tool-steps/file-to-text

Use a Tool step to convert files into text

The 'Convert a file to text' Tool step allows you to convert a supported files into text so that

## Add the 'Convert a file to text' Tool step to your Tool

<div>
  <iframe title="Add the Convert a file to text Tool step to your Tool" />
</div>

You can add the 'Convert a file to text' Tool step to your Tool by:

1. Creating a new Tool, then searching for the 'Convert a file to text' Tool step
2. Click 'Expand' to see the full Tool step
3. Add the file you want to convert to text
4. Click 'Run step' to run the Tool step

## What file types are supported?

## Advanced Settings

Supported formats are:

* PDF (no ocr)
* Word (`.doc`, `.docx`)
* Excel (`.xlsx`, `.xls`)
* CSV
* Audio (`.mp3`, `.mp4`, `.mp2`, `.aac`, `.wav`, `.flac`, `.pcm`, `.m4a`, `.ogg`, `.opu`, `.webm`)
* JSON
* TXT

<Note>Audio transcription is completed using the Deepgram model.</Note>

### LLM Friendly Format

A checkbox that, when selected, will convert your text into a format that is more friendly for LLMs.

## Common errors

<Accordion title="Unsupported file type">
  The error message similar to the one below indicates that a file that is not supported was added to the Tool step.
  `Unsupported file type: pdf`
</Accordion>


# Firecrawl Tool step
Source: https://relevanceai.com/docs/build/tools/tool-steps/firecrawl

Use Firecrawl as a step in your Tools to turn websites into LLM-ready data

The Firecrawl Tool step allows you to turn websites of your choosing into LLM-ready data that you can then feed into an LLM as part of your Tools and Agents.

## Integrating with Firecrawl

In order to use the Firecrawl Tool step, you need to sign up for Firecrawl at [firecrawl.dev](https://www.firecrawl.dev/) and get an API key. You then need to add this API key to Relevance AI.

## Add the Firecrawl Tool step to your Tool

<div>
  <iframe title="Add the Firecrawl Tool step to your Tool" />
</div>

You can add the Firecrawl Tool step to your Tool by:

1. Creating a new Tool, then searching for the 'Firecrawl Web Scraper' Tool step
2. Click 'Expand' to see the full Tool step
3. Add the URL you want to scrape in 'URL'
4. Check or uncheck 'Scrape Only' based on whether or not you want to scrape the URL you've added only
5. Check or uncheck 'Extract Main Content Only' based on whether or not you want to include nav, header, footers, etc.
6. Add paths you want to include here under 'Include Paths'
7. If 'Scrape Only' is unchecked, you can specify how many pages you want to crawl under 'Number of Pages'
8. Run the Tool step to see if works with your configuration!

## Advanced Settings

### Output Formats

You can select from the following Output Formats for your scrapes:

* Markdown (default)
* HTML
* Raw HTML
* Links
* Screenshot
* Screenshot of the full page
* JSON
* Change Tracking

## Common errors

<AccordionGroup>
  <Accordion title="Insufficient credits">
    The error message below indicates that you have run out of credits, or do not have enough credits, in your Firecrawl account to perform the Tool step you're running. You will need to upgrade your plan at Firecrawl to get more credits.

    ```
    {"success":false,"error":"Insufficient credits to perform this request. For more credits, you can upgrade your plan at https://firecrawl.dev/pricing or try changing the request limit to a lower value."}
    ```
  </Accordion>

  <Accordion title="Invalid token">
    The error message below indicates that your Firecrawl API Key is either incorrect or missing. You will need to add / replace this API Key in 'Integrations & API Keys'.

    ```
    {"error":"Unauthorized: Invalid token"}
    ```
  </Accordion>
</AccordionGroup>


# Add Label to Email
Source: https://relevanceai.com/docs/build/tools/tool-steps/gmail/add-label-to-email

The Add Label to Email tool step allows you to add a tag to an email

## Add the ‘Add Label to Email’ Tool step to your Tool

<div>
  <iframe title="Gmail Add Label to Email" />
</div>

You can add the ‘Add Label to Email’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Add Label to Email’ Tool step.
2. Click ‘Expand’ to see the full Tool step.
3. Select the account you would like to use.
4. Select the your target email to label.
5. Select the label you would like to use.
6. Click 'Run step' to test out the Tool.

## Parameters

<ParamField type="string">
  The unique identifier for the email conversation you want to label.
  <Note> For automated workflows: Use the GMAIL\_THREAD\_ID from your trigger event to target the specific email. </Note>
</ParamField>

<ParamField type="string">
  The label you would like to add to an email. If you are using this tool with input variable, you can retrieve your label IDs using the **List Labels** tool step.
</ParamField>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="How can I use this with an Agent trigger?">
    Use the `GMAIL_THREAD_ID` from the Gmail trigger as the value for the Message parameter.
  </Accordion>

  <Accordion title="How do I get a Label ID?">
    Use the “List Labels” tool step to fetch your available labels and their IDs, then select the appropriate one.
  </Accordion>
</AccordionGroup>


# List Labels
Source: https://relevanceai.com/docs/build/tools/tool-steps/gmail/list-labels

The Add Label to Email tool step allows you to add a tag to an email

## Add the ‘List Labels’ Tool step to your Tool

<div>
  <iframe title="Gmail List Labels" />
</div>

You can add the ‘List Labels’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘List Labels’ Tool step.
2. Click ‘Expand’ to see the full Tool step.
3. Select the account you would like to use.
4. Click 'Run step' to test out the Tool.


# Cancel Event in Google Calendar
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-calendar/cancel-event-in-google-calendar

The Cancel Event in Google Calendar tool step allows you to delete an event in your Google Calendar

## Add the ‘Cancel Event in Google Calendar’ Tool step to your Tool

<div>
  <iframe title="Add the cancel event in Google Calendar Tool step to your Tool" />
</div>

You can add the ‘Cancel Event in Google Calendar’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Cancel Event in Google Calendar’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Select the Google account you would like to use
4. Input the event ID of the event you want to delete
   <Note>To get the event ID, you can use a tool step like [Get Events from Google Calendar](/docs/build/tools/tool-steps/google-calendar/get-events-from-google-calendar)</Note>
5. Click 'Run step' to test out the Tool

## Advanced settings

### Google Calendar ID

Select the calendar you want to delete events from. If nothing is selected the primary calendar will be used.
<Note>To get your calendar ID, open the settings of the calendar you want to use, then go to “Integrate calendar” and copy the Calendar ID.</Note>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="How do I get the event ID?">
    You can use a tool step like [Get Events from Google Calendar](/docs/build/tools/tool-steps/google-calendar/get-events-from-google-calendar) to get an event ID.
  </Accordion>

  <Accordion title="How come im not getting a response even though the event was successfully deleted?">
    You won’t receive a response after successfully deleting a calendar event because there’s nothing more to confirm—once it’s gone, it’s gone. If you run it again on the same event ID, it will let you know that the event has already been deleted.
  </Accordion>
</AccordionGroup>


# Check Google Calendar Availability
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-calendar/check-google-calendar-availability

The Check Google Calendar Availability tool step allows you to check the availability of a calendar

## Add the ‘Check Google Calendar Availability’ Tool step to your Tool

<div>
  <iframe title="Add the Check Google Calendar Availability Tool step to your Tool" />
</div>

You can add the ‘Check Google Calendar Availability’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Check Google Calendar Availability’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Select the Google account you would like to use
4. Input the event start date<Tooltip>(ISO 8601 format)</Tooltip>
5. Input the event end date<Tooltip>(ISO 8601 format)</Tooltip>
6. Input the slot duration. This is the duration of the time slot you want to check for in minutes
7. Enter the start time of your workday in 24-hour format (e.g. 09:00)
8. Enter the end time of your workday in 24-hour format (e.g. 17:00)
9. Select your timezone
10. Click 'Run step' to test out the Tool

## Advanced settings

### Calendars and/or groups to query

Select the calendar/s you want to check the availability for. If nothing is selected the primary calendar will be used.
<Note>To get your calendar ID, open the settings of the calendar you want to use, then go to “Integrate calendar” and copy the Calendar ID.</Note>

## Common errors

<AccordionGroup>
  <Accordion title="Incorrect start/end date format">
    The start and end date must be in ISO 8601 format (YYYY-MM-DD).

    `EG: 2025-07-23`
  </Accordion>

  <Accordion title="End date is earlier than start date">
    The end date must be on or after the start date. Reversing the order will result in an error.
  </Accordion>

  <Accordion title="Calendar ID is incorrect or not found">
    Double-check that the Calendar ID is copied correctly from your calendar settings page. Invalid Calendar IDs will cause the tool to fail.
  </Accordion>
</AccordionGroup>


# Create Google Calendar Event
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-calendar/create-google-calendar-event

The Create Google Calendar Event tool step allows you to create an event in your Google Calendar

## Add the ‘Create Google Calendar Event’ Tool step to your Tool

<div>
  <iframe title="Add the Create Google Calendar Event Tool step to your Tool" />
</div>

You can add the ‘Create Google Calendar Event’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Create Google Calendar Event’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Select the Google account you would like to use.
4. Input the event title
5. Input the event description
6. Input the event start date and time <Tooltip>(ISO 8601 format)</Tooltip>
7. Input the event end date and time <Tooltip>(ISO 8601 format)</Tooltip>
8. Input the attendees email addresses in array format
   <Note>Example: `["john.doe@example.com", "jane.doe@example.com"]`</Note>
9. Select whether guests can modify the event
10. Click 'Run step' to test out the Tool

## Advanced settings

### Calendar to add event to

This is the calendar you want to add the event to. If you don't select a calendar, the event will be added to the primary calendar.

### Meeting Location

Can be a physical location or a link to a virtual meeting.

## Common errors

<AccordionGroup>
  <Accordion title="Incorrect start/end date and time format">
    The start and end date and time must be in ISO 8601 format.

    `EG: 2025-07-23T10:00:00Z`
  </Accordion>

  <Accordion title="Incorrect attendees email address format">
    Ensure that the attendees email addresses are valid and in array format.

    `EG: ["john.doe@example.com", "jane.doe@example.com"]`
  </Accordion>
</AccordionGroup>


# Get All Google Calendars Details
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-calendar/get-all-google-calendars-details

The Get All Google Calendars Details tool step allows you to get information about all your Google Calendars

## Add the ‘Get All Google Calendars Details’ Tool step to your Tool

<div>
  <iframe title="Add the Get All Google Calendars Details Tool step to your Tool" />
</div>

You can add the ‘Get All Google Calendars Details’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Get All Google Calendars Details’ Tool step
2. Select the Google account you would like to use
3. Click 'Run step' to test out the Tool

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What type of information can I get?">
    You can get the calendar ID, calendar name, description, access level and timezone of all your Google Calendars.
  </Accordion>

  <Accordion title="Does this include shared calendars?">
    Yes, this tool will return details for all calendars you have access to, including calendars that have been shared with you and calendars you've subscribed to.
  </Accordion>

  <Accordion title="Does this tool return calendar events?">
    No, this tool only returns calendar details (name, ID, description, etc.). To get calendar events, use the [Get Events from Google Calendar](/docs/build/tools/tool-steps/google-calendar/get-events-from-google-calendar) tool step.
  </Accordion>
</AccordionGroup>


# Get Events from Google Calendar
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-calendar/get-events-from-google-calendar

The Get Events from Google Calendar tool step allows you to get your upcoming events from your Google Calendar

## Add the ‘Get Events from Google Calendar’ Tool step to your Tool

<div>
  <iframe title="Add the Get Events from Google Calendar Tool step to your Tool" />
</div>

You can add the ‘Get Events from Google Calendar’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Get Events from Google Calendar’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Select the Google account you would like to use
4. Input the time to search from<Tooltip>(ISO 8601 format)</Tooltip>
5. Input the time to search to<Tooltip>(ISO 8601 format)</Tooltip>
6. Check whether your would like to include attendees and links
7. Click 'Run step' to test out the Tool

## Advanced settings

### Google Calendar ID

Select the calendar you want to get events for. If nothing is selected the primary calendar will be used.
<Note>To get your calendar ID, open the settings of the calendar you want to use, then go to “Integrate calendar” and copy the Calendar ID.</Note>

## Common errors

<AccordionGroup>
  <Accordion title="Incorrect start/end date format">
    The start and end date must be in ISO 8601 format (YYYY-MM-DD).

    `EG: 2025-07-23`
  </Accordion>

  <Accordion title="End date is earlier than start date">
    The end date must be on or after the start date. Reversing the order will result in an error.
  </Accordion>

  <Accordion title="Calendar ID is incorrect or not found">
    Double-check that the Calendar ID is copied correctly from your calendar settings page. Invalid Calendar IDs will cause the tool to fail.
  </Accordion>
</AccordionGroup>


# Get Google Calendar Event Details
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-calendar/get-google-calendar-event-details

The Get Google Calendar Event Details tool step allows you to get the details of a specific event in your Google Calendar

## Add the ‘Get Google Calendar Event Details’ Tool step to your Tool

<div>
  <iframe title="Add the Get Google Calendar Event Details Tool step to your Tool" />
</div>

You can add the ‘Get Google Calendar Event Details’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Get Google Calendar Event Details’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Select the Google account you would like to use.
4. Input the event ID
   <Note>To get the event ID, you can use a tool like [Get Events from Google Calendar](/docs/build/tools/tool-steps/google-calendar/get-events-from-google-calendar).</Note>
5. Click 'Run step' to test out the Tool

## Advanced settings

### Google Calendar ID

Input the calendar ID of the calendar you want to get the event details for. If nothing is selected the primary calendar will be used.
<Note>To get your calendar ID, open the settings of the calendar you want to use, then go to “Integrate calendar” and copy the Calendar ID.</Note>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="How do I get the event ID?">
    You can use the [Get Events from Google Calendar](/docs/build/tools/tool-steps/google-calendar/get-events-from-google-calendar) tool step to get the event ID.
  </Accordion>

  <Accordion title="How do I get the calendar ID?">
    To get your calendar ID, open the settings of the calendar you want to use, then go to “Integrate calendar” and copy the Calendar ID.
  </Accordion>

  <Accordion title="What details can I get from an event?">
    You can get the event name, description, start and end times, attendees, location, and meeting links.
  </Accordion>
</AccordionGroup>


# Append Image to Document
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-docs/append-image-to-document

The Append Image to Document tool step allows you to add an image to your doc

## Add the ‘Append Image to Document’ Tool step to your Tool

<div>
  <iframe title="Add the Append Text Tool step to your Tool" />
</div>

You can add the ‘Append Image to Document’ Tool step to your Tool by:

1. Create a ["File to URL"](/docs/build/tools/customize-tool/inputs#file-to-url) input field
2. Create a "Append Image to Document" tool step
3. Select the account you would like to use
4. Select the document you would like to append an image to.
5. Select the "File to URL" input variable you created in step 1.
6. Upload an image to the "File to URL" input field by dragging and dropping the image into the input field.
7. Check whether you would like to append at the beginning of the document.
8. Click 'Run step' to test out the Tool

## Parameters

<ParamField type="string">
  The document you would like to append an image to. You can get the document ID from the URL of the document between `/d/` and `/edit`.
</ParamField>

<ParamField type="string">
  The URL of the image you would like to append to the document.

  <Warning>
    You must upload the image using the "File to URL" input field
  </Warning>
</ParamField>

<ParamField type="boolean">
  When this option is checked, the image will be prepended to the document.
</ParamField>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can I append an image to a specific section of the document?">
    Currently, you can only append an image to the beginning or end of the document.
  </Accordion>

  <Accordion title="I keep getting an error when I try to append an image to the document?">
    Please ensure that the image is being uploaded to the "File to URL" input field.
  </Accordion>
</AccordionGroup>


# Append Text
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-docs/append-text

The Append Text tool step allows you to append text to a Google Doc

## Add the ‘Append Text’ Tool step to your Tool

<div>
  <iframe title="Add the Append Text Tool step to your Tool" />
</div>

You can add the ‘Append Text’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Append Text’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Select the account you would like to use
4. Select the document you would like to append text to.
5. Enter the text you would like to append to the document.
6. Check whether you would like to append at the beginning of the document.
7. Click 'Run step' to test out the Tool

## Parameters

<ParamField type="string">
  The document you would like to append text to. You can get the document ID from the URL of the document between `/d/` and `/edit`.
</ParamField>

<ParamField type="string">
  The text you would like to append to the document.
</ParamField>

<ParamField type="boolean">
  When this option is checked, the text will be prepended to the beginning of the document.
</ParamField>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can I append formatted text (bold, italic, lists, etc.)?">
    With the default template, you can only append plain text. However you can use this tool [Add/Update any content to Google Doc](https://app.relevanceai.com/marketplace/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/search?search=Add/Update+any+content+to+Google+Doc\&listing=e0a51c06-dccc-4eea-b8bc-40331ec41fbd) as a workaround to append formatted text.
  </Accordion>

  <Accordion title="Can I append text to a specific section of the document?">
    Currently, you can only append text to the beginning or end of the document.
  </Accordion>
</AccordionGroup>


# Create a New Document
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-docs/create-a-new-document

The Create a New Document tool step allows you to create a new document with preexisting content

## Add the ‘Create a New Document’ Tool step to your Tool

<div>
  <iframe title="Google Docs Create a New Document" />
</div>

You can add the ‘Create a New Document’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Create a New Document’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Select the account you would like to use
4. Enter the title of the document you would like to create.
5. Enter the content you would like to add to the document.
6. Select the folder you would like to create the document in.
7. Click 'Run step' to test out the Tool

## Parameters

<ParamField type="string">
  The name of the file to create.
</ParamField>

<ParamField type="string">
  The content you would like to create the document with.
</ParamField>

<ParamField type="string">
  The folder you would like to create the document in. If no folder is specified, the document will be created in the root folder.
</ParamField>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can I create a document with formatted text?">
    Currently, you can only create a new document with a blank template. However you can use this tool [Add/Update any content to Google Doc](https://app.relevanceai.com/marketplace/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/search?search=Add/Update+any+content+to+Google+Doc\&listing=e0a51c06-dccc-4eea-b8bc-40331ec41fbd) as a workaround to create a document with formatted text.
  </Accordion>

  <Accordion title="How can I get the folder ID?">
    You can get the folder ID by going to the folder in Google Docs and copying the ID from the URL. It should be the string after `folders/` in the URL.
  </Accordion>
</AccordionGroup>


# Create New Document from Template
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-docs/create-new-document-from-template

The Create New Document from Template tool step allows you to create new documents using existing documents as templates

## Add the ‘Create New Document from Template’ Tool step to your Tool

<div>
  <iframe title="Create New Document From Template" />
</div>

You can add the ‘Create New Document from Template’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Create New Document from Template’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Select the account you would like to use
4. Select the template you would like to use (this is just any existing document in your Google Drive)
5. Select the folder to save the new document in
6. Pick a name for the new document
7. Select the output file type (PDF, DOCX or both)
8. Enter the text you want to replace the placeholders with
9. Click 'Run step' to test out the Tool

## Parameters

<ParamField type="string">
  Select the document you would like to use as a template.
</ParamField>

<ParamField type="string">
  Specify the folder where the new document will be saved.
</ParamField>

<ParamField type="string">
  This will be the name of the new file.
</ParamField>

<ParamField type="options">
  This is the file type of the new document. Choose from PDF, DOCX or both.
</ParamField>

<ParamField type="json">
  This is a JSON object that contains the text placeholders and the text to replace them with.
</ParamField>

```json theme={null}
{
  // This will replace all instances of {{Name}} with "John Doe"
  "Name": "John Doe",
  // This will replace all instances of {{Date}} with "2025-01-01"
  "Date": "2025-01-01"
}
```

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Where do I get a template from?">
    You can use any existing document as a template. Add custom text to the template by using placeholders. Then, in the "Replace Text Placeholders" field, specify what each placeholder should be replaced with in the new document.
  </Accordion>

  <Accordion title="What are placeholders?">
    Placeholders are special markers in your template that get replaced with custom text when the document is created.\
    <Tip>To add a placeholder, wrap the name in double curly braces. For example: `{{Name}}` or `{{Date}}`.</Tip>
  </Accordion>

  <Accordion title="What happens if there is more than one instance of the same placeholder?">
    All instances of the same placeholder will be replaced.
  </Accordion>
</AccordionGroup>


# Get Document
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-docs/get-document

Get Document tool step allows you to retrieve the contents of a file in your Google Drive

## Add the ’Get Document’ Tool step to your Tool

<div>
  <iframe title="Google Docs Get Document" />
</div>

You can add the ‘Get Document’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Get Document’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Select the Google account you would like to use.
4. Select the Document you would like to fetch.
5. Check whether you want to include content from all tabs.
6. Select which tabs you want to include.
7. Click 'Run step' to test out the Tool

## Parameters

<ParamField type="string">
  The document you would like to append text to. You can get the document ID from the URL of the document between `/d/` and `/edit`.
</ParamField>

<ParamField type="boolean">
  If this option is unchecked, only the content from the main tab will be included in the output.
  When this option is checked, the content from all tabs will be included in the output.
  <Warning> This option must be enabled if Tab IDs are being provided. </Warning>
</ParamField>

<ParamField type="array">
  Specify which tabs should appear in the output.
</ParamField>

## Playground

<div>
  <iframe />
</div>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="How do I get the document ID?">
    To get the document ID, navigate to the document you want to include in the output and copy the ID from the URL (the long string between `/d/` and `/edit`).
  </Accordion>

  <Accordion title="How do I get the Tab IDs?">
    To locate the Tab ID, navigate to the specific tab you're interested in. You can find the tab ID in the URL after `tab=`.
  </Accordion>
</AccordionGroup>


# Replace Text
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-docs/replace-text

The Replace Text tool step allows you to replace text in a Google Doc

## Add the ‘Replace Text’ Tool step to your Tool

<div>
  <iframe title="Add the Replace Text Tool step to your Tool" />
</div>

You can add the ‘Replace Text’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Replace Text’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Select the account you would like to use
4. Select the document you would like to replace text in.
5. Enter the text you would like to replace in the document.
6. Enter your new text.
7. Check whether you want the text matching (in step 5) to be case-sensitive.
8. Click 'Run step' to test out the Tool

## Parameters

<ParamField type="string">
  The document you would like to append text to. You can get the document ID from the URL of the document between `/d/` and `/edit`.
</ParamField>

<ParamField type="string">
  Specify the exact text in the document that you want to replace. The replacement will only occur if the text matches exactly.

  <Warning>
    Be precise when entering the text to replace - all the instances of the text will be replaced.
  </Warning>
</ParamField>

<ParamField type="string">
  This is the text that will replace the existing text.
</ParamField>

<ParamField type="boolean">
  This option specifies whether the text matching should be case-sensitive.
</ParamField>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What happens if the 'Text to be replaced' is not found?">
    The tool step will still run successfully, but no changes will be made to the document.
  </Accordion>

  <Accordion title="What happens if there is more than one instance of the 'Text to be replaced'?">
    Any instances of the 'Text to be replaced' will be replaced.
  </Accordion>
</AccordionGroup>


# Google Search Tool step
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-search

Add a Google Search step to your Tools

The Google Search Tool step allows you to run a Google Search within your Tool, and use the outputs of this search in your LLM Tool steps.

## Add the Google Search Tool step to your Tool

<div>
  <iframe title="Add the Google Search Tool step to your Tool" />
</div>

You can add the Google Search Tool step to your Tool by:

1. Creating a new Tool, then click 'Google' under 'Steps'
2. Click 'Expand' to see the full Tool step
3. Add a Search Query to the field
4. Run the Tool step to see if works with your configuration!

## Advanced Settings

### Country

You can select which country / region you would like to search Google from.

### Locale

You can select which language is set for the Google Search.

## Tool step outputs

The Google Search Tool step allows you to access the organic search (as an array) completed on the search query entered, as well as:

* **People Also Ask**: you can access an array of questions searchers also ask about the query, as well as a snippet, title and link for each question
* **Related searches**: you can also access an array of related searches to the search query

If you wanted to use an LLM Tool step to then analyse these results, here is an example of how you could prompt engineer this, where `search_query` is the input variable used for the Google Search, and `google` is the variable for the Google Search Tool step.

```markdown theme={null}
You are analysing the Google results for the following query: {{search_query}}

A Google Search has already been completed for this query, and Google has returned the following info:

* People Also Ask: {{google.peopleAlsoAsk}}
* Related Searches: {{google.relatedSearches}}
* All of the latest search results: {{google.organic}}

Analyse these to give me a run down of the top results from the {{search_query}}, as well as what people are asking about this query, and related searches. Keep this analysis between 1 - 3 paragraphs. 
```


# Add Multiple Rows
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-sheets/add-multiple-rows

The Add Multiple Rows tool step allows you to add multiple rows to your Google Sheet

## Add the ‘Add Multiple Rows’ Tool step to your Tool

<div>
  <iframe title="Google Sheets Add Multiple Rows" />
</div>

You can add the ‘Add Multiple Rows’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Add Multiple Rows’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Using the dropdown menu:
   * Select the Google account you would like to use.
   * Select the Drive you would like to use (defaults to "My Drive").
   * Select the Spreadsheet you would like to add the row to.
   * Select the worksheet you would like to add the row to.
4. Enter the new Row Values. This should be an array of arrays.
   <Note>Example: `[["John Doe", "john.doe@example.com", "123 Main St"], ["Jane Doe", "jane.doe@example.com", "456 Main St"]]`</Note>
5. Check whether to reset the row formatting.
6. Click 'Run step' to test out the Tool

## Using in an Agent

There are 3 ways to configure the tool step when using in an agent

<Tabs>
  <Tab title="Using dropdown menus">
    This is the easiest way to configure the tool step. You can select the Drive, Spreadsheet and Worksheet from a list of options via the dropdown menu.

    <img alt="Google Sheets Add multiple rows" />
  </Tab>

  <Tab title="Using input variables">
    This option offers the most flexibility - you can use the same tool step for multiple Google Sheets, however you will need to manually provide the necessary Google Sheet IDs.

    <img alt="Google Sheets Add multiple rows" />
  </Tab>

  <Tab title="Mix of dropdown menus and input variables">
    This option offers the best of both worlds. You can use the dropdown menu to select the Drive, Spreadsheet and Worksheet, and then use an input variable for "Row Values" to allow the agent to dynamically insert data to your Google sheet.

    <img alt="Google Sheets Add multiple rows" />
  </Tab>
</Tabs>

## Parameters

<ParamField type="string">
  The Drive you want to use. By default, the tool uses "My Drive".
</ParamField>

<ParamField type="string">
  To get the spreadsheet ID, navigate to your Google Sheet and copy the ID from the URL (the long string between `/d/` and `/edit`).
  <Tip> [You can use this tool to parse the worksheet, spreadsheet and drive IDs from a url](https://app.relevanceai.com/notebook/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/01e4918b-6c0a-423d-8daa-a625ced25e23)</Tip>
</ParamField>

<ParamField type="string">
  To get the worksheet ID, open the specific worksheet you want to use and copy the ID from the URL (appears after `#gid=`).
  <Tip> [You can also use this tool to parse the worksheet, spreadsheet and drive IDs from a url](https://app.relevanceai.com/notebook/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/01e4918b-6c0a-423d-8daa-a625ced25e23)</Tip>
</ParamField>

<ParamField type="array of arrays">
  The data being inserted will need to be in an array of arrays format.

  **Example:**

  ```json theme={null}
  [
    ["John Doe", "john.doe@example.com", "123 Main St"], // row 1
    ["Jane Doe", "jane.doe@example.com", "456 Main St"] // row 2
    // ...
  ]
  ```
</ParamField>

<ParamField type="boolean">
  If this option is unchecked, the format will be preserved when inserting data. If this option is checked, the data will be inserted as plain text.
</ParamField>

## What is an array of arrays?

An array is a data structure similar to a CSV row, where each element is separated by a comma, and the entire row is enclosed in square brackets.
An array can have many different types of data, including strings, numbers, <Tooltip>booleans</Tooltip>, and even other arrays.

```json theme={null}
["Dog", 1, true]
```

An array of arrays is just multiple arrays put together. This is what an array of arrays looks like:

```json theme={null}
[
  ["Column 1 Row 1", "Column 2 Row 1", "Column 3 Row 1"],
  ["Column 1 Row 2", "Column 2 Row 2", "Column 3 Row 2"],
  ["Column 1 Row 3", "Column 2 Row 3", "Column 3 Row 3"]
]
```

When you combine multiple arrays together, you can see that the data is structured like a Google Sheet.

### Playground

<div>
  <iframe />
</div>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What is the difference between using the green input variables and the dropdown menu?">
    **The dropdown menu is the easiest way to configure the tool step**. You can select the Drive, Spreadsheet and Worksheet from a list of options via the dropdown menu.

    **The input variables offer more flexibility** e.g. you can use the same tool step for multiple Google Sheets, however you will need to manually provide the necessary Google Sheet IDs.
  </Accordion>

  <Accordion title="Can I use a mix of both dropdown menu and green input variables?">
    If you are using the tool in an agent, this is the recommended approach. You can use the dropdown menu to select the fields that don't need to change (the ID fields), and then use the green input variables for "Row Values" to allow the agent to dynamically retrieve data from your Google sheet.

    <Warning>If you are only using the input variables, you will need to provide the necessary Google Sheet IDs for the tool step to function properly. At the minimum, this includes the spreadsheet ID and the worksheet ID.</Warning>
  </Accordion>
</AccordionGroup>


# Add Single Row
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-sheets/add-single-row

The Add Single Row tool step allows you to add a row to your Google Sheet

## Add the ‘Add Single Row’ Tool step to your Tool

<div>
  <iframe title="Add the Add Single Row Tool step to your Tool" />
</div>

You can add the ‘Add Single Row’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Add Single Row’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Using the dropdown menu:
   * Select the Google account you would like to use.
   * Select the Drive you would like to use (defaults to "My Drive").
   * Select the Spreadsheet you would like to add the row to.
   * Select the worksheet you would like to add the row to.
4. Select whether the first row is a header row.
   <Note>If this option is checked, each header will appear as a separate input field in the tool step for you to fill in.</Note>
5. Click 'Run step' to test out the Tool

## Using in an Agent

There are 3 ways to configure the tool step when using in an agent

<Tabs>
  <Tab title="Using dropdown menus">
    This is the easiest way to configure the tool step. You can select the Drive, Spreadsheet and Worksheet from a list of options via the dropdown menu.

    <img alt="Google Sheets Add Single Row" />
  </Tab>

  <Tab title="Using input variables">
    This option offers the most flexibility - you can use the same tool step for multiple Google Sheets, however you will need to manually provide the necessary Google Sheet IDs.

    <img alt="Google Sheets Add Single Row" />
  </Tab>

  <Tab title="Mix of dropdown menus and input variables">
    This option offers the best of both worlds. You can use the dropdown menu to select the Drive, Spreadsheet and Worksheet, and then use an input variable for "Values" to allow the agent to dynamically insert data to your Google sheet.

    <img alt="Google Sheets Add Single Row" />
  </Tab>
</Tabs>

## Parameters

<ParamField type="string">
  The Drive you want to use. By default, the tool uses "My Drive".
</ParamField>

<ParamField type="string">
  To get the spreadsheet ID, navigate to your Google Sheet and copy the ID from the URL (the long string between `/d/` and `/edit`).
  <Tip> [You can use this tool to parse the worksheet, spreadsheet and drive IDs from a url](https://app.relevanceai.com/notebook/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/01e4918b-6c0a-423d-8daa-a625ced25e23)</Tip>
</ParamField>

<ParamField type="string">
  To get the worksheet ID, open the specific worksheet you want to use and copy the ID from the URL (appears after `#gid=`).
  <Tip> [You can also use this tool to parse the worksheet, spreadsheet and drive IDs from a url](https://app.relevanceai.com/notebook/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/01e4918b-6c0a-423d-8daa-a625ced25e23)</Tip>
</ParamField>

<ParamField type="boolean">
  When this option is checked, each column will appear as a separate input field in the tool step for you to fill in. **This option will be ignored when using dynamic input fields**.
</ParamField>

<ParamField type="array">
  The data being inserted will need to be in an array format.
  <Note>Example: `["John Doe", "john.doe@example.com", "123 Main St"]`</Note>
</ParamField>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What is the difference between using the green input variables and the dropdown menu?">
    **The dropdown menu is the easiest way to configure the tool step**. You can select the Drive, Spreadsheet and Worksheet from a list of options via the dropdown menu.

    **The input variables offer more flexibility** e.g. you can use the same tool step for multiple Google Sheets, however you will need to manually provide the necessary Google Sheet IDs.
  </Accordion>

  <Accordion title="Can I use a mix of both dropdown menu and green input variables?">
    If you are using the tool in an agent, this is the recommended approach. You can use the dropdown menu to select the fields that don't need to change, and then use the green input variables for "Range" to allow the agent to dynamically retrieve data from your Google sheet.

    <Warning>If you are only using the input variables, you will need to provide the necessary Google Sheet IDs for the tool step to function properly. At the minimum, this includes the spreadsheet ID and the worksheet ID.</Warning>
  </Accordion>
</AccordionGroup>


# Clear Cell
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-sheets/clear-cell

The Clear Cell tool step allows you to clear a cell in your Google Sheet

## Add the ‘Clear Cell’ Tool step to your Tool

<div>
  <iframe title="Google Sheets Clear Cell" />
</div>

You can add the ‘Clear Cell’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Clear Cell’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Using the dropdown menu:
   * Select the Google account you would like to use.
   * Select the Drive you would like to use (defaults to "My Drive").
   * Select the Spreadsheet you would like to clear the cell in.
   * Select the worksheet you would like to clear the cell in.
4. Enter the cell you want to clear (e.g. "A1").
5. Click 'Run step' to test out the Tool

## Using in an Agent

There are 3 ways to configure the tool step when using in an agent

<Tabs>
  <Tab title="Using dropdown menus">
    This is the easiest way to configure the tool step. You can select the Drive, Spreadsheet and Worksheet from a list of options via the dropdown menu.

    <img alt="Google Sheets Clear Cell" />
  </Tab>

  <Tab title="Using input variables">
    This option offers the most flexibility - you can use the same tool step for multiple Google Sheets, however you will need to manually provide the necessary Google Sheet IDs.

    <img alt="Google Sheets Clear Cell" />
  </Tab>

  <Tab title="Mix of dropdown menus and input variables">
    This option offers the best of both worlds. You can use the dropdown menu to select the Drive, Spreadsheet and Worksheet, and then use an input variable for "Cell" to give the agent the flexibility to clear different cells.

    <img alt="Google Sheets Clear Cell" />
  </Tab>
</Tabs>

## Parameters

<ParamField type="string">
  The Drive you want to use. By default, the tool uses "My Drive".
</ParamField>

<ParamField type="string">
  To get the spreadsheet ID, navigate to your Google Sheet and copy the ID from the URL (the long string between `/d/` and `/edit`).
  <Tip> [You can use this tool to parse the worksheet, spreadsheet and drive IDs from a url](https://app.relevanceai.com/notebook/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/01e4918b-6c0a-423d-8daa-a625ced25e23)</Tip>
</ParamField>

<ParamField type="string">
  To get the worksheet ID, open the specific worksheet you want to use and copy the ID from the URL (appears after `#gid=`).
  <Tip> [You can also use this tool to parse the worksheet, spreadsheet and drive IDs from a url](https://app.relevanceai.com/notebook/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/01e4918b-6c0a-423d-8daa-a625ced25e23)</Tip>
</ParamField>

<ParamField type="string">
  The cell you want to clear (e.g. "A1").
</ParamField>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What is the difference between using the green input variables and the dropdown menu?">
    **The dropdown menu is the easiest way to configure the tool step**. You can select the Drive, Spreadsheet and Worksheet from a list of options via the dropdown menu.

    **The input variables offer more flexibility** e.g. you can use the same tool step for multiple Google Sheets, however you will need to manually provide the necessary Google Sheet IDs.
  </Accordion>

  <Accordion title="Can I use a mix of both dropdown menu and green input variables?">
    If you are using the tool in an agent, this is the recommended approach. You can use the dropdown menu to select the Drive, Spreadsheet and Worksheet that don't need to change, and then use the green input variables for "Cell" to allow the agent to dynamically clear different cells in your Google sheet.
    <Warning>If you are only using the input variables, you will need to provide the necessary Google Sheet IDs for the tool step to function properly.</Warning>
  </Accordion>
</AccordionGroup>


# Clear Rows
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-sheets/clear-rows

The Clear Rows tool step allows you to clear rows in your Google Sheet

## Add the ‘Clear Rows’ Tool step to your Tool

<div>
  <iframe title="Google Sheets Clear Rows" />
</div>

You can add the ‘Clear Rows’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Clear Rows’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Using the dropdown menu:
   * Select the Google account you would like to use.
   * Select the Drive you would like to use (defaults to "My Drive").
   * Select the Spreadsheet you would like to clear the rows in.
   * Select the worksheet you would like to clear the rows in.
4. Enter the row number you want to start clearing from.
5. Enter the row number you want to stop clearing at.
6. Click 'Run step' to test out the Tool

## Using in an Agent

There are 3 ways to configure the tool step when using in an agent

<Tabs>
  <Tab title="Using dropdown menus">
    This is the easiest way to configure the tool step. You can select the Drive, Spreadsheet and Worksheet from a list of options via the dropdown menu.

    <img alt="Google Sheets Clear Rows" />
  </Tab>

  <Tab title="Using input variables">
    This option offers the most flexibility - you can use the same tool step for multiple Google Sheets, however you will need to manually provide the necessary Google Sheet IDs.

    <img alt="Google Sheets Clear Rows" />
  </Tab>

  <Tab title="Mix of dropdown menus and input variables">
    This option offers the best of both worlds. You can use the dropdown menu to select the Drive, Spreadsheet and Worksheet, and then use an input variable for "Start Index" and "End Index" to give the agent the flexibility to clear different rows.

    <img alt="Google Sheets Clear Rows" />
  </Tab>
</Tabs>

## Parameters

<ParamField type="string">
  The Drive you want to use. By default, the tool uses "My Drive".
</ParamField>

<ParamField type="string">
  To get the spreadsheet ID, navigate to your Google Sheet and copy the ID from the URL (the long string between `/d/` and `/edit`).
  <Tip> [You can use this tool to parse the worksheet, spreadsheet and drive IDs from a url](https://app.relevanceai.com/notebook/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/01e4918b-6c0a-423d-8daa-a625ced25e23)</Tip>
</ParamField>

<ParamField type="string">
  To get the worksheet ID, open the specific worksheet you want to use and copy the ID from the URL (appears after `#gid=`).
  <Tip> [You can also use this tool to parse the worksheet, spreadsheet and drive IDs from a url](https://app.relevanceai.com/notebook/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/01e4918b-6c0a-423d-8daa-a625ced25e23)</Tip>
</ParamField>

<ParamField type="number">
  The row number you want to start clearing from (inclusive).
</ParamField>

<ParamField type="number">
  The row number you want to stop clearing at (exclusive).
</ParamField>

<Tip>Example: If you want to clear rows 1 to 5, you would enter 1 for "Start Index" and 6 for "End Index".</Tip>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What is the difference between using the green input variables and the dropdown menu?">
    **The dropdown menu is the easiest way to configure the tool step**. You can select the Drive, Spreadsheet and Worksheet from a list of options via the dropdown menu.

    **The input variables offer more flexibility** e.g. you can use the same tool step for multiple Google Sheets, however you will need to manually provide the necessary Google Sheet IDs.
  </Accordion>

  <Accordion title="Can I use a mix of both dropdown menu and green input variables?">
    If you are using the tool in an agent, this is the recommended approach. You can use the dropdown menu to select the Drive, Spreadsheet and Worksheet that don't need to change, and then use the green input variables for "Start Index" and "End Index" to allow the agent to dynamically clear different rows in your Google sheet.
    <Warning>If you are only using the input variables, you will need to provide the necessary Google Sheet IDs for the tool step to function properly.</Warning>
  </Accordion>

  <Accordion title="What is the difference between the Google Sheets clear row and Delete rows tool step?">
    When you use [Delete Rows](/docs/build/tools/tool-steps/google-sheets/delete-rows), the remaining rows shift up to fill the gap. When you use Clear row, the rows are cleared in place (the data is removed, but the rows themselves remain the same).
  </Accordion>
</AccordionGroup>


# Copy Worksheet
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-sheets/copy-worksheet

The Copy Worksheet tool step allows you to copy a worksheet from one Google Sheet to another

## Add the ‘Copy Worksheet’ Tool step to your Tool

<div>
  <iframe title="Google Sheets Copy Worksheet" />
</div>

You can add the ‘Copy Worksheet’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Copy Worksheet’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Using the dropdown menu:
   * Select the account you would like to use.
   * Select the Drive you would like to use.
   * Select the Spreadsheet that the worksheet is in.
   * Select the worksheet you would like to copy.
   * Select the Spreadsheet you would like to copy the worksheet to.
4. Click 'Run step' to test out the Tool

## Parameters

<ParamField type="string">
  The Drive you want to use. By default, the tool uses "My Drive".
</ParamField>

<ParamField type="string">
  Enter the ID of the spreadsheet where the original worksheet is located.
</ParamField>

<ParamField type="string">
  Enter the ID of the worksheet you want to copy.
</ParamField>

<ParamField type="string">
  Enter the ID of the spreadsheet where you want to copy the worksheet to.
</ParamField>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can I copy a worksheet to the same spreadsheet?">
    Yes, you can copy a worksheet to the same spreadsheet. This behaviour would be the same as duplicating a worksheet.
  </Accordion>

  <Accordion title="Will the formatting be preserved?">
    Yes, all cell values, formulas, and formatting in the original worksheet will be copied to the new one.
  </Accordion>

  <Accordion title="Does this tool copy scripts, macros, or protected ranges?">
    No, this tool only copies the worksheet contents (data and formatting). Scripts, macros, and protected ranges will not be copied.
  </Accordion>
</AccordionGroup>


# Create Column
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-sheets/create-column

The Create Worksheet tool step allows you to create a new Google Sheet

## Add the ‘Create Column’ Tool step to your Tool

<div>
  <iframe title="Google Sheets Create Column" />
</div>

You can add the ‘Create Column’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Create Column’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Using the dropdown menu:
   * Select the account you would like to use.
   * Select the Drive you would like to use (defaults to "My Drive").
   * Select the Spreadsheet you would like to create the column in.
4. * Select the worksheet you would like to create the column in.
5. Enter the letter of the column where you would like to create the column.
6. Click 'Run step' to test out the Tool

## Parameters

<ParamField type="string">
  The Drive you want to use. By default, the tool uses "My Drive".
</ParamField>

<ParamField type="string">
  The ID of the spreadsheet you want to create the column in.
</ParamField>

<ParamField type="string">
  The ID of the worksheet you want to create the column in.
</ParamField>

<ParamField type="string">
  The location of the new column you want to create. The new column will be inserted to the right of the column letter you provide.
</ParamField>


# Create Spreadsheet
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-sheets/create-spreadsheet

The Create Spreadsheet tool step allows you to create a new Google Sheet

## Add the ‘Create Spreadsheet’ Tool step to your Tool

<div>
  <iframe title="Google Sheets Create Spreadsheet" />
</div>

You can add the ‘Create Spreadsheet’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Create Spreadsheet’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Select the Drive you would like to use (defaults to "My Drive").
4. Select the name of the new spreadsheet you would like to create.
5. Select the folder you would like to create the spreadsheet in.
6. **Optional:** select the template you would like to use to create the spreadsheet.
7. Click 'Run step' to test out the Tool

## Using in an Agent

There are 3 ways to configure the tool step when using in an agent

<Tabs>
  <Tab title="Using dropdown menus">
    This is the easiest way to configure the tool step. You can select the Drive, Folder and Spreadsheet from a list of options via the dropdown menu.

    <img alt="Google Sheets Create Spreadsheet" />
  </Tab>

  <Tab title="Using input variables">
    This option offers the most flexibility - you can use the same tool step for creating multiple spreadsheets, however you will need to manually provide the necessary Google Sheet IDs.

    <img alt="Google Sheets Create Spreadsheet" />
  </Tab>

  <Tab title="Mix of dropdown menus and input variables">
    This option offers the best of both worlds. You can use the dropdown menu to select the Drive, Folder and Spreadsheet, and then use an input variable for "Title" to allow the agent to dynamically pick a name for the new spreadsheet.

    <img alt="Google Sheets Create Spreadsheet" />
  </Tab>
</Tabs>

## Parameters

<ParamField type="string">
  The Drive you want to use. By default, the tool uses "My Drive".
</ParamField>

<ParamField type="string">
  The title of the new spreadsheet you want to create.
</ParamField>

<ParamField type="string">
  The folder you want to create the spreadsheet in. To get the folder ID, navigate to the folder and copy the ID from the URL (the long ID after `/folders/`).
</ParamField>

<ParamField type="string">
  Use one of your existing spreadsheets as a template. Leave this field blank to create a new spreadsheet from scratch.
  <Tip> [You can use this tool to parse the spreadsheet ID from a url](https://app.relevanceai.com/notebook/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/01e4918b-6c0a-423d-8daa-a625ced25e23)</Tip>
</ParamField>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What is the difference between using the green input variables and the dropdown menu?">
    **The dropdown menu is the easiest way to configure the tool step**. You can select the Drive, Folder and Spreadsheet from a list of options via the dropdown menu.

    **The input variables offer more flexibility** e.g. you can use the same tool step for creating multiple spreadsheets, however you will need to manually provide the necessary Google Sheet IDs.
  </Accordion>

  <Accordion title="Can I use a mix of both dropdown menu and green input variables?">
    If you are using the tool in an agent, this is the recommended approach. You can use the dropdown menu to select the fields that don't need to change, and then use the green input variables for "Title" to allow the agent to dynamically pick a name for the new sheet.
  </Accordion>
</AccordionGroup>


# Create Worksheet
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-sheets/create-worksheet

The Create Worksheet tool step allows you to create a new Google Sheet

## Add the ‘Create Worksheet’ Tool step to your Tool

<div>
  <iframe title="Google Sheets Create Worksheet" />
</div>

You can add the ‘Create Worksheet’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Create Worksheet’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Using the dropdown menu:
   * Select the account you would like to use.
   * Select the Drive you would like to use (defaults to "My Drive").
   * Select the Spreadsheet you would like to create the worksheet in.
4. Enter the name of the new worksheet you are creating.
5. Click 'Run step' to test out the Tool

## Using in an Agent

There are 3 ways to configure the tool step when using in an agent

<Tabs>
  <Tab title="Using dropdown menus">
    This is the easiest way to configure the tool step. You can select the Drive and Spreadsheet from a list of options via the dropdown menu.

    <img alt="Google Sheets Create Worksheet" />
  </Tab>

  <Tab title="Using input variables">
    This option offers the most flexibility - you can use the same tool step for creating multiple worksheets, however you will need to manually provide the necessary Google Sheet IDs.

    <img alt="Google Sheets Create Worksheet" />
  </Tab>

  <Tab title="Mix of dropdown menus and input variables">
    This option offers the best of both worlds. You can use the dropdown menu to select the Drive and Spreadsheet, and then use an input variable for "Title" to allow the agent to dynamically pick a name for the new worksheet.

    <img alt="Google Sheets Create Worksheet" />
  </Tab>
</Tabs>

## Parameters

<ParamField type="string">
  The Drive you want to use. By default, the tool uses "My Drive".
</ParamField>

<ParamField type="string">
  The ID of the spreadsheet you want to create the worksheet in.
  <Tip> [You can use this tool to parse the spreadsheet ID from a url](https://app.relevanceai.com/notebook/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/01e4918b-6c0a-423d-8daa-a625ced25e23)</Tip>
</ParamField>

<ParamField type="string">
  The title of the new worksheet you want to create.
</ParamField>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What is the difference between using the green input variables and the dropdown menu?">
    **The dropdown menu is the easiest way to configure the tool step**. You can select the Drive and Spreadsheet from a list of options via the dropdown menu.

    **The input variables offer more flexibility** e.g. you can use the same tool step for creating multiple worksheets, however you will need to manually provide the necessary Google Sheet IDs.
  </Accordion>

  <Accordion title="Can I use a mix of both dropdown menu and green input variables?">
    If you are using the tool in an agent, this is the recommended approach. You can use the dropdown menu to select the fields that don't need to change, then use the green input variables for "Title" to allow the agent to dynamically pick a name for the new sheet.
  </Accordion>
</AccordionGroup>


# Delete Rows
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-sheets/delete-rows

The Delete Rows tool step allows you to delete rows in your Google Sheet

## Add the ‘Delete Rows’ Tool step to your Tool

<div>
  <iframe title="Google Sheets Delete Rows" />
</div>

You can add the ‘Delete Rows’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Delete Rows’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Using the dropdown menu:
   * Select the Google account you would like to use.
   * Select the Drive you would like to use (defaults to "My Drive").
   * Select the Spreadsheet you would like to delete the rows in.
   * Select the worksheet you would like to delete the rows in.
4. Enter the row number you want to start deleting from.
5. Enter the row number you want to stop deleting at.
6. Click 'Run step' to test out the Tool

## Using in an Agent

There are 3 ways to configure the tool step when using in an agent

<Tabs>
  <Tab title="Using dropdown menus">
    This is the easiest way to configure the tool step. You can select the Drive, Spreadsheet and Worksheet from a list of options via the dropdown menu.

    <img alt="Google Sheets Delete Rows" />
  </Tab>

  <Tab title="Using input variables">
    This option offers the most flexibility - you can use the same tool step for multiple Google Sheets, however you will need to manually provide the necessary Google Sheet IDs.

    <img alt="Google Sheets Delete Rows" />
  </Tab>

  <Tab title="Mix of dropdown menus and input variables">
    This option offers the best of both worlds. You can use the dropdown menu to select the Drive, Spreadsheet and Worksheet, and then use an input variable for "Start Index" and "End Index" to allow the agent to delete different rows.

    <img alt="Google Sheets Delete Rows" />
  </Tab>
</Tabs>

## Parameters

<ParamField type="string">
  The Drive you want to use. By default, the tool uses "My Drive".
</ParamField>

<ParamField type="string">
  To get the spreadsheet ID, navigate to your Google Sheet and copy the ID from the URL (the long string between `/d/` and `/edit`).
  <Tip> [You can use this tool to parse the worksheet, spreadsheet and drive IDs from a url](https://app.relevanceai.com/notebook/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/01e4918b-6c0a-423d-8daa-a625ced25e23)</Tip>
</ParamField>

<ParamField type="string">
  To get the worksheet ID, open the specific worksheet you want to use and copy the ID from the URL (appears after `#gid=`).
  <Tip> [You can also use this tool to parse the worksheet, spreadsheet and drive IDs from a url](https://app.relevanceai.com/notebook/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/01e4918b-6c0a-423d-8daa-a625ced25e23)</Tip>
</ParamField>

<ParamField type="number">
  The row number you want to start deleting from (inclusive).
</ParamField>

<ParamField type="number">
  The row number you want to stop deleting at (exclusive).
</ParamField>

<Tip>Example: If you want to delete rows 1 to 5, you would enter 1 for "Start Row" and 6 for "End Row".</Tip>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What is the difference between using the green input variables and the dropdown menu?">
    **The dropdown menu is the easiest way to configure the tool step**. You can select the Drive, Spreadsheet and Worksheet from a list of options via the dropdown menu.

    **The input variables offer more flexibility** e.g. you can use the same tool step for multiple Google Sheets, however you will need to manually provide the necessary Google Sheet IDs.
  </Accordion>

  <Accordion title="Can I use a mix of both dropdown menu and green input variables?">
    If you are using the tool in an agent, this is the recommended approach. You can use the dropdown menu to select the Drive, Spreadsheet and Worksheet that don't need to change, and then use the green input variables for "Start Row" and "End Row" to allow the agent to dynamically delete different rows in your Google sheet.
    <Warning>If you are only using the input variables, you will need to provide the necessary Google Sheet IDs for the tool step to function properly.</Warning>
  </Accordion>
</AccordionGroup>


# Delete Worksheet
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-sheets/delete-worksheet

The Delete Worksheet tool step allows you to delete a worksheet in your Google Sheet

## Add the ‘Delete Worksheet’ Tool step to your Tool

<div>
  <iframe title="Google Sheets Delete Worksheet" />
</div>

You can add the ‘Delete Worksheet’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Delete Worksheet’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Using the dropdown menu:
   * Select the Google account you would like to use.
   * Select the Drive you would like to use (defaults to "My Drive").
   * Select your spreadsheet.
   * Select the worksheet you would like to delete.
4. Click 'Run step' to test out the Tool

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="How do i get the Drive ID?">
    To get the Drive ID, open Google Drive in your browser and navigate to the folder or shared drive you want to use. The drive ID is the long string of letters and numbers in the URL after `/drives/`.
  </Accordion>

  <Accordion title="How do i get the Spreadsheet ID?">
    You can get the Spreadsheet ID by navigating to your Google Sheet and copying the ID from the URL (the long string between `/d/` and `/edit`).
  </Accordion>

  <Accordion title="How do i get the Worksheet ID?">
    You can get the Worksheet ID by opening the specific worksheet you want to use and copying the ID from the URL (appears after `#gid=`).
  </Accordion>
</AccordionGroup>


# Find Row
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-sheets/find-row

The Find Row tool step allows you to find a row in your Google Sheet

## Add the ‘Find Row’ Tool step to your Tool

<div>
  <iframe title="Google Sheets Find Row" />
</div>

You can add the ‘Find Row’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Find Row’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Using the dropdown menu:
   * Select the Google account you would like to use.
   * Select the Drive you would like to use (defaults to "My Drive").
   * Select the Spreadsheet you would like to find the row in.
   * Select the worksheet you would like to find the row in.
4. Enter the letter of the column you want to search in.
5. Enter the value you want to find.
   <Note>The value you enter will be case-sensitive.</Note>
6. Click 'Run step' to test out the Tool

## Using in an Agent

There are 3 ways to configure the tool step when using in an agent

<Tabs>
  <Tab title="Using dropdown menus">
    This is the easiest way to configure the tool step. You can select the Drive, Spreadsheet and Worksheet from a list of options via the dropdown menu.

    <img alt="Google Sheets Find Row" />
  </Tab>

  <Tab title="Using input variables">
    This option offers the most flexibility - you can use the same tool step for multiple Google Sheets, however you will need to manually provide the necessary Google Sheet IDs.

    <img alt="Google Sheets Find Row" />
  </Tab>

  <Tab title="Mix of dropdown menus and input variables">
    This option offers the best of both worlds. You can use the dropdown menu to select the Drive, Spreadsheet and Worksheet, and then use an input variable for "Column Letter" and "Value" to allow the agent to work for different values.

    <img alt="Google Sheets Find Row" />
  </Tab>
</Tabs>

## Parameters

<ParamField type="string">
  The Drive you want to use. By default, the tool uses "My Drive".
</ParamField>

<ParamField type="string">
  To get the spreadsheet ID, navigate to your Google Sheet and copy the ID from the URL (the long string between `/d/` and `/edit`).
  <Tip> [You can use this tool to parse the worksheet, spreadsheet and drive IDs from a url](https://app.relevanceai.com/notebook/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/01e4918b-6c0a-423d-8daa-a625ced25e23)</Tip>
</ParamField>

<ParamField type="string">
  To get the worksheet ID, open the specific worksheet you want to use and copy the ID from the URL (appears after `#gid=`).
  <Tip> [You can also use this tool to parse the worksheet, spreadsheet and drive IDs from a url](https://app.relevanceai.com/notebook/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/01e4918b-6c0a-423d-8daa-a625ced25e23)</Tip>
</ParamField>

<ParamField type="string">
  The letter of the column of the sheet you are searching in.
</ParamField>

<ParamField type="string">
  The value you want to search for in the column.
</ParamField>

<ParamField type="boolean">
  When this option is checked, the tool will return the entire row in the results.
</ParamField>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What is the difference between using the green input variables and the dropdown menu?">
    **The dropdown menu is the easiest way to configure the tool step**. You can select the Drive, Spreadsheet and Worksheet from a list of options via the dropdown menu.

    **The input variables offer more flexibility** e.g. you can use the same tool step for multiple Google Sheets, however you will need to manually provide the necessary Google Sheet IDs.
  </Accordion>

  <Accordion title="Can I use a mix of both dropdown menu and green input variables?">
    If you are using the tool in an agent, this is the recommended approach. You can use the dropdown menu to select the Drive, Spreadsheet and Worksheet that don't need to change, and then use the green input variables for "Column Letter" and "Value" to allow the agent to dynamically search for different values in your Google sheet.
    <Warning>If you are only using the input variables, you will need to provide the necessary Google Sheet IDs for the tool step to function properly.</Warning>
  </Accordion>
</AccordionGroup>


# Get Cell
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-sheets/get-cell

The Get Cell tool step allows you to retrieve a cell from your Google Sheet

## Add the ‘Get Cell’ Tool step to your Tool

<div>
  <iframe title="Google Sheets Get Cell" />
</div>

You can add the ‘Get Cell’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Get Cell’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Using the dropdown menu:
   * Select the Google account you would like to use.
   * Select the Drive you would like to use (defaults to "My Drive").
   * Select the Spreadsheet you would like to fetch from.
   * Select the worksheet you would like to fetch from.
   * Select the Cell you would like to fetch.
4. Click 'Run step' to test out the Tool

## Parameters

<ParamField type="string">
  The Drive you want to use. By default, the tool uses "My Drive".
</ParamField>

<ParamField type="string">
  The ID of the spreadsheet you want to fetch from.
</ParamField>

<ParamField type="string">
  The ID of the worksheet you want to fetch from.
</ParamField>

<ParamField type="string">
  The coordinates of the cell you want to fetch. For example, "A1", "B2", etc.
</ParamField>


# Get Spreadsheet by ID
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-sheets/get-spreadsheet-by-id

The Get Spreadsheet by ID tool step allows you to retrieve a spreadsheet from your Google Drive

## Add the ‘Get Spreadsheet by ID’ Tool step to your Tool

<div>
  <iframe title="Google Sheets Get Spreadsheet by ID" />
</div>

You can add the ‘Get Spreadsheet by ID’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Get Spreadsheet by ID’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Using the dropdown menu:
   * Select the Google account you would like to use.
   * Select the Drive you would like to use (defaults to "My Drive").
   * Select the Spreadsheet you would like to fetch.
4. Click 'Run step' to test out the Tool

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="How do I get the Drive ID?">
    To get the Drive ID, navigate to your Google Drive and copy the ID from the URL (the long string after `/folders/`).
    <Tip> [You can also use this tool to parse the drive ID from a url](https://app.relevanceai.com/notebook/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/01e4918b-6c0a-423d-8daa-a625ced25e23)</Tip>
  </Accordion>

  <Accordion title="How do I get the spreadsheet ID?">
    To get the spreadsheet ID, navigate to your Google Sheet and copy the ID from the URL (the long string between `/d/` and `/edit`).
    <Tip> [You can also use this tool to parse the spreadsheet ID from a url](https://app.relevanceai.com/notebook/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/01e4918b-6c0a-423d-8daa-a625ced25e23)</Tip>
  </Accordion>
</AccordionGroup>


# Get Values in Range
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-sheets/get-values-in-range

The Get Values in Range tool step allows you to retrieve data from your Google Sheet

## Add the ‘Get Values in Range’ Tool step to your Tool

<div>
  <iframe title="Google Sheets Get Values in Range" />
</div>

You can add the ‘Get Values in Range’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Get Values in Range’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Using the dropdown menu:
   * Select the Google account you would like to use.
   * Select the Drive you would like to use (defaults to "My Drive").
   * Select the Spreadsheet you would like to get the values from.
   * Select the worksheet you would like to get the values from.
4. Select the range of cells you would like to get the values from (e.g. `A1:B2`).
5. Click 'Run step' to test out the Tool

## Using in an Agent

There are 3 ways to configure the tool step when using in an agent

<Tabs>
  <Tab title="Using dropdown menus">
    This is the easiest way to configure the tool step. You can select the Drive, Spreadsheet and Worksheet from a list of options via the dropdown menu.

    <img alt="Google Sheets Get Values in Range" />
  </Tab>

  <Tab title="Using input variables">
    This option offers the most flexibility - you can use the same tool step for multiple Google Sheets, however you will need to manually provide the necessary Google Sheet IDs.

    <img alt="Google Sheets Get Values in Range" />
  </Tab>

  <Tab title="Mix of dropdown menus and input variables">
    This option offers the best of both worlds. You can use the dropdown menu to select the Drive, Spreadsheet and Worksheet, and then use an input variable for "Range" to allow the agent to dynamically retrieve data to your Google sheet.

    <img alt="Google Sheets Get Values in Range" />
  </Tab>
</Tabs>

## Parameters

<ParamField type="string">
  The Drive you want to use. By default, the tool uses "My Drive".
</ParamField>

<ParamField type="string">
  To get the spreadsheet ID, navigate to your Google Sheet and copy the ID from the URL (the long string between `/d/` and `/edit`).
  <Tip> [You can use this tool to parse the worksheet, spreadsheet and drive IDs from a url](https://app.relevanceai.com/notebook/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/01e4918b-6c0a-423d-8daa-a625ced25e23)</Tip>
</ParamField>

<ParamField type="string">
  To get the worksheet ID, open the specific worksheet you want to use and copy the ID from the URL (appears after `#gid=`).
  <Tip> [You can also use this tool to parse the worksheet, spreadsheet and drive IDs from a url](https://app.relevanceai.com/notebook/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/01e4918b-6c0a-423d-8daa-a625ced25e23)</Tip>
</ParamField>

<ParamField type="string">
  Specify the range of cells to retrieve, starting from the top-left cell to the bottom-right cell.

  <img alt="Google Sheets Add Single Row" />

  `A2:B3` will return a 2x2 block containing `Bella`, `Luna`, `Dog`, and `Cat`.
  <Note>Leave the "Range" field empty to retrieve the entire sheet.</Note>
</ParamField>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What is the difference between using the green input variables and the dropdown menu?">
    **The dropdown menu is the easiest way to configure the tool step**. You can select the Drive, Spreadsheet and Worksheet from a list of options via the dropdown menu.

    **The input variables offer more flexibility** e.g. you can use the same tool step for multiple Google Sheets, however you will need to manually provide the necessary Google Sheet IDs.
  </Accordion>

  <Accordion title="Can I use a mix of both dropdown menu and green input variables?">
    If you are using the tool in an agent, this is the recommended approach. You can use the dropdown menu to select the fields that don't need to change, and then use the green input variables for "Range" to allow the agent to dynamically retrieve data from your Google sheet.

    <Warning>If you are only using the input variables, you will need to provide the necessary Google Sheet IDs for the tool step to function properly. At the minimum, this includes the spreadsheet ID and the worksheet ID.</Warning>
  </Accordion>
</AccordionGroup>


# List Worksheets
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-sheets/list-worksheets

The List Worksheets tool step allows you to list all the worksheets in a Google Sheet

## Add the ‘List Worksheets’ Tool step to your Tool

<div>
  <iframe title="Google Sheets List Worksheets" />
</div>

You can add the ‘List Worksheets’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘List Worksheets’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Using the dropdown menu:
   * **Select the Google account** you would like to use.
   * **Select the Drive** you would like to use (defaults to "My Drive").
   * **Select the Spreadsheet** you would like to get the worksheets from.
4. Click 'Run step' to test out the Tool

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="How do I use this tool in an agent?">
    For an example of how to use this tool in an agent, see [Add single row](../google-sheets/add-single-row).
  </Accordion>

  <Accordion title="How do i get the spreadsheet ID?">
    To get the spreadsheet ID, navigate to your Google Sheet and copy the ID from the URL (the long string between `/d/` and `/edit`).
    <Tip> [You can use this tool to parse the spreadsheet ID from a url](https://app.relevanceai.com/notebook/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/01e4918b-6c0a-423d-8daa-a625ced25e23)</Tip>
  </Accordion>
</AccordionGroup>


# Update Cell
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-sheets/update-cell

The Update Cell tool step allows you to update any cell in your Google Sheet

## Add the ‘Update Cell’ Tool step to your Tool

<div>
  <iframe title="Google Sheets Update Cell" />
</div>

You can add the ‘Update Cell’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Update Cell’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Using the dropdown menu:
   * Select the Google account you would like to use.
   * Select the Drive you would like to use (defaults to "My Drive").
   * Select the Spreadsheet you would like to update.
   * Select the worksheet you would like to update.
4. Enter the coordinates of the cell you would like to update.
5. Input the new cell value.
6. Click 'Run step' to test out the Tool

## Using in an Agent

There are 3 ways to configure the tool step when using in an agent

<Tabs>
  <Tab title="Using dropdown menus">
    This is the easiest way to configure the tool step. You can select the Drive, Spreadsheet and Worksheet from a list of options via the dropdown menu.

    <img alt="Google Sheets Update Cell" />
  </Tab>

  <Tab title="Using input variables">
    This option offers the most flexibility - you can use the same tool step for multiple Google Sheets, however you will need to manually provide the necessary Google Sheet IDs.

    <img alt="Google Sheets Update Cell" />
  </Tab>

  <Tab title="Mix of dropdown menus and input variables">
    This option offers the best of both worlds. You can use the dropdown menu to select the Drive, Spreadsheet and Worksheet, and then use an input variable for "Cell" and "Value" to allow the agent to dynamically update different cells in your Google sheet.

    <img alt="Google Sheets Update Cell" />
  </Tab>
</Tabs>

## Parameters

<ParamField type="string">
  The Drive you want to use. By default, the tool uses "My Drive".
</ParamField>

<ParamField type="string">
  To get the spreadsheet ID, navigate to your Google Sheet and copy the ID from the URL (the long string between `/d/` and `/edit`).
  <Tip> [You can use this tool to parse the worksheet, spreadsheet and drive IDs from a url](https://app.relevanceai.com/notebook/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/01e4918b-6c0a-423d-8daa-a625ced25e23)</Tip>
</ParamField>

<ParamField type="string">
  To get the worksheet ID, open the specific worksheet you want to use and copy the ID from the URL (appears after `#gid=`).
  <Tip> [You can also use this tool to parse the worksheet, spreadsheet and drive IDs from a url](https://app.relevanceai.com/notebook/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/01e4918b-6c0a-423d-8daa-a625ced25e23)</Tip>
</ParamField>

<ParamField type="string">
  The notation of the cell you want to update. For example, "A1", "B2", etc.
</ParamField>

<ParamField type="string">
  The new value you want to update the cell with.
</ParamField>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What is the difference between using the green input variables and the dropdown menu?">
    **The dropdown menu is the easiest way to configure the tool step**. You can select the Drive, Spreadsheet and Worksheet from a list of options via the dropdown menu.

    **The input variables offer more flexibility** e.g. you can use the same tool step for multiple Google Sheets, however you will need to manually provide the necessary Google Sheet IDs.
  </Accordion>

  <Accordion title="Can I use a mix of both dropdown menu and green input variables?">
    If you are using the tool in an agent, this is the recommended approach. You can use the dropdown menu to select the fields that don't need to change, and then use the green input variables for "Cell" and "Cell Value" to allow the agent to dynamically retrieve data from your Google sheet.

    <Warning>If you are only using the input variables, you will need to provide the necessary Google Sheet IDs for the tool step to function properly. At the minimum, this includes the spreadsheet ID and the worksheet ID.</Warning>
  </Accordion>
</AccordionGroup>


# Update Multiple Rows
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-sheets/update-multiple-rows

The Update Multiple Rows tool step allows you to update multiple rows in your Google Sheet

## Add the ‘Update Multiple Rows’ Tool step to your Tool

<div>
  <iframe title="Google Sheets Update Multiple Rows" />
</div>

You can add the ‘Update Multiple Rows’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Update Multiple Rows’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Using the dropdown menu:
   * Select the Google account you would like to use.
   * Select the Drive you would like to use (defaults to "My Drive").
   * Select the Spreadsheet you would like to update rows in.
   * Select the worksheet you would like to update rows in.
4. Enter the range you would like to update. This should be in the format of `A1:C5`.
5. Enter the new Row Values. This should be an array of arrays.
   <Note>Example: `[["John Doe", "john.doe@example.com", "123 Main St"], ["Jane Doe", "jane.doe@example.com", "456 Main St"]]`</Note>
6. Click 'Run step' to test out the Tool

## Using in an Agent

There are 3 ways to configure the tool step when using in an agent

<Tabs>
  <Tab title="Using dropdown menus">
    This is the easiest way to configure the tool step. You can select the Drive, Spreadsheet and Worksheet from a list of options via the dropdown menu.

    <img alt="Google Sheets Update multiple rows" />
  </Tab>

  <Tab title="Using input variables">
    This option offers the most flexibility - you can use the same tool step for multiple Google Sheets, however you will need to manually provide the necessary Google Sheet IDs.

    <img alt="Google Sheets Update multiple rows" />
  </Tab>

  <Tab title="Mix of dropdown menus and input variables">
    This option offers the best of both worlds. You can use the dropdown menu to select the Drive, Spreadsheet and Worksheet, and then use an input variable for "Range" and "Row Values" to allow the agent to dynamically update data in your Google sheet.

    <img alt="Google Sheets Update multiple rows" />
  </Tab>
</Tabs>

## Parameters

<ParamField type="string">
  The Drive you want to use. By default, the tool uses "My Drive".
</ParamField>

<ParamField type="string">
  To get the spreadsheet ID, navigate to your Google Sheet and copy the ID from the URL (the long string between `/d/` and `/edit`).
  <Tip> [You can use this tool to parse the worksheet, spreadsheet and drive IDs from a url](https://app.relevanceai.com/notebook/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/01e4918b-6c0a-423d-8daa-a625ced25e23)</Tip>
</ParamField>

<ParamField type="string">
  To get the worksheet ID, open the specific worksheet you want to use and copy the ID from the URL (appears after `#gid=`).
  <Tip> [You can also use this tool to parse the worksheet, spreadsheet and drive IDs from a url](https://app.relevanceai.com/notebook/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/01e4918b-6c0a-423d-8daa-a625ced25e23)</Tip>
</ParamField>

<ParamField type="string">
  The range of cells to update. The update will begin at the starting cell of the range (e.g., if the range is `A10:C20`, updates begin at `A10`).

  <Warning>
    Make sure the specified range fully covers the rows and columns you are updating.\
    For example, if you are updating 3 columns across the first 5 rows, the range should be at least `A1:C5`.
  </Warning>
</ParamField>

<ParamField type="array of arrays">
  The data being updated will need to be in an array of arrays format.

  **Example:**

  ```json theme={null}
  [
    ["John Doe", "john.doe@example.com", "123 Main St"], // row 1
    ["Jane Doe", "jane.doe@example.com", "456 Main St"] // row 2
    // ...
  ]
  ```
</ParamField>

## What is an array of arrays?

An array is a data structure similar to a CSV row, where each element is separated by a comma, and the entire row is enclosed in square brackets.
An array can have many different types of data, including strings, numbers, <Tooltip>booleans</Tooltip>, and even other arrays.

```json theme={null}
["Dog", 1, true]
```

An array of arrays is just multiple arrays put together. This is what an array of arrays looks like:

```json theme={null}
[
  ["Column 1 Row 1", "Column 2 Row 1", "Column 3 Row 1"],
  ["Column 1 Row 2", "Column 2 Row 2", "Column 3 Row 2"],
  ["Column 1 Row 3", "Column 2 Row 3", "Column 3 Row 3"]
]
```

When you combine multiple arrays together, you can see that the data is structured like a Google Sheet.

### Playground

<div>
  <iframe />
</div>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What is the difference between using the green input variables and the dropdown menu?">
    **The dropdown menu is the easiest way to configure the tool step**. You can select the Drive, Spreadsheet and Worksheet from a list of options via the dropdown menu.

    **The input variables offer more flexibility** e.g. you can use the same tool step for multiple Google Sheets, however you will need to manually provide the necessary Google Sheet IDs.
  </Accordion>

  <Accordion title="Can I use a mix of both dropdown menu and green input variables?">
    If you are using the tool in an agent, this is the recommended approach. You can use the dropdown menu to select the fields that don't need to change (the ID fields), and then use the green input variables for "Range" and "Row Values" to allow the agent to dynamically update data in your Google sheet.

    <Warning>If you are only using the input variables, you will need to provide the necessary Google Sheet IDs for the tool step to function properly. At the minimum, this includes the spreadsheet ID and the worksheet ID.</Warning>
  </Accordion>
</AccordionGroup>


# Update Row
Source: https://relevanceai.com/docs/build/tools/tool-steps/google-sheets/update-row

The Update Row tool step allows you to update any row in your Google Sheet

## Add the ‘Update Row’ Tool step to your Tool

<div>
  <iframe title="Google Sheets Update Row" />
</div>

You can add the ‘Update Row’ Tool step to your Tool by:

1. Creating a new Tool, then searching for the ‘Update Row’ Tool step
2. Click ‘Expand’ to see the full Tool step
3. Using the dropdown menu:
   * Select the Google account you would like to use.
   * Select the Drive you would like to use (defaults to "My Drive").
   * Select the Spreadsheet you would like to update.
   * Select the worksheet you would like to update.
4. Select the row number you would like to update.
5. Select whether the first row has headers.
6. Enter the values you would like to update.
7. Click 'Run step' to test out the Tool

## Using in an Agent

There are 3 ways to configure the tool step when using in an agent

<Tabs>
  <Tab title="Using dropdown menus">
    This is the easiest way to configure the tool step. You can select the Drive, Spreadsheet and Worksheet from a list of options via the dropdown menu.

    <img alt="Google Sheets Update Row" />
  </Tab>

  <Tab title="Using input variables">
    This option offers the most flexibility - you can use the same tool step for multiple Google Sheets, however you will need to manually provide the necessary Google Sheet IDs.

    <img alt="Google Sheets Update Row" />
  </Tab>

  <Tab title="Mix of dropdown menus and input variables">
    This option offers the best of both worlds. You can use the dropdown menu to select the Drive, Spreadsheet and Worksheet, and then use an input variable for "Row Number" and "Values" to allow the agent to dynamically update different rows in your Google sheet.

    <img alt="Google Sheets Update Row" />
  </Tab>
</Tabs>

## Parameters

<ParamField type="string">
  The Drive you want to use. By default, the tool uses "My Drive".
</ParamField>

<ParamField type="string">
  To get the spreadsheet ID, navigate to your Google Sheet and copy the ID from the URL (the long string between `/d/` and `/edit`).
  <Tip> [You can use this tool to parse the worksheet, spreadsheet and drive IDs from a url](https://app.relevanceai.com/notebook/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/01e4918b-6c0a-423d-8daa-a625ced25e23)</Tip>
</ParamField>

<ParamField type="string">
  To get the worksheet ID, open the specific worksheet you want to use and copy the ID from the URL (appears after `#gid=`).
  <Tip> [You can also use this tool to parse the worksheet, spreadsheet and drive IDs from a url](https://app.relevanceai.com/notebook/f1db6c/d64db999bc25-4ea0-b926-30b350f69243/01e4918b-6c0a-423d-8daa-a625ced25e23)</Tip>
</ParamField>

<ParamField type="number">
  The row number you want to update.
</ParamField>

<ParamField type="boolean">
  When this option is checked, each column will appear as a separate input field in the tool step for you to fill in. **This option will be ignored when using dynamic input fields**.
</ParamField>

<ParamField type="array">
  The data being inserted will need to be in an array format.
  <Note>Example: `["John Doe", "john.doe@example.com", "123 Main St"]`</Note>
</ParamField>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What is the difference between using the green input variables and the dropdown menu?">
    **The dropdown menu is the easiest way to configure the tool step**. You can select the Drive, Spreadsheet and Worksheet from a list of options via the dropdown menu.

    **The input variables offer more flexibility** e.g. you can use the same tool step for multiple Google Sheets, however you will need to manually provide the necessary Google Sheet IDs.
  </Accordion>

  <Accordion title="Can I use a mix of both dropdown menu and green input variables?">
    If you are using the tool in an agent, this is the recommended approach. You can use the dropdown menu to select the fields that don't need to change, and then use the green input variables for "Range" to allow the agent to dynamically retrieve data from your Google sheet.

    <Warning>If you are only using the input variables, you will need to provide the necessary Google Sheet IDs for the tool step to function properly. At the minimum, this includes the spreadsheet ID and the worksheet ID.</Warning>
  </Accordion>
</AccordionGroup>


# Advanced Knowledge Search (2m context)
Source: https://relevanceai.com/docs/build/tools/tool-steps/knowledge/advanced-knowledge-search-2m

The 'Advanced Knowledge Search (2m context)' Tool allows you to search your knowledge table with a prompt using CAG

## Add the 'Advanced Knowledge Search (2m)' Tool step to your Tool

<div>
  <iframe title="Add the Advanced Knowledge Search (2m) Tool step to your Tool" />
</div>

You can add the 'Advanced Knowledge Search (2m context) Tool' step to your Tool by:

1. Creating a new Tool, then searching for the Advanced Knowledge Search (2m context) Tool step.
2. Click "Expand" to see the full Tool step.
3. Enter your search query in the **Query** field.
4. Select the knowledge source you would like to use.
5. Select your Gemini model *(currently there are only 2 models available)*

## Advanced settings

### System prompt

The system prompt is the very first message in a conversation that instructs the model on its role or the personality of the AI. Examples can be expert researcher, analyst or tutor.

### Temperature

The temperature setting controls how “creative” or “random” the model’s responses will be. Lower temperatures make the model more focused and deterministic, while higher temperatures make it more creative and unpredictable.

### Enable Grounding

Grounding integrates Google Search results to inform and support the model’s responses.

### Page size

Page size controls the number of documents that will be retrieved from the knowledge set.


# Delete Knowledge
Source: https://relevanceai.com/docs/build/tools/tool-steps/knowledge/delete-knowledge

The 'Delete Knowledge' Tool step allows your agent to remove information from your knowledge table — making it easy to manage and keep your data relevant and up to date.

***

## Add the 'Delete Knowledge' Tool step to your Tool

<div>
  <iframe title="Add the Delete Knowledge Tool step to your Tool" />
</div>

You can add the 'Delete Knowledge' Tool step to your Tool by:

1. Creating a new Tool, then clicking `+ Add Step`
2. Searching for and adding the 'Delete Knowledge' Tool step
3. Clicking ‘Expand’ to see the full Tool step
4. Selecting the knowledge table you would like to access
5. Filter condition: “and” vs “or” when matching data to delete:

   * Use `and` when all specified conditions must match for a row to be deleted.
   * Use `or` when any of the conditions can match.

   <Note>
     Filter condition only applies if you include more than one object in the filter
   </Note>
6. Filters: Accepts a JSON selection of objects, where each object corresponds to a row. Add more objects as outlined in the formatting examples below to delete multiple rows at once.
7. Click `Run step` to test out the Tool

## Common errors

<Accordion title="Formatting issues">
  This indicates that there is a formatting issue in the "Filters" input field. Make sure your object is formatted correctly. `Expected property name or '}' in JSON at...`
</Accordion>

<Accordion title="Incorrect data type">
  This means the Filter being used is not a JSON. `Invalid JSON`
</Accordion>

## Formatting Examples

### Example 1: Delete single condition

```json theme={null}
// All poodles will be deleted
[
  {
    "breed": "poodle"
  }
]
```

### Example 2: Delete single condition

```json theme={null}
// All small poodles will be deleted
[
  {
    "breed": "poodle",
	"size": "small"
  }
]
```

### Example 3: Delete multiple conditions (*or*)

```json theme={null}
// All small poodles will be deleted AND any dog aged 4 will be deleted
[
  {
    "breed": "poodle",
    "size": "small"
  }, 
  {
    "age": 4
  }
]
```

### Example 3: Delete multiple conditions (*and*)

```json theme={null}
// Only small poodles that are aged 4 will be deleted
[
  {
    "breed": "poodle",
    "size": "small"
  }, 
  {
    "age": 4
  }
]
```

<Note>
  Ensure to separate each object with a comma if inserting multiple rows.
</Note>

## Using Variables in JSON

You can use variables from Tool inputs and other Tool steps in the **Filters** field. For detailed information on how to use variables in JSON for Knowledge Tool steps, including variable syntax and typing, see [Using Variables in JSON](/docs/build/tools/tool-steps/knowledge/using-variables-in-json).

For Knowledge Tool steps, all variables should be wrapped in quotes in the JSON format: `"{{variable_name}}"`. The filter values will be matched against your knowledge table data based on the column types.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What's the difference between using 'AND' and 'OR'?">
    * Use **AND** when *all* filter conditions must be true for a row to be deleted.
    * Use **OR** when *any* one of the conditions can match.

    <Note>
      This only applies when you include more than one object in the filter array.
    </Note>
  </Accordion>

  <Accordion title="Can I delete multiple rows at once?">
    Yes. Any row that matches the filter condition you have specified will be deleted.
  </Accordion>

  <Accordion title="What if my filters don't match any rows?">
    If no rows match the provided conditions, no data will be deleted. Make sure your filter values exactly match the data in the knowledge set (e.g., casing, field names, etc.).
  </Accordion>
</AccordionGroup>


# Get Knowledge
Source: https://relevanceai.com/docs/build/tools/tool-steps/knowledge/get-knowledge

The Get Knowledge feature allows you to search and retrieve information directly from your knowledge table without RAG

## Add the ‘Get Knowledge’ Tool step to your Tool

<div>
  <iframe title="Add the Get Knowledge Tool step to your Tool" />
</div>

You can add the Get Knowledge’ Tool step to your Tool by:

1. Creating a new Tool, then clicking `+ Add Step`
2. Searching for and addingthe ‘Get Knowledge’ Tool step
3. Clicking ‘Expand’ to see the full Tool step
4. Selecting the Knowledge table you would like to access
5. Filter condition: “and” vs “or” when matching data to update:
   * Use **and** when all specified conditions must match for a row to be deleted.
   * Use **or** when any of the conditions can match.
   <Note>
     Filter condition only applies if you include more than one object in the filter
   </Note>
6. Filters: Accepts a JSON selection of objects, where each object corresponds to a row. Add more objects as outlined in the formatting examples below to delete multiple rows at once.
7. Click `Run step` to test out the Tool

## Advanced settings

### Fields to include

This is the fields (columns) that will be included in the response. By default, all fields are included.

<Tip> When targeting a specific field, you will need to prefix each field with `data.`. For example, if you have a field name `breed`, you will instead use `data.breed`. </Tip>

### Number of records to retrieve

This is the number of records to retrieve. By default, **20** records are retrieved.

## Common errors

<AccordionGroup>
  <Accordion title="Formatting issues">
    This indicates that there is a formatting issue in the "New data to insert" input field. Make sure your object is formatted correctly. `Expected property name or '}' in JSON at...`
  </Accordion>

  <Accordion title="Incorrect data type">
    This means the data being inserted into the knowledge table is not a JSON. `Invalid JSON`
  </Accordion>
</AccordionGroup>

## Formatting Examples

### Example 1: Retrieve all large dogs

```json theme={null}
[
  {
    "size": "large"
  }
]
```

### Example 2: Retrieve all large French Bulldogs

```json theme={null}
[
  {
    "breed": "French Bulldog",
    "size": "large"
  },
]
```

### Example 3: Retrieve all large dogs OR small poodles

```json theme={null}
[
  {
    "size": "large",
    "breed": "French Bulldog"
  },
  {
    "size": "small",
    "breed": "poodle"
  }
]
```

<Note>
  Ensure to separate each object with a comma if inserting multiple rows.
</Note>

## Using Variables in JSON

You can use variables from Tool inputs and other Tool steps in the **Filters** field. For detailed information on how to use variables in JSON for Knowledge Tool steps, including variable syntax and typing, see [Using Variables in JSON](/docs/build/tools/tool-steps/knowledge/using-variables-in-json).

For Knowledge Tool steps, all variables should be wrapped in quotes in the JSON format: `"{{variable_name}}"`. The filter values will be matched against your knowledge table data based on the column types.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What is the difference between get knowledge and advanced knowledge search?">
    Get knowledge allows you to search and retrieve information directly from your knowledge table without RAG.
  </Accordion>

  <Accordion title="What format should I use for the filter?">
    The filter should be a JSON array of objects. Each object should have the field name and the value you want to filter by.
  </Accordion>

  <Accordion title="What's the difference between using 'AND' and 'OR'?">
    * Use **AND** when *all* filter conditions must be true for a row to be retrieved.
    * Use **OR** when *any* one of the conditions can match.

    <Note>
      This only applies when you include more than one object in the filter array.
    </Note>
  </Accordion>
</AccordionGroup>


# Insert Knowledge
Source: https://relevanceai.com/docs/build/tools/tool-steps/knowledge/insert-knowledge

The 'Insert Knowledge' Tool step lets your agent save information by adding new rows to your knowledge table—making it easy to capture and store data dynamically.

## Add the 'Insert Knowledge' Tool step to your Tool

<div>
  <iframe title="Add the Insert Knowledge Tool step to your Tool" />
</div>

You can add the 'Insert Knowledge' Tool step to your Tool by:

1. Creating a new Tool, then clicking `+ Add Step`
2. Searching for and adding the 'Insert Knowledge' Tool step
3. Clicking 'Expand' to see the full Tool step
4. Selecting the Knowledge table you would like to access
5. New data to insert: Accepts a JSON selection of objects, where each object corresponds to a row. Add more objects as outlined in the formatting examples below to insert multiple rows at once.
6. Use the **Sync on upload** field to determine whether the data should be vectorized when added to the knowledge set. In most cases, you'll want this enabled as it allows your agents to semantically search and retrieve relevant information from your knowledge table.
7. Click `Run step` to test out the Tool

## Common errors

<Accordion title="Formatting issues">
  This indicates that there is a formatting issue in the "New data to insert" input field. Make sure your object is formatted correctly. `Expected property name or '}' in JSON at...`
</Accordion>

<Accordion title="Incorrect data type">
  This means the data being inserted into the knowledge table is not a JSON. `Invalid JSON`
</Accordion>

<Accordion title="Tool input type mismatch">
  When creating a Tool that wraps the Insert Knowledge Tool step for use with AI agents, ensure the Tool input for new data is configured as a **JSON input type** rather than a Text input. The Insert Knowledge Tool step expects a JSON array of objects (as shown in the formatting examples above), so the Tool input should match this format to properly pass data from agents to the step.

  When setting up the Tool input:

  * Use the **JSON** input type (not Text)
  * Optionally provide a JSON Schema that matches the structure of your knowledge table to guide agents on the expected format
  * Reference the Tool input variable in the Insert Knowledge Tool step's **New data to insert** field using variable mode (`{{input_variable_name}}`)

  This ensures that the data format passed from the Tool input to the Tool step remains consistent and properly structured.
</Accordion>

## Formatting Examples

### Example 1: Insert single row

```json theme={null}
[
  {
    "breed": "poodle",
    "size": "small"
  }
]
```

### Example 2: Insert multiple rows

```json theme={null}
[
  {
    "breed": "poodle",
    "size": "small"
  }, 
  {
    "breed": "great dane",
    "size": "large"
  }
]
```

<Note>
  Ensure to separate each object with a comma if inserting multiple rows.
</Note>

## Using Variables in JSON

You can use variables from Tool inputs and other Tool steps in the **New data to insert** field. For detailed information on how to use variables in JSON for Knowledge Tool steps, including variable syntax and typing, see [Using Variables in JSON](/docs/build/tools/tool-steps/knowledge/using-variables-in-json).

For Knowledge Tool steps, all variables should be wrapped in quotes in the JSON format: `"{{variable_name}}"`. The type of the inserted value is determined by the column type defined in your knowledge table.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What happens if I don't enable 'Sync on Upload'?">
    If "Sync on Upload" is disabled, the data will be stored but not vectorized. This means your agents won't be able to semantically search or reference that data until you manually vectorize it.
  </Accordion>

  <Accordion title="Can I update or delete rows later?">
    Yes. After inserting data, you can manage your knowledge set using the available tools this includes updating or deleting rows as needed.
  </Accordion>

  <Accordion title="What data types can I insert?">
    You can insert text, numbers, booleans, or any JSON-serializable value. Just make sure each object in the array follows the correct formatting.
  </Accordion>
</AccordionGroup>


# Knowledge Search
Source: https://relevanceai.com/docs/build/tools/tool-steps/knowledge/knowledge-search

The 'Knowledge Search' Tool allows you to search knowledge tables with RAG

## Add the 'Knowledge Search' Tool step to your Tool

<div>
  <iframe title="Add the Knowledge Search tool step to your Tool" />
</div>

You can add the 'Knowledge Search' Tool step to your Tool by:

1. Creating a new Tool, then searching for the Knowledge Search Tool step.
2. Click "Expand" to see the full Tool step.
3. Select the Knowledge set to use.
4. Enter your search query in the **Query** field.
5. Select the search type:
   * **Vector**: Finds results based on meaning (semantic similarity). Great when you want answers even if the exact words don’t match.
   * **Keyword**: Finds results based on exact word match. Useful when looking for specific terms or phrases.

## Advanced Settings

### Vector field

The vector field is the column used for searching your knowledge set using semantic similarity. By default, all fields are included unless you choose specific ones.

### Output all fields

* When enabled, all fields from your knowledge set will be shown - including the similarity score.
* When disabled, only the main content field is returned.

### Page size

The number of results to return.

### Similarity threshold

All results are given a score between 0 and 1 based on how similar the result is to your query. This threshold is the minimum score for a result to be returned.

<Tip>
  Increase this value to get fewer but more accurate results. Lower it to be
  more inclusive.
</Tip>

### Raw filters

If you want to filter your search to a specific subset of entries in your knowledge set, you can use raw filters. Filters are defined as JSON objects.

**OR Logic with Arrays:** When using `exact_match` filter type, you can pass an array to the `condition_value` field to create OR semantics. This allows you to match records where the field equals ANY value in the array.

**Examples:**

* Single value: `"condition_value": "x"` → matches records where field equals x
* Array value: `"condition_value": ["a", "b"]` → matches records where field equals a OR b

**Example filter with OR logic:**

```json theme={null}
{
  "field": "call_id",
  "filter_type": "exact_match",
  "condition": "==",
  "condition_value": ["45eb244f-5008-4b7f-bfab-6f4db2189d2e", "4a678557-700e-407f-ab62-23d0decf1b52"]
}
```

This filter will return results where `call_id` matches either of the two provided values.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What is the difference between Knowledge Search and Advanced Knowledge Search?">
    * Knowledge Search is easier to use out of the box, with sensible default
      options that work well for most use cases. - [Advanced Knowledge
      Search](/docs/build/tools/tool-steps/knowledge/advanced-knowledge-search) gives you
      greater control and fine-tuning capabilities for more precise retrieval —
      but may be slower in comparison.
  </Accordion>

  <Accordion title="When should I use Advanced Knowledge Search instead of the regular one?">
    * If you’re unsure which to choose, start with Knowledge Search. - Knowledge
      search works better out of the box, whilst advanced knowledge search gives
      you more control and fine-tuning options.
  </Accordion>
</AccordionGroup>


# Update Knowledge
Source: https://relevanceai.com/docs/build/tools/tool-steps/knowledge/update-knowledge

The 'Update Knowledge' Tool step lets you update existing rows in your knowledge table—making it easy to keep your data current and accurate.

## Add the 'Update Knowledge' Tool step to your Tool

<div>
  <iframe title="Add the Update Knowledge Tool step to your Tool" />
</div>

You can add the 'Update Knowledge' Tool step to your Tool by:

1. Creating a new Tool, then clicking `+ Add Step`
2. Searching for and adding the 'Update Knowledge' Tool step
3. Clicking 'Expand' to see the full Tool step
4. Selecting the knowledge table you would like to access
5. Filter condition: "and" vs "or" when matching data to update:
   * Use `and` when all specified conditions must match for a row to be updated.
   * Use `or` when any of the conditions can match.
   <Note>
     Filter condition only applies if you include more than one object in the filter
   </Note>
6. Filters: Accepts a JSON selection of objects, where each object corresponds to a row. Add more objects as outlined in the formatting examples below to match multiple rows at once.
7. Update value: Provide an object with the new values to update. Each key should match a column name in your knowledge table, and its corresponding value will replace the current value in the matching row.
8. Use the **Sync on upload** field to determine whether the data should be vectorized when added to the knowledge set. In most cases, you'll want this enabled as it allows your agents to semantically search and retrieve relevant information from your knowledge table.
9. Click `Run step` to test out the Tool

## Common errors

<Accordion title="Formatting issues">
  This indicates that there is a formatting issue in the "New data to insert" input field. Make sure your object is formatted correctly. `Expected property name or '}' in JSON at...`
</Accordion>

<Accordion title="Incorrect data type">
  This means the data being inserted into the knowledge table is not a JSON. `Invalid JSON`
</Accordion>

## Formatting Examples

### Example 1: Replace all poodles with bulldogs

```json theme={null}
// Filters
[
  {
    "breed": "poodle"
  }
]
```

```json theme={null}
// Update value
[
  {
    "breed": "bulldog"
  }
]
```

### Example 2: Replace all small poodles with large bulldogs

```json theme={null}
// Filters
[
  {
    "breed": "poodle",
    "size": "small"
  }
]
```

```json theme={null}
// Update value
[
  {
    "breed": "bulldog",
    "size": "large"
  }
]
```

### Example 3: Replace all poodles *and* bulldogs with cats

```json theme={null}
// Filters
[
  {
    "breed": "poodle"
  },
  {
    "breed": "bulldog"
  }
]
```

```json theme={null}
// Update value
[
  {
    "breed": "cat"
  }
]
```

<Note>
  Ensure to separate each object with a comma if filtering multiple rows.
</Note>

## Using Variables in JSON

You can use variables from Tool inputs and other Tool steps in the **Filters** and **Update value** fields. For detailed information on how to use variables in JSON for Knowledge Tool steps, including variable syntax and typing, see [Using Variables in JSON](/docs/build/tools/tool-steps/knowledge/using-variables-in-json).

For Knowledge Tool steps, all variables should be wrapped in quotes in the JSON format: `"{{variable_name}}"`. The type of the updated value is determined by the column type defined in your knowledge table, and values will be automatically converted to match the column type during the update.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What happens if I don't enable 'Sync on Upload'?">
    If "Sync on Upload" is disabled, the data will be stored but not vectorized. This means your agents won't be able to semantically search or reference that data until you manually vectorize it.
  </Accordion>

  <Accordion title="How do I update a field to be empty/blank?">
    You can update a field to be empty/blank by setting the value to empty quote marks `""`.
  </Accordion>

  <Accordion title="What data types can I update with?">
    You can use text, numbers, booleans, or any JSON-serializable value. Just make sure each object in the array follows the correct formatting.
  </Accordion>
</AccordionGroup>


# Using Variables in JSON
Source: https://relevanceai.com/docs/build/tools/tool-steps/knowledge/using-variables-in-json

Learn how to use variables from Tool inputs and other Tool steps in JSON fields for Knowledge Tool steps

When building Tools with Knowledge Tool steps (Insert, Update, Get, Delete Knowledge), you can reference inputs from Tool inputs or outputs from other Tool steps using variable syntax `{{variable_name}}`. This allows you to dynamically insert data from previous steps or user inputs into your knowledge operations.

## Variable Syntax

You can reference:

* **Tool inputs**: `{{input_variable_name}}`
* **Previous Tool step outputs**: `{{steps.step_name.output.variable_name}}`

Press the  button next to any JSON input field to switch to variable mode and reference these values.

## Variable Typing for Knowledge Tool Steps

For Knowledge Tool steps (Insert, Update, Get, Delete), **all variables should be wrapped in quotes** in the JSON format: `"{{variable_name}}"`.

The type of the inserted value is determined by the column type defined in your knowledge table. The system will automatically convert the value to match the column type during insertion.

Make sure the column types in your knowledge table match the data types you want to store:

* **Number columns**: Values will be converted to numbers on insert, even though they're quoted in the JSON
* **Text/String columns**: Values will be stored as text
* **Boolean columns**: Values will be stored as booleans

## Examples

### Example: Using Variables in Filters

```json theme={null}
[
  {
    "breed": "{{user_breed_input}}",
    "size": "{{user_size_input}}",
    "age": "{{previous_step.output.age}}"
  }
]
```

### Example: Using Variables in Data Objects

```json theme={null}
[
  {
    "breed": "{{user_breed_input}}",
    "size": "{{user_size_input}}",
    "age": "{{user_age_input}}",
    "weight": "{{previous_step.output.weight}}",
    "is_active": "{{previous_step.output.is_active}}"
  }
]
```

In these examples, all variables are wrapped in quotes as required for Knowledge Tool steps. The final type depends on your knowledge table column types:

* If the `breed` and `size` columns are text columns, the values will be inserted as strings
* If the `age` and `weight` columns are number columns, the values will be converted to numbers on insert, even though they're quoted in the JSON
* The `is_active` column type determines whether the value is stored as a boolean

## Variable Typing for Other Tool Steps

**Note**: When passing values to other Tool steps that accept JSON input (not Knowledge Tool steps), the typing works differently:

* `"{{variable}}"` will be passed as a string
* `{{variable}}` (without quotes) will be passed as a number or boolean. The JSON input box may show a syntax error, but it will still run correctly.

## See Also

* [Insert Knowledge](/docs/build/tools/tool-steps/knowledge/insert-knowledge)
* [Update Knowledge](/docs/build/tools/tool-steps/knowledge/update-knowledge)
* [Get Knowledge](/docs/build/tools/tool-steps/knowledge/get-knowledge)
* [Delete Knowledge](/docs/build/tools/tool-steps/knowledge/delete-knowledge)


# 'RapidAPI - Get a LinkedIn Profile/Company and Posts' Tool step
Source: https://relevanceai.com/docs/build/tools/tool-steps/linkedin/get-linkedin-profile

Fetch a LinkedIn profile / company and its posts

The 'RapidAPI - Get a LinkedIn Profile/Company and Posts' Tool step allows you to search through LinkedIn profiles through Relevance AI.

<Note>
  You can learn more about the LinkedIn integration [here](/docs/integrations/popular-integrations/linkedin).
</Note>

## Add the 'RapidAPI - Get a LinkedIn Profile/Company and Posts' Tool step to your Tool

<div>
  <iframe title="Add the RapidAPI - Get a LinkedIn Profile/Company and Posts Tool step to your Tool" />
</div>

You can add the 'RapidAPI - Get a LinkedIn Profile/Company and Posts' Tool step to your Tool by:

1. Creating a new Tool, then searching for the 'RapidAPI - Get a LinkedIn Profile/Company and Posts' Tool step
2. Click 'Expand' to see the full Tool step
3. Enter the URL of the company / profile you want to get in the URL field
4. Select the Method you want to use in the Method field. You can choose from:
   * Get User Posts
   * Get User Profile
   * Get Company Posts
   * Get Company Profile
5. Click 'Run step' to test / run the Tool step, then click 'data' to see your output

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Why are LinkedIn Tools so expensive?">
    This Tool uses [RapidAPI](https://rapidapi.com/) to perform actions in LinkedIn. The credits charged are to cover the cost of this service through RapidAPI. We recommend upgrading to a subscription in order to use this Tool.
  </Accordion>
</AccordionGroup>


# 'Unipile - LinkedIn Action' Tool step
Source: https://relevanceai.com/docs/build/tools/tool-steps/linkedin/linkedin-action

Perform actions on LinkedIn using this Tool step

The 'Unipile - LinkedIn Action' Tool step allows you to search through LinkedIn profiles through Relevance AI.

<Note>
  You can learn more about the LinkedIn integration [here](/docs/integrations/popular-integrations/linkedin).
</Note>

## Add the 'Unipile - LinkedIn Action' Tool step to your Tool

<div>
  <iframe title="Add the Unipile - LinkedIn Action Tool step to your Tool" />
</div>

You can add the 'Unipile - LinkedIn Action' Tool step to your Tool by:

1. Creating a new Tool, then searching for the 'Unipile - LinkedIn Action' Tool step
2. Click 'Expand' to see the full Tool step
3. Choose the Action you want to perform in the Action field
4. Connect or select your LinkedIn account in the 'LinkedIn Account' field
5. Enter the parameters for the action you selected, and then click 'Run step' to run the Tool step

## Parameters

Parameters differ for each action in this Tool step, but the LinkedIn account is always required.

<Tabs>
  <Tab title="Get All Chats">
    <ParamField type="string">
      A filter to target items created before the datetime (exclusive). Must be an ISO 8601 UTC datetime (YYYY-MM-DDTHH:MM:SS.sssZ).
    </ParamField>

    <ParamField type="string">
      A filter to target items created after the datetime (exclusive). Must be an ISO 8601 UTC datetime (YYYY-MM-DDTHH:MM:SS.sssZ).
    </ParamField>

    <ParamField type="string">
      A cursor for pagination purposes. To get the next page of entries, you need to make a new request and fulfill this field with the cursor received in the previous request. This process should be repeated until all entries have been retrieved.
    </ParamField>

    <ParamField type="string">
      The maximum number of results to return. This can be between 1 to 250.
    </ParamField>
  </Tab>

  <Tab title="Get All Messages From Chat">
    <ParamField type="string">
      The ID of the chat.
    </ParamField>

    <ParamField type="string">
      A filter to target items created before the datetime (exclusive). Must be an ISO 8601 UTC datetime (YYYY-MM-DDTHH:MM:SS.sssZ).
    </ParamField>

    <ParamField type="string">
      A filter to target items created after the datetime (exclusive). Must be an ISO 8601 UTC datetime (YYYY-MM-DDTHH:MM:SS.sssZ).
    </ParamField>

    <ParamField type="string">
      A cursor for pagination purposes. To get the next page of entries, you need to make a new request and fulfill this field with the cursor received in the previous request. This process should be repeated until all entries have been retrieved.
    </ParamField>

    <ParamField type="string">
      The maximum number of results to return. This can be between 1 to 250.
    </ParamField>
  </Tab>

  <Tab title="Get User Profile">
    <ParamField type="string">
      The username of who you wish to message. This can be found in their profile URL after [https://www.linkedin.com/in/](https://www.linkedin.com/in/)
    </ParamField>

    <ParamField type="checkbox">
      Whether the profile visit should be notified to the viewee or not. Default is false.
    </ParamField>
  </Tab>

  <Tab title="Create A Post">
    <ParamField type="string">
      The caption to your post.
    </ParamField>

    <ParamField type="string">
      The attachments to your post.
    </ParamField>
  </Tab>

  <Tab title="Send Comment">
    <ParamField type="string">
      The ID of the post to comment on
    </ParamField>

    <ParamField type="string">
      The caption for your comment.
    </ParamField>
  </Tab>

  <Tab title="Get InMail Balance">
    No parameters are available for this action.
  </Tab>

  <Tab title="Send Invitation">
    <ParamField type="string">
      The username of who you wish to message. This can be found in their profile URL after [https://www.linkedin.com/in/](https://www.linkedin.com/in/)
    </ParamField>

    <ParamField type="string">
      A message to go with the invitation. This is max. 200 characters for free accounts, and max. 300 characters for Premium accounts.
    </ParamField>

    <ParamField type="string">
      The conversation you would like messages sent to.
    </ParamField>
  </Tab>

  <Tab title="Send Message In A Chat">
    <ParamField type="string">
      The ID of the chat.
    </ParamField>

    <ParamField type="string">
      The message you wish to send.
    </ParamField>
  </Tab>

  <Tab title="Start New Chat">
    <ParamField type="dropdown">
      Choose from Classic, Recruiter and Sales Navigator.
    </ParamField>

    <ParamField type="string">
      The username of who you wish to message. This can be found in their profile URL after [https://www.linkedin.com/in/](https://www.linkedin.com/in/)
    </ParamField>

    <ParamField type="string">
      Add a title to the message.
    </ParamField>

    <ParamField type="string">
      The message you wish to send.
    </ParamField>
  </Tab>
</Tabs>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Why are LinkedIn Tools so expensive?">
    This Tool uses [Unipile](https://unipile.com/) to perform actions in LinkedIn. The credits charged are to cover the cost of this service through Unipile. We recommend upgrading to a subscription in order to use this Tool.
  </Accordion>
</AccordionGroup>


# 'Crustdata - Search LinkedIn Profiles' Tool step
Source: https://relevanceai.com/docs/build/tools/tool-steps/linkedin/search-linkedin-profiles

Search through LinkedIn profiles using this Tool step

The 'Crustdata - Search LinkedIn Profiles' Tool step allows you to search through LinkedIn profiles through Relevance AI.

<Note>You can learn more about the LinkedIn integration [here](/docs/integrations/popular-integrations/linkedin).</Note>

## Add the 'Crustdata - Search LinkedIn Profiles' Tool step to your Tool

<div>
  <iframe title="Add the Crustdata - Search LinkedIn Profiles Tool step to your Tool" />
</div>

You can add the 'Crustdata - Search LinkedIn Profiles' Tool step to your Tool by:

1. Creating a new Tool, then searching for the 'Crustdata - Search LinkedIn Profiles' Tool step
2. Click 'Expand' to see the full Tool step
3. Enter a keyword(s) in the 'Keyword' field
4. Adjust the limit of results you wish to receive in the 'Limit' field
5. Open Advanced Settings to search through many more parameters
6. Click 'Run step' to run the Tool step and then click 'data' to view the output

## Advanced Settings

The majority of parameters to search through can be found in Advanced Settings.

<ParamField type="string">
  The first name field to use for filtering.
</ParamField>

<ParamField type="string">
  The last name field to use for filtering.
</ParamField>

<ParamField type="string">
  The current company to use for filtering. Can provide the company name or LinkedIn URL.
</ParamField>

<ParamField type="string">
  Companies to exclude from current company field. Can provide the company name or LinkedIn URL.
</ParamField>

<ParamField type="string">
  The past company to use for filtering. Can provide the company name or LinkedIn URL.
</ParamField>

<ParamField type="string">
  Companies to exclude from past company field. Can provide the company name or LinkedIn URL.
</ParamField>

<ParamField type="dropdown">
  The type of company to use for filtering.
</ParamField>

<ParamField type="string">
  The current title to use for filtering.
</ParamField>

<ParamField type="string">
  Titles to exclude from the current title field.
</ParamField>

<ParamField type="string">
  The past title to use for filtering.
</ParamField>

<ParamField type="string">
  The headquarters location of the company.
</ParamField>

<ParamField type="string">
  Headquarters locations to exclude.
</ParamField>

<ParamField type="dropdown">
  The headcount size of the company.
</ParamField>

<ParamField type="string">
  The region to use for filtering.
</ParamField>

<ParamField type="string">
  Regions to exclude.
</ParamField>

<ParamField type="string">
  The industry to use for filtering.
</ParamField>

<ParamField type="string">
  Industries to exclude.
</ParamField>

<ParamField type="dropdown">
  The profile language of the person.
</ParamField>

<ParamField type="dropdown">
  The seniority level to use for filtering.
</ParamField>

<ParamField type="dropdown">
  Seniority levels to exclude.
</ParamField>

<ParamField type="dropdown">
  The number of years at the current company.
</ParamField>

<ParamField type="dropdown">
  The number of years in the current position.
</ParamField>

<ParamField type="dropdown">
  The total years of experience.
</ParamField>

<ParamField type="dropdown">
  The department to use for filtering.
</ParamField>

<ParamField type="dropdown">
  Departments to exclude.
</ParamField>

<ParamField type="checkbox">
  Whether the profile has posted on LinkedIn in the last 30 days.
</ParamField>

<ParamField type="checkbox">
  Whether the profile recently changed jobs in the last 90 days.
</ParamField>

<ParamField type="checkbox">
  Whether the profile has been in the news.
</ParamField>

<ParamField type="string">
  The school attended.
</ParamField>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Why are LinkedIn Tools so expensive?">
    This Tool uses [Crustdata](https://crustdata.com/) to search through LinkedIn profiles. The credits charged are to cover the cost of this service through CrustData. We recommend upgrading to a subscription in order to use this Tool.
  </Accordion>

  <Accordion title="Do I need to connect my LinkedIn account to run this Tool?">
    No, connecting your LinkedIn account is not required for this Tool step.
  </Accordion>
</AccordionGroup>


# LLM Tool step
Source: https://relevanceai.com/docs/build/tools/tool-steps/llms/llm-tool-step

Use a LLM as a step in your Tools

The LLM Tool Step is a versatile Tool step in Relevance AI that allows you to inject the power of large language models into your tools. It supports complex reasoning, classification, summarization, and text generation tasks — tailored to your use case.

## Add the LLM Tool step to your Tool

<div>
  <iframe title="Add the LLM Tool step to your Tool" />
</div>

You can add the LLM Tool step to your Tool by:

1. Creating a new Tool, then clicking 'LLM' under Steps
2. Click 'Expand' to see the full Tool step
3. Add the Prompt you want to the Prompt field
4. Choose the model you want to use
5. Click the Settings icon if you want to access Advanced Settings

## Advanced Settings

If you're a pro at using LLMs, you can also access these Advanced Settings.

### Fallback Model

You can select a Fallback Model if the model you've selected fails for any reason. This is a good setting to select to make sure your Tool works!

### System prompt

A system prompt is composed of notes, instruction and guides normally guiding AI to assume a certain role, or to follow a specific format, or limitations.

An example of a System prompt would be: `You are an expert on the solar system. Answer the following questions in a concise and informative manner`.

### Temperature

Temperature is a hyperparameter, ranging in (0,1), that affects the randomness (sometimes referred to as creativity) of the LLMs' response.

Higher randomness/creativity/diversity is expected of higher temperature values.

However, responses might also lose the right context.

### Conversation history

Click on `+ Add row` and you can add lines of conversation taking place between a "user" and "ai".

<Info>The only acceptable roles are "ai" and "user".</Info>

Conversation history is useful in conversational agents and help AI to know more about the situation.

### Thinking / Reasoning Configuration

For OpenAI, Claude and Gemini models, you can select a thinking configuration. Enabling this will make the model 'think' more before answering, which is useful for complex problems.

Reasoning / thinking tokens are charged as output tokens and will be added to the cost of your LLM Tool step.

<Warning>If model and vendor options are mismatched, this option will be ignored.</Warning>

### Force response format

For certain models such as OpenAI, response format can be forced using this field.

## Common errors

<AccordionGroup>
  <Accordion title="Prompt is too long. Please reduce prompt in length.">
    The error message below indicate that the provided prompt includes more tokens than what the choses model allows. To resolve the issue, you can use a model that supports higher number of tokens.

    ```
    1. 400: {"message":"aviary.backend.llm.error_handling.PromptTooLongError: Input too long. Received 5002 tokens, but the maximum input length is 4090 tokens.","internal_message":"aviary.backend.server.openai_compat.openai_exception.OpenAIHTTPException","code":400,"type":"PromptTooLongError","param":{}}
    2. Token limit for each model
    ```
  </Accordion>

  <Accordion title="Too large data">
    When using Relevance to handle large inputs by selecting the most relevant entries, if the input data is too large, you need to upload it as a dataset use it as a [knowledge](/docs/get-started/core-concepts/knowledge) in your Tool. Maximum size for non-knowledge data is 131,072 tokens (\~90kb).

    ```
    Data is too large for the 'Most relevant data'. Consider adding the data to knowledge.
    ```
  </Accordion>

  <Accordion title="Rate limit">
    This error happens when the used API key is set to a different rate limit compared to what Relevance uses by default. Trying again with different intervals of pause helps with this issue.

    ```
    429: {"message":"Rate limit reached for default-gpt-4 in organization org-... on tokens per min. Limit: 40000 / min. 
    Please try again in 1ms. Contact us through our help center at help.openai.com if you continue to have issues.",
    "type":"tokens","param":null,"code":"rate_limit_exceeded"}
    ```
  </Accordion>

  <Accordion title="Temperature errors">
    There is a temperature parameter under LLM advanced settings. The below error occurs if the entered value is out of the accepted (0,1) range.

    ```
    400: {"message":"-0.5 is less than the minimum of 0 - 'temperature'","type":"invalid_request_error","param":null,"code":null}
    ```
  </Accordion>
</AccordionGroup>


# Python Tool step
Source: https://relevanceai.com/docs/build/tools/tool-steps/python-code/code-python

Empower your agent with custom code using one of the world's most popular and beginner-friendly languages

The Python Tool step allows you to use Python to take outputs from other Tool steps and manipulate data for other Tool steps and LLMs.

<Warning>In order to use this Tool step, we would recommend having a beginner level of understanding of Python programming, or use our Chat to walk you through implementing it at [chat.relevanceai.com](https://chat.relevanceai.com).</Warning>

## What is Python?

Python is a widely used programming language known for its clarity and simplicity. It's popular among beginners and experts alike because it's easy to read, quick to write, and versatile enough for tasks ranging from web development and automation to data analysis and machine learning. With a massive library ecosystem and strong community support, Python is a powerful tool for solving real-world problems efficiently.

## Python Package Ecosystem

One of Python's greatest strengths is its extensive package ecosystem. You can install and use any package available on the [Python Package Index (PyPI)](https://pypi.org/), which hosts over 693,000 projects with millions of releases. This means you have access to virtually any functionality you need across countless domains.

### Popular Packages by Use Case

<AccordionGroup>
  <Accordion title="Data Analysis & Manipulation">
    * `pandas` - Data manipulation and analysis
    * `numpy` - Numerical computing and arrays
    * `matplotlib` - Data visualization and plotting
    * `seaborn` - Statistical data visualization
    * `plotly` - Interactive data visualization
  </Accordion>

  <Accordion title="Web Scraping & HTTP Requests">
    * `requests` - HTTP library for making web requests
    * `beautifulsoup4` - Web scraping and HTML parsing
    * `scrapy` - Web crawling framework
    * `selenium` - Browser automation and testing
  </Accordion>

  <Accordion title="Machine Learning & AI">
    * `scikit-learn` - Machine learning library
    * `tensorflow` - Deep learning framework
    * `pytorch` - Deep learning and neural networks
    * `transformers` - Pre-trained language models
    * `openai` - OpenAI API client
  </Accordion>

  <Accordion title="Data Processing & File Handling">
    * `openpyxl` - Excel file manipulation
    * `python-docx` - Word document processing
    * `PyPDF2` - PDF file manipulation
    * `pillow` - Image processing library
    * `csv` - Built-in CSV file handling
  </Accordion>

  <Accordion title="API Integration & Communication">
    * `flask` - Lightweight web framework
    * `fastapi` - Modern web API framework
    * `requests-oauthlib` - OAuth authentication
    * `twilio` - SMS and communication APIs
    * `stripe` - Payment processing
  </Accordion>

  <Accordion title="Database & Storage">
    * `sqlalchemy` - SQL toolkit and ORM
    * `psycopg2` - PostgreSQL adapter
    * `pymongo` - MongoDB driver
    * `redis` - Redis database client
  </Accordion>
</AccordionGroup>

To find and explore available packages, visit [PyPI.org](https://pypi.org/) where you can search through the vast collection of Python packages. Simply add any package you need to your Python Tool step using the 'PyPI Packages' section.

## Adding the Python Tool step into your Tool

<div>
  <iframe title="Add the Python Tool step to your Tool" />
</div>

You can add the Python Tool step to your Tool by:

1. Creating a new Tool, then clicking or searching for the Python Tool step
2. Click '+ New item' under 'PyPI Packages' if you need to install any packages, and enter the package name you need.
3. Enter your code in the 'Code' field, including any imports you need.
4. Click 'Run step' to run the Tool step

<Info>In order to access the output of this codeblock, you will need to return a value in the `return` statement.</Info>

## Accessing variables

You can access Tool input variables, other Tool step outputs, and API keys you've created in Relevance in your Python Tool step.

### Input variables

You can access Input variables by simply using the variable name directly, where 'name' is the variable for your input.

```python3 theme={null}
name
```

### Other Tool step outputs

You can access other Tool step outputs by using the variable name directly, where 'llm' is the variable for the Tool step you're trying to access the output of.

```python3 theme={null}
llm
```

This returns the full output object. To access a specific field within that output, use dictionary notation:

```python3 theme={null}
llm["answer"]
```

For example, if an LLM step returns `{"answer": "Hello!"}`, then:

* `llm` returns `{"answer": "Hello!"}`
* `llm["answer"]` returns `"Hello!"`

### API keys

You can access one of your saved API keys by using this syntax, where 'api\_key' is the title of your API key.

```python3 theme={null}
"{{secrets.chains_api_key}}"
```

## OAuth Authentication

Python code steps support OAuth authentication through the `Integration` helper class, allowing you to make authenticated API calls to third-party services without manually managing OAuth tokens.

To use OAuth authentication:

1. Add an OAuth account input to your tool and assign it a variable name
2. Access the account ID in your Python code via `params['your_oauth_input_name']`
3. Create an `Integration` instance with the provider name and account ID
4. Use the `api_call()` method to make authenticated requests

For complete documentation on the `Integration` class, including detailed examples, method parameters, and best practices, see the [Python Helper Functions documentation](/docs/build/tools/tool-steps/python-code/code-python-helper-functions#integration).

<Note>
  While Python code steps can use OAuth authentication through the Integration helper class, they do not support [User Level Authentication](/docs/enterprise/user-level-authentication). If you need individual users to authenticate with their own accounts, you should use dedicated OAuth-based tool steps instead of Python code.
</Note>

## Advanced Settings

<Note>Several advanced settings (Session ID, Number of GPUs, Number of CPU Cores, Memory Size, and Session Timeout) are only available when the backend is set to 'Modal Labs'.</Note>

### Backend

You can choose between two backend services to run the code, 'Modal Labs' or 'Daytona'.

### Runtime Commands

You can include a list of Linux terminal commands to be run before running your code.

### Session ID

Define a custom session that saves all local variables between Python executions.

### Enable 'Long Output' Mode

Allow the output to be longer than 127,000 characters, in exchange for 2-4 seconds of extra latency.

### Number of GPUs

Allows you to specify a number of GPUs for your code to use.

### Number of CPU Cores

Allows you to specify a number of CPU cores for your code to use.

### Memory Size

Allows you to specify an amount of RAM for your code to use.

### Session Timeout

Allows you to specify (in seconds) how long the code should run before timing out.

### Raise Error

Allows you to specify how errors are raised:

* **Error with Full Traceback** - Shows the complete error traceback including file names, line numbers, and the full call stack. This is useful for debugging complex issues.
* **Error** - Shows only the error message without the detailed traceback. This provides a cleaner output for production use.
* **Print Error to Stderr** - Prints the error to stderr instead of raising it, allowing your code to continue execution even when errors occur.

### Enable fallback

If the selected backend fails, when this is selected, your code will attempt to run via the other backend.

## Common errors

<AccordionGroup>
  <Accordion title="Code execution timed out">
    This error indicates that the code took longer than the seconds allowed in the 'Session timeout' Advanced Setting, or exceeded our hard limit of four hours.
  </Accordion>

  <Accordion title="No return statement">
    This error indicates that there is no return statement in the code. You will not be able to access the output of this codeblock unless you pass data in a return statement. If the transformed section of the output is showing `null`, it means that there is nothing being returned from your code.

    ```
    WARNING: There is no return statement in this code. To access the output of this codeblock, return a value.
    ```
  </Accordion>
</AccordionGroup>

## Additional resources

<Columns>
  <Card title="Helper Functions" href="/build/tools/tool-steps/python-code/code-python-helper-functions">
    Access functions that can help you code Python Tool steps
  </Card>

  <Card title="Transformations" href="/build/tools/tool-steps/python-code/code-studio-transformation-python-helper-functions">
    Use in-built Tool steps in Python
  </Card>
</Columns>


# Python Helper Functions
Source: https://relevanceai.com/docs/build/tools/tool-steps/python-code/code-python-helper-functions

Natively Supported Helper Functions

This client provides an interface to interact with the Relevance API. It includes functions to run various steps, insert and retrieve data, and upload temporary files.

## **Functions**

### Insert data

```python Python theme={null}
insert_data(dataset_id: str, data: List[Dict[str, Any]])
```

Inserts data into a Relevance dataset.

**Arguments**

* `dataset_id`: The ID of the dataset to insert into.
* `data`: A list of dictionaries containing the data to insert.

**Returns**

* The response from the API as a JSON object.

### Retrieve data

```python Python theme={null}
retrieve_data(dataset_id: str, page_size: int = None, include_fields: List[str] = None)
```

Retrieves data from a Relevance dataset.

**Arguments**

* `dataset_id`: The ID of the dataset to retrieve from.
* `page_size`: The number of results to return per page (optional).
* `include_fields`: A list of fields to include in the response (optional).

**Returns**

* The response from the API as a JSON object.

### Retrieve All Data

```python Python theme={null}
retrieve_all(dataset_id: str, page_size: int = 1000, include_fields: List[str] = None) -> List[Dict[str, Any]]:
```

Retrieves all data from a Relevance dataset, paginated to handle large datasets.

**Arguments**

* `dataset_id`: The ID of the dataset to retrieve from.
* `page_size`: The number of results to return per page. Defaults to 1000 (optional).
* `include_fields`: A list of fields to include in the response. Defaults to None (optional).

**Returns**

* A list of dictionaries containing the retrieved data. Each dictionary represents a document from the dataset.

Example

### Upload a temporary file

```python Python theme={null}
insert_temp_file(file_path_or_bytes: str, ext: str = None)
```

Uploads a temporary file to Relevance.

**Arguments**

* `file_path_or_bytes`: The path to the file or the file contents as bytes.
* `ext`: The file extension (optional).

**Returns**

* A dictionary containing the download URL of the uploaded file.

### Prompt completion

```python Python theme={null}
prompt_completion(prompt: str, model: int = None)
```

Runs the `prompt_completion` step with the given prompt and model.

**Arguments**

* `prompt`: The prompt to complete.
* `model`: The model to use for completion (optional).

**Returns**

* The response from the API as a JSON object.

### Run a step

```python Python theme={null}
run_step(step_name: str, params: Dict[str, Any])
```

Runs a Relevance step with the given name and parameters.

**Arguments**

* `step_name`: The name of the step to run.
* `params`: A dictionary of parameters to pass to the step.

**Returns**

* The response from the API as a JSON object.

## **Classes**

### Integration

```python Python theme={null}
Integration(provider_name: str, account_id: str)
```

The `Integration` class provides a convenient way to make authenticated API calls to OAuth-connected services in your Python code. It automatically handles OAuth token management and makes authenticated requests to third-party APIs.

**Constructor Arguments**

* `provider_name`: The name of the OAuth provider/integration (e.g., `'dataforseo'`, `'hubspot'`, `'slack'`).
* `account_id`: The account ID from your OAuth account input. This is accessed via `params['your_oauth_input_name']` where `your_oauth_input_name` is the variable name of your OAuth account input.

**Methods**

#### api\_call()

```python Python theme={null}
integration.api_call(method: str, url: str, body: Dict[str, Any] = None, headers: Dict[str, str] = None, params: Dict[str, Any] = None)
```

Makes an authenticated HTTP request to the specified URL using the OAuth credentials associated with the integration.

**Arguments**

* `method`: The HTTP method to use (e.g., `'GET'`, `'POST'`, `'PUT'`, `'DELETE'`).
* `url`: The full URL endpoint to make the request to.
* `body`: Optional dictionary or list containing the request body (for POST, PUT, etc.). Supports both JSON objects (dictionaries) and JSON arrays (lists). Will be automatically serialized to JSON.
* `headers`: Optional dictionary of additional HTTP headers to include in the request.
* `params`: Optional dictionary of query parameters to append to the URL.

**Returns**

* The API response as a parsed JSON object (dictionary). The response is automatically parsed, so you can directly access the data without additional JSON parsing.

## **Usage Examples**

### Insert data

```python Python theme={null}
data = [{"field1": "value1", "field2": "value2"}, {"field1": "value3", "field2": "value4"}]
response = insert_data("my_dataset", data)
```

### Retrieve data

```python Python theme={null}
response = retrieve_data("my_dataset", page_size=10, include_fields=["field1", "field2"])
```

### Retrieve all

```python Python theme={null}
response = retrieve_all("my_dataset", page_size=500, include_fields=["field1", "field2"])
```

### Upload a temporary file

Note: Make sure to replace the `region` variable with your actual region.

```python Python theme={null}
file_path = "path/to/file.txt"
response = insert_temp_file(file_path)
```

### Prompt completion

```python Python theme={null}
response = prompt_completion("My prompt", model="openai-gpt35")
```

### Run a step

```python Python theme={null}
response = run_step("my_step", {"param1": "value1", "param2": "value2"})
```

### Integration

```python Python theme={null}
# Get OAuth account ID from input parameter
account_id = params['my_oauth_account']

# Create Integration instance
integration = Integration('google', account_id)

# Make authenticated API call with object body
response = integration.api_call(
    method='GET',
    url='https://www.googleapis.com/oauth2/v2/userinfo',
    params={'limit': 10}
)

# Access the parsed JSON response
user_info = response
```

### Integration with Array Body

```python Python theme={null}
# Example: Bulk update operation with array body
account_id = params['my_oauth_account']
integration = Integration('confluence', account_id)

# Make API call with array body
response = integration.api_call(
    method='PUT',
    url='https://your-domain.atlassian.net/wiki/api/v2/pages/123/properties',
    body=[
        {"key": "status", "value": "reviewed"},
        {"key": "priority", "value": "high"}
    ],
    headers={'accept': 'application/json'}
)
```


# 'Slack - Retrieve Messages' Tool step
Source: https://relevanceai.com/docs/build/tools/tool-steps/slack/retrieve-messages

Retrieve messages from a Slack channel using Relevance AI

The 'Slack - Retrieve Messages' Tool step allows you to retrieve messages from a Slack channel in Relevance AI.

<Note>You can learn more about the Slack integration [here](/docs/integrations/popular-integrations/slack).</Note>

<Warning>This integration for Slack was developed by our team and is not affiliated with Slack. Please contact our Support for any questions/queries you may have.</Warning>

## Add the 'Slack - Retrieve Messages' Tool step to your Tool

<div>
  <iframe title="Add the Python Tool step to your Tool" />
</div>

You can add the 'Slack - Retrieve Messages' Tool step to your Tool by:

1. Creating a new Tool, then searching for the 'Slack: Retrieve Messages' Tool step
2. Click 'Expand' to see the full Tool step
3. Select your Slack account
4. Enter the name of the Slack channel you want to retrieve messages from
5. Specify the number of messages you want to retrieve
6. Test your Tool step by clicking 'Run step'


# 'Slack - Send Message' Tool step
Source: https://relevanceai.com/docs/build/tools/tool-steps/slack/send-message

Send a message to a Slack channel from Relevance AI

The 'Slack - Send Message' Tool step allows you to send a message to one of your Slack channels from Relevance AI.

<Note>You can learn more about the Slack integration [here](/docs/integrations/popular-integrations/slack).</Note>

<Warning>This integration for Slack was developed by our team and is not affiliated with Slack. Please contact our Support for any questions/queries you may have.</Warning>

## Add the 'Slack - Send Message' Tool step to your Tool

<div>
  <iframe title="Add the Python Tool step to your Tool" />
</div>

You can add the 'Slack - Send Message' Tool step to your Tool by:

1. Creating a new Tool, then searching for the 'Slack: Send Message' Tool step
2. Click 'Expand' to see the full Tool step
3. Select the Slack channel you want to send the message to
4. Add your message in the 'Message' field
5. Click 'Run step' to test out your Tool step


# 'Trello - Add Checklist' Tool step
Source: https://relevanceai.com/docs/build/tools/tool-steps/trello/add-checklist

Add a checklist to a card in Trello

The 'Trello - Add Checklist' Tool step allows you to add a checklist to one of your Trello cards.

<Note>You can learn more about the Trello integration [here](/docs/integrations/popular-integrations/trello).</Note>

## Add the 'Trello - Add Checklist' Tool step to your Tool

<div>
  <iframe title="Add the Python Tool step to your Tool" />
</div>

You can add the 'Trello - Add Checklist' Tool step to your Tool by:

1. Creating a new Tool, then searching for the 'Trello - Add Checklist' Tool step
2. Click 'Expand' to see the full Tool step
3. Select or add your Trello account
4. Select the board your card is on
5. Select the card you want to add a checklist to
6. Name the checklist in the 'Checklist Name' field
7. If you wish to copy the items from another checklist, add this checklist's name to 'Copy from Checklist'
8. Use the 'Position' field to choose whether the checklist is added above or below other checklists on the card
9. Click 'Run step' to test out the Tool step!

<Info>You will not be able to add checklist items to the checklist item you create in this Tool step unless you copy items from another checklist using the 'Copy from Checklist' field. To create new items, you should add another Tool / Tool step using the 'Trello - Create a Checklist Item' Tool step.</Info>


# 'Trello - Add Comment' Tool step
Source: https://relevanceai.com/docs/build/tools/tool-steps/trello/add-comment

Add a comment to a card in Trello

The 'Trello - Add Comment' Tool step allows you to add a comment to one of your Trello cards.

<Note>You can learn more about the Trello integration [here](/docs/integrations/popular-integrations/trello).</Note>

## Add the 'Trello - Add Comment' Tool step to your Tool

<div>
  <iframe title="Add the Python Tool step to your Tool" />
</div>

You can add the 'Trello - Add Comment' Tool step to your Tool by:

1. Creating a new Tool, then searching for the 'Trello - Add Comment' Tool step
2. Click 'Expand' to see the full Tool step
3. Select or add your Trello account
4. Select the board your card is on
5. Select the card you want to add a comment to
6. Add the text that you want to add in a comment to 'Comment'
7. Click 'Run step' to run your Tool step


# 'Trello - Archive Card' Tool step
Source: https://relevanceai.com/docs/build/tools/tool-steps/trello/archive-card

Archive a card in Trello

The 'Trello - Archive Card' Tool step allows you to archive a card in Trello from Relevance AI.

<Note>You can learn more about the Trello integration [here](/docs/integrations/popular-integrations/trello).</Note>

## Add the 'Trello - Archive Card' Tool step to your Tool

<div>
  <iframe title="Add the 'Trello - Archive Card' Tool step to your Tool" />
</div>

You can add the 'Trello - Archive Card' Tool step to your Tool by:

1. Creating a new Tool, then searching for the 'Trello - Add Comment' Tool step
2. Click 'Expand' to see the full Tool step
3. Select or add your Trello account
4. Select the board your card is on
5. Select the card you want to archive
6. Click 'Run step' to run your Tool step


# 'Trello - Create a Checklist Item' Tool step
Source: https://relevanceai.com/docs/build/tools/tool-steps/trello/create-checklist-item

Add a checklist item to a checklist in Trello

The 'Trello - Create a Checklist Item' Tool step allows you to add a checklist item to one of your Trello checklists.

<Note>You can learn more about the Trello integration [here](/docs/integrations/popular-integrations/trello).</Note>

## Add the 'Trello - Create a Checklist Item' Tool step to your Tool

<Note>In order to create a checklist item, you need to create a checklist first. You can do this in Trello or by using the [Add Checklist Tool step](/docs/build/tools/tool-steps/trello/add-checklist).</Note>

<div>
  <iframe title="Add the Trello - Create a Checklist Item Tool step to your Tool" />
</div>

You can add the 'Create a Checklist Item' Tool step to your Tool by:

1. Creating a new Tool, then searching for the 'Create a Checklist Item' Tool step
2. Click 'Expand' to see the full Tool step
3. Select or add your Trello account
4. Add the board your checklist is on
5. Add the card your checklist is on
6. Select the name of the checklist you want to add a checklist item to
7. Add the name of the checklist item in the 'Name' field
8. Use the 'Position' dropdown to specify whether the checklist item is added to the top or bottom of the checklist
9. Use the 'Checked' checkbox to specify whether the checklist item should be created as checked or not
10. Click 'Run step' to test your Tool step!


# Add, create, edit and connect Agents to your Workforce
Source: https://relevanceai.com/docs/build/workforces/build-an-ai-workforce/add-agents

Adding agents to your Workforce allows you to create powerful multi-agent systems where specialized AI agents work together to accomplish complex tasks. 

## Adding Agents to Your Workforce

You can add Agents you've already created to your Workforce by following these instructions

1. Open your Workforce, and drag or click an Agent onto your Workforce
2. Using the right-hand sidebar, select the Agent you want to add to your Workforce

## Create Agents within your Workforce

You can also create Agents within the Workforce builder by following these instructions:

<div>
  <iframe title="How to create Agents from the Workforce builder in Relevance AI" />
</div>

1. Open your Workforce, and drag or click an Agent onto your Workforce
2. Then, when the right-hand sidebar opens, click 'Create new agent'
3. You should then be able to name, describe and change the icon of your Agent
4. Afterwards, you can change your Agent's Prompt, add Tools, and access all sections of your Agent aside from the ones mentioned below from the Workforce builder!

<Warning>
  You cannot edit your Agent's triggers from the Workforce builder, as Workforce's have their own Triggers. You also cannot access your Agent's Advanced Settings.
</Warning>

## Edit Agents within your Workforce

You can also edit your Agents by clicking on them within the Workforce.

<div>
  <iframe title="How to edit Agents in the Workforce builder in Relevance AI" />
</div>

1. First, click on the Agent you want to edit
2. Then, the right-hand sidebar should open, allowing you to make any edits you need
3. After you've made an edit, you will need to save your Agent before continuing to make edits to other parts of your Workforce

<Warning>
  When you edit an Agent from the Workforce builder, it changes the settings for that Agent everywhere. We do not have Workforce-specific Agent settings.
</Warning>

## Connect Agents in your Workforce together

Once you've added two or more Agents to your Workforce, you can connect them together by following these instructions:

1. Click and drag from one Agent to create a connection line to another Agent
2. When connecting Agents, you'll be prompted to select a connection type:

* **AI connection**: The Agent makes its own decision on whether to talk to another Agent
* **Next step**: An Agent will always hand over what it was doing to the next Agent, regardless of its finished state

3. After creating a connection, click on the connecting line to configure edge settings
   * For '**AI connection**' connections:
     * Complete the "When to call this agent" box with natural language instructions
     * Set approval mode (auto run, approval required, or let agent decide)
     * Configure max auto runs as a guardrail
     * Choose between 'always start a new task' or 'always continue with one task'
   * For '**Next step**' connections:
     * Choose between 'always start a new task' or 'always continue with one task'
4. Click the 'Save' button to save your configuration

## Connection type

When connecting Agents, you can configure how they communicate with each other:

### AI connection

Use natural language to explain how the Agent should behave. The Agent will follow your instructions but will escalate if it needs help or clarification.

* **Communication Settings**: Explain in natural language when this Agent should be called. For example, if a manager Agent is delegating to another Agent, explain the logic for when to delegate to that specific Agent.
* **Approval mode**:
  * **Auto run**: The Agent will complete the task without approval
  * **Approval required**: The Agent will draft the task but ask for approval before completing
  * **Let Agent decide**: The Agent will complete the task if confident, otherwise will ask for approval
  * **Max auto runs**: Set a limit on the number of times the Agent auto runs

### Next step

This setting forces one Agent to pass the task to another Agent. The handover must happen regardless of the first Agent's state.

## Best Practices for Adding Agents

1. **Define Clear Roles**
   * Give each Agent a specific role and responsibility within your workforce
   * Avoid overlapping responsibilities that could cause confusion
2. **Create Logical Workflows**
   * Arrange Agents in a way that follows a natural progression of tasks
   * Consider the information flow between Agents when designing your workforce
3. **Use Descriptive Connection Instructions**
   * When using "AI connection", provide clear, detailed instructions about when to delegate
   * Include specific criteria or conditions that should trigger the handover
4. **Test Incrementally**
   * Add and test Agents one at a time to ensure each connection works as expected
   * Use the Task View to monitor how Agents interact with each other
5. **Set Appropriate Approval Levels**
   * For critical tasks, use "Approval required" to maintain human oversight
   * For routine tasks, consider "Auto run" to maximize efficiency

## Monitoring Your Multi-Agent Workforce

After adding Agents to your Workforce:

1. Go to the 'Task' view to see your Workforce in action
2. Here you can trigger your Workforce via manual messages/conversation
3. Monitor when specific Agents need approvals
4. Observe how Agents communicate and hand off tasks to each other

## Troubleshooting

<AccordionGroup>
  <Accordion title="Agents aren't communicating as expected">
    Check the connection settings between Agents and ensure the instructions are clear and specific.
  </Accordion>

  <Accordion title="Workflow stops unexpectedly">
    Check if an Agent is waiting for approval in the Task View.
  </Accordion>

  <Accordion title="Agent makes incorrect decisions about when to delegate">
    Refine the natural language instructions in the "When to call this agent" field.
  </Accordion>
</AccordionGroup>


# Add Conditions
Source: https://relevanceai.com/docs/build/workforces/build-an-ai-workforce/add-conditions

Conditions in Workforce act as powerful decision-making components that allow you to create intelligent workflows with conditional logic.

## Overview

Conditions enable you to build dynamic, responsive workflows by creating branching paths based on specific criteria. Rather than having a linear workflow where every step always executes in the same order, conditions let you implement decision points that determine which path your workflow should take.

Workforce offers two types of conditions to suit different use cases:

* **LLM-based conditions**: Use natural language to describe complex logic that requires interpretation
* **Rule-based conditions**: Use structured filters for deterministic, precise comparisons

## When to Use Conditions

Conditions are particularly valuable when you need to:

* Process data differently based on specific values or states
* Create branching workflows with different paths for different scenarios
* Implement quality control checks before proceeding to critical steps
* Filter out certain items from further processing
* Execute tools only on specific days, times, or under certain circumstances

## Adding a Condition to Your Workforce

To add a condition to your workforce:

1. Navigate to the "Build" section of your Workforce
2. Locate the "Conditions" option in the node selection panel
3. Drag a condition node onto your workspace
4. Connect the condition to the appropriate tools or agents

## Choosing a Condition Type

When you configure a condition node, you'll see a **Condition type** dropdown with two options:

### LLM-based Conditions

LLM-based conditions use natural language to evaluate complex logic. The AI interprets your description and determines whether the condition is met.

**Best for:**

* Complex logic that's easier to express in natural language
* Evaluating unstructured data like text content
* Scenarios requiring interpretation or context understanding
* Time-based conditions (e.g., "if it's a weekday")

**Example use cases:**

* "Go down this path if the email contains an urgent request"
* "Follow this route if the customer sentiment is negative"
* "Use this path if today is a weekday"

### Rule-based Conditions

Rule-based conditions use structured filters to compare specific values. These are deterministic and don't require AI interpretation.

**Best for:**

* Precise comparisons of known values
* Checking tool outputs against expected values
* Numeric comparisons
* Validating that specific fields exist or have certain values
* Performance-critical paths where you need guaranteed behavior

**Example use cases:**

* Checking if a tool's status output equals "SUCCESS"
* Verifying a numeric value is greater than a threshold
* Confirming a required field is not empty

## Configuring LLM-based Conditions

Setting up LLM-based conditions is simple thanks to natural language configuration:

1. Click on your condition node to open its settings
2. Select **LLM-based** from the Condition type dropdown
3. Enter a natural language description of your condition, such as:
   * "Go down this path if the day is a weekday"
   * "Follow this route if the customer score is above 80"
   * "Use this path if the email contains the word 'urgent'"

## Configuring Rule-based Conditions

Rule-based conditions let you build precise filters using a structured interface:

### Basic Setup

1. Click on your condition node to open its settings
2. Select **Rule-based** from the Condition type dropdown
3. Click **Add filter** to create your first rule

### Filter Components

Each filter has three parts:

#### 1. Target

The **Target** is the value you want to evaluate. Use the variable picker to select from upstream node outputs.

The picker displays available variables in this format:

```
<Node Name> / output / <field name>
```

For example:

* `Check Status / output / result` - the "result" field from a tool called "Check Status"
* `Get User Data / output / email` - the "email" field from a tool called "Get User Data"

<Note>The target resolves to whatever value that field contains. If the field contains an object, you're comparing the entire object, not a nested property within it.</Note>

#### 2. Operator

Choose how to compare the target value:

| Operator        | Description                      | Example Use Case                   |
| --------------- | -------------------------------- | ---------------------------------- |
| **equals**      | Exact match                      | Check if status equals "SUCCESS"   |
| **contains**    | Target includes the source value | Check if text contains "urgent"    |
| **starts with** | Target begins with source value  | Check if ID starts with "CUST-"    |
| **ends with**   | Target ends with source value    | Check if filename ends with ".pdf" |
| **exists**      | Target field has any value       | Verify a field is present          |
| **empty**       | Target field is empty or null    | Check if optional field is blank   |
| **>**           | Greater than (numeric)           | Check if score > 80                |
| **\<**          | Less than (numeric)              | Check if count \< 10               |
| **>=**          | Greater than or equal to         | Check if age >= 18                 |
| **\<=**         | Less than or equal to            | Check if price \<= 100             |

#### 3. Source

The **Source** is the value you're comparing against. You can either:

* Enter a constant value (e.g., "SUCCESS", 100, "urgent")
* Use the variable picker to select another field from upstream nodes

### Match Modes

When you have multiple filters, choose how they should be evaluated:

* **Match All** (AND logic): All filters must be true for the condition to pass
* **Match Any** (OR logic): At least one filter must be true for the condition to pass

### Understanding Target Paths

The target path determines what value gets evaluated. Here's how it works:

**Example 1: Simple string output**

```
Tool: "Check Status"
Output: "SUCCESS"
Target: Check Status / output
Comparison: equals "SUCCESS" ✓
```

**Example 2: Object output**

```
Tool: "Validate Data"
Output: {"status": "SUCCESS", "code": 200}
Target: Validate Data / output
Comparison: equals {"status": "SUCCESS", "code": 200} ✓
```

<Warning>If a tool returns an object like `{"status":"SUCCESS"}`, the target picker shows `Tool Name / output`, which refers to the entire object. You cannot drill into nested fields like `status` through the picker.</Warning>

### Nested Field Limitation and Workaround

Rule-based conditions cannot access nested fields within objects. If your tool returns:

```json theme={null}
{
  "status": "SUCCESS",
  "data": {
    "user_id": 123
  }
}
```

You **cannot** directly access `status` or `data.user_id` through the target picker.

**Workaround:** Structure your tools to output flat values as separate named outputs:

Instead of returning a single object, configure your tool to return:

* Output 1: `status` = "SUCCESS"
* Output 2: `user_id` = 123

This way, each value is accessible as a separate target in your rule-based condition.

## Practical Examples

### Example 1: Check Tool Success Status

**Scenario:** Only proceed if a validation tool succeeded

**Setup:**

* Condition type: Rule-based
* Filter 1:
  * Target: `Validate Input / output / status`
  * Operator: equals
  * Source: "SUCCESS"

### Example 2: Numeric Threshold Check

**Scenario:** Route high-value orders differently

**Setup:**

* Condition type: Rule-based
* Filter 1:
  * Target: `Calculate Total / output / amount`
  * Operator: >
  * Source: 1000

### Example 3: Multiple Conditions (AND Logic)

**Scenario:** Process only if status is complete AND priority is high

**Setup:**

* Condition type: Rule-based
* Match mode: Match All
* Filter 1:
  * Target: `Get Status / output / status`
  * Operator: equals
  * Source: "complete"
* Filter 2:
  * Target: `Get Priority / output / level`
  * Operator: equals
  * Source: "high"

### Example 4: Multiple Conditions (OR Logic)

**Scenario:** Flag items that are either urgent or high-value

**Setup:**

* Condition type: Rule-based
* Match mode: Match Any
* Filter 1:
  * Target: `Check Priority / output / priority`
  * Operator: equals
  * Source: "urgent"
* Filter 2:
  * Target: `Calculate Value / output / amount`
  * Operator: >=
  * Source: 5000

### Example 5: Check Field Exists

**Scenario:** Only proceed if an optional field was provided

**Setup:**

* Condition type: Rule-based
* Filter 1:
  * Target: `Get Input / output / optional_field`
  * Operator: exists

### Example 6: String Pattern Matching

**Scenario:** Route customer IDs that start with a specific prefix

**Setup:**

* Condition type: Rule-based
* Filter 1:
  * Target: `Get Customer / output / customer_id`
  * Operator: starts with
  * Source: "PREMIUM-"

## Best Practices

### When to Use Each Condition Type

**Use LLM-based conditions when:**

* You need to evaluate natural language or unstructured text
* The logic is complex and easier to express in plain English
* You're working with time-based or contextual conditions
* Exact matching isn't critical

**Use rule-based conditions when:**

* You need precise, deterministic comparisons
* You're validating tool outputs against known values
* You're working with numeric thresholds
* Performance and reliability are critical
* You want to avoid AI interpretation overhead

### General Tips

1. **Keep conditions focused**: Each condition should evaluate a clear, specific criterion
2. **Test thoroughly**: Verify your conditions work with different inputs and edge cases
3. **Use descriptive node names**: Make it easy to identify what each condition checks
4. **Consider the data structure**: For rule-based conditions, ensure your tools output data in a format that's easy to filter
5. **Combine condition types**: Use LLM-based for complex logic and rule-based for precise checks in the same workflow

## Troubleshooting

### Rule-based Condition Not Working

**Problem:** Your rule-based condition isn't evaluating correctly

**Solutions:**

1. Check that the target path points to the correct field
2. Verify the operator matches your comparison type (e.g., use numeric operators for numbers)
3. Ensure the source value matches the expected data type
4. Test with a simpler condition to isolate the issue
5. Check if the upstream tool is actually outputting the expected value

### Cannot Access Nested Fields

**Problem:** You need to check a value inside an object, but it's not available in the picker

**Solution:** Restructure your tool to output flat values as separate named outputs instead of nested objects.

### Condition Always Passes or Fails

**Problem:** Your condition doesn't seem to evaluate properly

**Solutions:**

1. For rule-based: Verify the target field actually contains data
2. For LLM-based: Make your natural language description more specific
3. Check the match mode (Match All vs Match Any) if using multiple filters
4. Review the execution logs to see what values are being compared

## Related Features

* [Condition to Tool Configuration](/docs/build/workforces/workforce-features/condition-to-tool-configuration): Learn how to connect conditions to tools
* [Edge Settings](/docs/build/workforces/build-an-ai-workforce/edge-settings): Configure how nodes connect in your workforce
* [Add Tools](/docs/build/workforces/build-an-ai-workforce/add-tools): Learn about adding and configuring tools in your workforce


# Add Tools
Source: https://relevanceai.com/docs/build/workforces/build-an-ai-workforce/add-tools

Tools are essential components in your AI Workforce that enable agents to perform specific actions and complete tasks.

## What Are Tools in Workforce?

In the Workforce, tools function as independent nodes that can be connected to agents, triggers, and even other tools. Unlike the traditional approach where tools are tightly coupled to specific agents, Workforce allows for a more flexible and modular design:

* Tools exist as standalone components on the workspace
* Multiple agents can access the same tool
* Tools can be connected to other tools to create sequential workflows
* Tools can be easily substituted without having to reconfigure each agent

<Note>When you edit a tool that is used by multiple agents, a warning banner appears in the tool editor showing how many agents are affected. You can clone the tool to make changes that only apply to specific agents. See [Editing shared tools](/docs/build/tools/create-a-tool#editing-shared-tools) for details.</Note>

## How to Add Tools to Your Workforce

Adding tools to your Workforce is a straightforward process:

1. Open your Workforce and navigate to the "Build" section
2. Locate the tools panel in the Workforce interface
3. Drag any tool from your project onto the workspace
4. Position the tool where you want it on the canvas

## Connecting Tools in Your Workforce

Once you've added tools to your workspace, you can create powerful connections:

### Agent to Tool Connections

To connect an agent to a tool:

1. Click on the agent node to reveal connection points
2. Drag a connection line from the agent to your desired tool
3. Configure the connection type:
   * **AI connection**: The agent will determine when to use the tool based on natural language instructions
   * **Next step**: The agent will always use the tool when the connection is triggered

### Tool to Tool Connections

One of the most powerful features of the Workforce Builder is the ability to connect tools to other tools:

1. Click on the source tool to reveal connection points
2. Drag a connection line to the destination tool
3. This creates a workflow where the output of one tool becomes the input for the next

For example, you might connect a tool that sends an email to another tool that summarizes responses, creating an automated follow-up workflow.

## Edge Settings for Tool Connections

When connecting tools in your Workforce, you can configure the following edge settings:

### Connection type

* **AI connection**: Use natural language to explain when the tool should be used
* **Next step**: Always execute the tool when the connection is triggered

### Approval mode

* **Auto run**: The tool will execute without requiring approval
* **Approval required**: The tool will draft the action but ask for human approval before executing
* **Let agent decide**: The agent will determine whether to auto-run or request approval based on context
* **Max auto runs**: Set limits on how many times a tool can auto-run as a safety guardrail

## Tool to Tool Workflows

The ability to connect tools directly to other tools is what makes the Workforce Builder so powerful. This workflow-style approach allows you to:

1. Create sequential processes where each tool performs a specific function
2. Build complex automations without writing code
3. Easily modify or replace individual components without disrupting the entire workflow
4. Maintain a visual representation of your business processes

For example, after sending an email with one tool, you could automatically hand over to another tool that monitors for responses, and then to a third tool that generates appropriate follow-ups.

## Using Conditions with Tools

For more advanced workflows, you can add conditions to your tool connections:

1. Add a condition node to your workspace
2. Connect a tool to the condition
3. Connect the condition to other tools
4. Configure the condition logic (e.g., "only send email if day of week is weekday")

Conditions act as IF statements in your workflow, allowing for branching logic and more sophisticated automations.

## Best Practices for Tools in Workforce Builder

To get the most out of tools in your Workforce:

1. **Keep tools focused**: Each tool should perform a specific, well-defined function
2. **Use descriptive names**: Name your tools clearly to understand their purpose at a glance
3. **Document connections**: Use the description fields to explain how tools should work together
4. **Test thoroughly**: Before deploying, test your tool connections with various inputs
5. **Monitor performance**: Regularly review how your tools are performing and refine as needed

## Related Features

* [Workforce Task View](https://relevanceai.com/docs/workforce-features-and-capabilities/workforce-task-view): Monitor and manage your workforce tasks and tool executions
* [Edge Settings](https://relevanceai.com/docs/build-an-ai-workforce/edge-settings): Configure how nodes connect and communicate in your workforce
* [Add Conditions](https://relevanceai.com/docs/build-an-ai-workforce/add-conditions): Learn how to add conditional logic to your workforce flows

By effectively utilizing tools in the Workforce Builder, you can create sophisticated, autonomous workflows that handle complex business processes with minimal human intervention.


# Add Triggers
Source: https://relevanceai.com/docs/build/workforces/build-an-ai-workforce/add-triggers

Triggers are the starting points that initiate workflows in your AI Workforce. They determine when and how your agents begin their work, serving as the entry points to your entire workflow system.

## Understanding Triggers

Triggers are specialized nodes in the Workforce that activate your agents based on specific events or conditions. They function as the initial connection point in your workflow, determining when your AI agents should start performing their tasks.

### Types of Triggers

#### Manual Triggers

When you first create a workforce, a manual trigger is automatically added to your workspace. This allows you to directly initiate your workflow by sending a message to your agent, similar to how you would interact with a standalone agent in Relevance AI.

Manual triggers are ideal for:

* Testing your workflow during development
* On-demand tasks that require human initiation
* Workflows that need to begin with specific human input or context

#### Integration Triggers

Integration triggers connect your workforce to external systems, allowing your agents to automatically respond to events from those systems. These triggers enable your AI Workforce to monitor and react to activities in your connected tools and platforms.

Common integration triggers include:

* Email triggers that activate when new messages arrive
* Calendar triggers that respond to meeting invitations or reminders
* CRM triggers that react to new lead creation or status changes
* Webhook triggers that respond to custom events from any system

#### Recurring Schedule Triggers

Recurring schedule triggers allow your AI Workforce to run automatically at set times without manual intervention. This is perfect for automating repetitive tasks that need to happen on a regular basis.

Use recurring schedules for:

* Daily reports that need to be generated every morning
* Weekly data processing tasks
* Monthly analytics and summaries
* Periodic system checks and maintenance tasks

#### External Triggers (Zapier)

External triggers like Zapier are configured and managed outside of Relevance AI. In Zapier, you can use the "Message Workforce" event to send data to your Workforce whenever a Zap fires — connecting your Workforce to any app or service that Zapier supports without needing a native integration in Relevance AI.

Since Zapier triggers are managed entirely within Zapier's platform, refer to Zapier's documentation for setup and configuration of specific Zaps.

## Adding Triggers to Your Workforce

### Adding Your First Trigger

1. When you open your workforce, you'll see a manual trigger already available in your workspace.
2. Connect this trigger to an agent by dragging a line from the trigger to your chosen agent.
3. This connection establishes the entry point for your workflow.

### Adding Additional Triggers

To add more triggers to your workforce:

1. Navigate to the Workforce interface.

2. Locate the "Triggers" section in the node palette.

3. Drag your desired trigger type onto the workspace.

4. Position it appropriately in your workflow diagram.

5. Connect the trigger to the agent that should respond first when the trigger activates.

## Configuring Triggers

Each trigger type has specific configuration options that determine its behavior:

### Manual Trigger Configuration

To trigger your workforce manually:

1. Switch to the "Task" tab in the Relevance AI interface.
2. Use the message input field to send a manual message to your workforce.
3. Provide natural language instructions that will initiate your workflow.
4. Your connected agent will receive this message and begin processing according to your workflow configuration.

This manual interaction allows you to start your workflow with specific context and instructions whenever needed.

### Integration Trigger Configuration

Integration triggers require specific setup to connect with external systems:

1. Select the integration trigger in your workspace.
2. Complete the authentication process for the connected service.
3. Configure trigger conditions (e.g., specific email subjects, calendar event types).
4. Set any filters to ensure the trigger only activates for relevant events.
5. Test the connection to verify proper setup.

### Recurring Schedule Trigger Configuration

Recurring schedules can be configured using either simple schedule options or advanced cron expressions:

#### Simple Schedule Options

| Schedule Type | What It Does                        | Perfect For                          |
| ------------- | ----------------------------------- | ------------------------------------ |
| **Daily**     | Runs every day at a time you choose | Morning reports, daily check-ins     |
| **Weekly**    | Runs on specific days each week     | Team updates, weekly newsletters     |
| **Monthly**   | Runs on chosen dates each month     | Monthly reports, regular maintenance |
| **Custom**    | Runs on specific dates you select   | One-off tasks or irregular schedules |

#### Advanced: Cron Expressions

For more precise control over scheduling, you can use **Cron expressions**. Relevance AI uses **AWS EventBridge cron format**, which requires **six fields**:

```
minute hour day-of-month month day-of-week year
```

**Cron Expression Format:**

| Field            | Values          | Wildcards      |
| ---------------- | --------------- | -------------- |
| **Minute**       | 0-59            | , - \* /       |
| **Hour**         | 0-23            | , - \* /       |
| **Day-of-month** | 1-31            | , - \* ? / L W |
| **Month**        | 1-12 or JAN-DEC | , - \* /       |
| **Day-of-week**  | 1-7 or SUN-SAT  | , - \* ? L #   |
| **Year**         | 1970-2199       | , - \* /       |

**Special Characters:**

* `*` (asterisk) - all values
* `?` (question mark) - no specific value (used in day-of-month or day-of-week)
* `-` (dash) - range of values
* `,` (comma) - additional values
* `/` (forward slash) - increments
* `L` - last day of month or week
* `W` - nearest weekday
* `#` - nth occurrence of a day in a month

**Common Cron Expression Examples:**

```
0 10 * * ? *        # Every day at 10:00 AM
0 0 * * MON *       # Every Monday at midnight
0 9 1 * ? *         # First day of every month at 9:00 AM
0 0 * * MON-FRI *   # Weekdays at midnight
*/15 * * * ? *      # Every 15 minutes
0 9,17 * * ? *      # Daily at 9:00 AM and 5:00 PM
0 0 1,15 * ? *      # 1st and 15th of every month at midnight
0 8 ? * 2 *         # Every Monday (day 2) at 8:00 AM
0 10 L * ? *        # Last day of every month at 10:00 AM
0 9 ? * 2#1 *       # First Monday of every month at 9:00 AM
```

<Note>
  AWS EventBridge cron expressions use 6 fields, unlike traditional Unix cron which uses 5 fields. Always include the year field, and use `?` for either day-of-month or day-of-week when specifying the other.
</Note>

## Pausing triggers

### Pause all triggers

<div>
  <iframe title="How to pause all triggers" />
</div>

The "Pause all triggers" feature allows you to temporarily stop integrated triggers in your workforce without disconnecting integrations or deleting trigger configurations. This feature is accessible from the workforce interface and provides a quick way to control automated workflow execution.

#### What gets paused

* Integration triggers (Gmail, Slack, CRM, etc.)
* Webhook triggers
* Recurring schedule triggers
* All other integrated triggers

#### What continues working

* Manual triggers (user message received)
* Direct user interactions

<Note>
  Manual triggers like "user message received" are designed to always remain active and cannot be paused. This ensures users can always interact with the workforce, even when integrated triggers are paused.
</Note>

## Best Practices for Using Triggers

### Strategic Trigger Placement

* **Start with clarity**: Position triggers at the logical beginning points of your workflows.
* **Minimize redundancy**: Avoid having multiple triggers that serve the same purpose.
* **Consider user experience**: For manual triggers, ensure they're intuitive and provide clear guidance.

### Trigger Maintenance

* Regularly test triggers to ensure they're functioning correctly.
* Monitor trigger activity to identify any performance issues.
* Update trigger configurations when connected systems change.

### Recurring Schedule Best Practices

* **Start small**: Begin with one simple recurring task before setting up complex schedules.
* **Test thoroughly**: Run your workforce manually first to ensure it works as expected.
* **Monitor initially**: Check the first few automated runs to confirm everything works correctly.
* **Use cron for precision**: When you need exact timing control, use cron expressions instead of simple schedules.
* **Consider dependencies**: Ensure your workforce can complete within the scheduled interval to avoid overlapping runs.

### Security Considerations

* Review authentication settings for integration triggers regularly.

* Implement appropriate access controls for manual triggers.

* Consider data privacy implications when configuring trigger conditions.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Why is my trigger not activating?">
    Verify connection settings and authentication. For integration triggers, check that the connected service has the proper permissions. For recurring schedules, ensure the cron expression or schedule configuration is correct.
  </Accordion>

  <Accordion title="How do I prevent duplicate activations?">
    Check for overlapping trigger conditions. Review all triggers in your workforce to ensure they don't respond to the same events. For recurring schedules, verify that you don't have multiple schedules set for the same time.
  </Accordion>

  <Accordion title="Why are my workflow responses delayed?">
    Examine workflow complexity and agent configuration. Complex workflows with multiple agents may take longer to complete. Check your agent settings and consider optimizing the workflow structure.
  </Accordion>

  <Accordion title="Can I temporarily pause a recurring schedule?">
    Yes, you can disable recurring schedule triggers without deleting them. Simply toggle off the schedule in your workforce's trigger settings when you need a pause.
  </Accordion>

  <Accordion title="What timezone are scheduled runs based on?">
    Scheduled runs use your account's configured timezone. You can adjust this in your workforce settings to ensure tasks run at the correct local time.
  </Accordion>

  <Accordion title="What happens when I click 'Pause all triggers'?">
    When you click "Pause all triggers", all integrated triggers in your workforce are temporarily disabled. This includes integration triggers (Gmail, Slack, CRM monitors), webhook triggers, and recurring schedule triggers. However, manual triggers like "user message received" continue to work normally, allowing you to interact with your workforce through the Task View at any time.
  </Accordion>

  <Accordion title="Why does my workforce still respond to messages after pausing all triggers?">
    Manual triggers, such as "user message received", are designed to always remain active and cannot be paused. These triggers represent direct user interactions and ensure you can always communicate with your workforce. The "Pause all triggers" feature only affects integrated triggers (integrations, webhooks, and schedules) that run without user initiation.
  </Accordion>

  <Accordion title="How do I test my triggers?">
    Test triggers based on their type:

    **For Manual and Integration Triggers:**

    1. Navigate to the "Task" tab in your Workforce
    2. For manual triggers: Send a test message to initiate the workflow
    3. For integration triggers: Trigger a test event from the connected service
    4. Monitor the execution in the [Workforce Task View](https://relevanceai.com/docs/build/workforces/workforce-features/workforce-task-view) to verify proper activation

    **For Recurring Schedule Triggers:**

    * **Note:** Test activations are not currently supported for recurring schedule triggers
    * **Best practice:** Test your workflow manually first using a manual trigger before setting up the recurring schedule
    * Monitor the first few automated runs after activation
    * Check the Task View after scheduled runs to verify execution

    **Checking Logs and Error Messages:**

    1. **Workforce Task View:** Navigate to the "Task" section in your Workforce to view:
       * All current and recent activities
       * Failed tasks with error messages
       * Agent notes and execution details
       * Complete conversation histories
       * [Learn more about Workforce Task View](https://relevanceai.com/docs/build/workforces/workforce-features/workforce-task-view)

    2. **Tool Logs:** Within individual tools, access the "Logs" tab to view execution details. [Learn more about Tools](https://relevanceai.com/docs/get-started/core-concepts/tools)

    3. **Agent Task History:** For agents with recurring schedules, view error details in the agent's task history. [Learn more about Agent Triggers](https://relevanceai.com/docs/build/agents/build-your-agent/triggers)

    **Additional Verification Steps:**

    * Verify that connected agents have appropriate permissions
    * Ensure all necessary API keys and credentials are current
    * Check trigger configuration settings (filters, conditions, authentication)
  </Accordion>
</AccordionGroup>


# Edge Settings
Source: https://relevanceai.com/docs/build/workforces/build-an-ai-workforce/edge-settings

Edge Settings control how nodes are connected to each other in your AI Workforce.

## Connection types

When connecting nodes in your Workforce, you can choose between two primary connection types:

### AI connection

This option gives your agent autonomy to determine when to use a connection based on natural language instructions.

**Configuration Options:**

* **Communication Settings**: Provide natural language instructions explaining when and how the agent should use this connection. For example, if you're connecting a manager agent to a specialist agent, you might write: "Delegate to this agent when a customer inquiry requires technical expertise about our enterprise product features."
* **Approval mode**: Control how the agent executes tasks through this connection:
  * **Auto run**: The agent will automatically execute the task without requiring approval.
  * **Approval required**: The agent will draft the task but wait for your approval before completing it.
  * **Let agent decide**: The agent will complete the task if it has sufficient context and confidence; otherwise, it will ask for human approval or clarification.
* **Max auto runs**: Set a limit on the number of times the agent can automatically execute this connection without human intervention:
  * **No limit**: The agent can use this connection as many times as needed.
  * **Custom limit**: Specify a maximum number of automatic executions as a guardrail.
* **Task behavior**: Choose how tasks flow through this connection:
  * **Create new task**: Each time this connection is used, it creates a separate task instance.
  * **Continue same task**: Maintain continuity by keeping all work within the same task context.

### Next step

This option creates a mandatory connection where one node always passes control to another node when triggered.

**Configuration Options:**

* **Task behavior**: Choose how tasks flow through this forced connection:
  * **Create new task**: Each handover creates a separate task instance.
  * **Continue same task**: Maintain continuity by keeping all work within the same task context.

## Practical applications

### Agent-to-Agent communication

When connecting agents, edge settings determine how they collaborate:

* **Manager-Specialist Relationship**: Configure a manager agent to delegate specific types of inquiries to specialist agents using "AI connection" with clear instructions about when to delegate.
* **Sequential Processing**: Use "Next step" to create a workflow where one agent always passes work to another after completing its part, such as having a research agent always pass findings to a content creation agent.

### Agent-to-Tool communication

Edge settings control how agents interact with tools:

* **Conditional Tool Usage**: With "AI connection", an agent can determine when a particular tool is appropriate based on the context.
* **Approval Workflows**: For sensitive operations like sending emails or making purchases, configure "Approval required" to ensure human oversight.
* **Automated Sequences**: Use "Next step" to ensure an agent always follows a specific process, such as always logging customer interactions after completing a support conversation.

### Tool-to-Tool communication

Create sophisticated automation workflows:

* **Data Processing Pipelines**: Connect tools in sequence to transform, analyze, and act on data without requiring agent intervention at each step.
* **Conditional Branching**: Use conditions between tools to create decision trees that process information differently based on specific criteria.

## Best practices

1. **Start Simple**: Begin with basic connections and gradually increase complexity as you understand how your Workforce operates.
2. **Use Clear Instructions**: When using "AI connection", provide specific, unambiguous instructions about when and how to use connections.
3. **Implement Guardrails**: Use approval requirements and auto-run limits for critical operations to maintain control while allowing automation.
4. **Test Thoroughly**: Verify that your edge settings produce the expected behavior across different scenarios before deploying to production.
5. **Monitor Performance**: Regularly review how your connections are being used and refine edge settings based on actual performance.

## Parallel Tool Calls compatibility <Badge>Beta</Badge>

When an agent has [Parallel Tool Calls](/docs/build/agents/build-your-agent/agent-settings/language-model#parallel-tool-calls-beta) enabled, set outbound connections from that agent to use **Create new task** as their Task Behavior to take advantage of parallel execution. Parallel handoffs require independent task instances that can execute simultaneously.

Connections set to "Continue same task" still work — but those handoffs run serially, so you won't see the speed benefit Parallel Tool Calls is meant to provide. The Workforce Builder highlights these connections in yellow as a reminder.

**How to update a connection for parallel execution:**

1. Select the highlighted (yellow) connection in the Workforce Builder
2. Open **Edge Settings**
3. Set **Task Behavior** to **Create new task**

Repeat for each connection on agents that have Parallel Tool Calls enabled.

<Info>
  Parallel Tool Calls is enabled per agent in [Advanced Settings > Language Model](/docs/build/agents/build-your-agent/agent-settings/language-model#parallel-tool-calls-beta). It is currently a Beta feature.
</Info>

## Limitations

* **One-way Communication**: Currently, connections in the Workforce Builder are limited to one-way communication. Bidirectional communication between nodes is not supported.
* **Decision Complexity**: While conditions can create branching workflows, complex decision trees may require careful planning and potentially multiple condition nodes.

## Related features

The Edge Settings functionality works closely with other Workforce Builder components:

* **Nodes**: The building blocks (triggers, agents, tools, conditions) that are connected using edges.
* **Configurations**: The specific relationship types between nodes that determine workflow possibilities.
* **Workforce Task View**: Where you can monitor your Workforce's operations and approve tasks when required.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can I change edge settings after creating a connection?">
    Yes, you can modify edge settings at any time by selecting the connection in the Workforce Builder interface.
  </Accordion>

  <Accordion title="How do I know if I should use AI connection or Next step?">
    Use AI connection when you want the agent to decide whether to use the connection based on context and natural language instructions. Use Next step when one node should always pass control to another regardless of context.
  </Accordion>

  <Accordion title="What happens if I set a limit on auto runs and the limit is reached?">
    Once the auto run limit is reached, the agent will request human approval before using that connection again.
  </Accordion>

  <Accordion title="Can I create conditional connections based on specific data values?">
    Yes, you can use condition nodes to create branches in your workflow based on specific criteria or data values.
  </Accordion>
</AccordionGroup>


# Introduction to Nodes
Source: https://relevanceai.com/docs/build/workforces/build-an-ai-workforce/introduction-to-nodes

Nodes are the fundamental building blocks of a workforce in Relevance AI. They represent all the components you can place on your workspace to create automated workflows. 

## What Are Nodes?

Nodes are the technical term for all the elements you can place and connect in the Workforce interface. They come from graph theory, where a graph consists of nodes (points) and edges (connections between points). In Workforce Builder:

* **Nodes** are the individual components you place on the workspace
* **Edges** are the connections between nodes that define how they interact

## Types of Nodes

There are four main types of nodes in Workforce:

### 1. Triggers

Triggers are the starting points of your workflows. They initiate the execution of your workforce when specific conditions are met.

**Examples of triggers include:**

* Manual triggers (allowing you to message an agent directly)
* Integration triggers (such as when an email is received)
* Scheduled triggers (running at specific times)

Triggers are essential because they determine when and how your workforce begins its work.

### 2. Agents

Agents are AI-powered entities that can reason, make decisions, and complete tasks. In Workforce, agents are the same as those you've configured elsewhere in Relevance AI, but now they can be visually connected to other components.

Agents can:

* Process information from triggers
* Make decisions about next steps
* Use tools to accomplish tasks
* Hand over work to other agents

### 3. Tools

Tools are specific capabilities that perform defined actions. Unlike in the previous system where tools were tightly coupled to specific agents, in Workforce, tools exist independently and can be connected to multiple agents or other tools.

This modular approach offers several advantages:

* One tool can be used by multiple agents
* Tools can be connected to other tools in sequence
* Tools can be easily substituted (e.g., changing from Notion to Confluence) without having to reconfigure each agent

### 4. Conditions

Conditions act as decision points in your workflow, similar to IF statements in programming. They evaluate specific criteria and direct the flow based on whether those criteria are met.

**For example:**

* A condition might check if the current day is a weekday
* If true, it might trigger an email tool
* If false, the workflow might end or take a different path

Conditions help create more sophisticated, branching workflows that respond dynamically to different situations.

## Working with Nodes in Workforce

### Placing Nodes

To add nodes to your workspace:

1. Open your workforce and navigating to "Build"
2. Drag the desired node type (trigger, agent, tool, or condition) from the sidebar onto the workspace
3. Position it where you want it to appear in your workflow

### Connecting Nodes

To establish relationships between nodes:

1. Click on the source node
2. Drag a connection line to the target node
3. Configure the connection settings (such as communication style)

### Node Configuration

Each node type has specific settings that can be configured:

* **Triggers**: Configure when and how they activate
* **Agents**: Set communication preferences, approval requirements, and behavior settings
* **Tools**: Define inputs, steps, and outputs
* **Conditions**: Specify the criteria to evaluate and the paths to take based on results

## Best Practices for Working with Nodes

1. **Start Simple**: Begin with a manual trigger connected to a single agent to test functionality
2. **Plan Your Workflow**: Before adding many nodes, sketch out the desired workflow to understand the necessary components
3. **Use Meaningful Names**: Give your nodes descriptive names that indicate their purpose
4. **Test Incrementally**: Add and test nodes in small batches rather than building the entire workflow at once
5. **Consider Approval Settings**: For critical actions, configure nodes to require human approval before proceeding

## Common Node Configurations

### Trigger → Agent

This is the most basic configuration, where a trigger initiates an agent. For example, a manual trigger allows you to message an agent directly.

### Agent → Agent

This configuration enables one agent to communicate with another. There are two communication styles:

* **Decide when to use**: The agent determines when to involve another agent based on natural language instructions
* **Next step**: The first agent always passes the task to the second agent

### Agent → Tool

This allows an agent to use a specific tool to complete a task, such as sending a message in a Slack channel.

### Tool → Tool

This creates a workflow-like sequence where one tool's output becomes another tool's input, without requiring agent intervention between steps.

### Condition → Tool

This setup evaluates a condition before deciding whether to execute a tool, adding logic to your workflows.

## Limitations and Considerations

* Currently, communication between agents is limited to one-way communication. Bidirectional communication is not yet supported.
* When designing complex workflows, consider the approval settings carefully to maintain appropriate human oversight.
* The more nodes in a workflow, the more important it becomes to monitor performance in the Task View.

## Related Features

For more information on how to use nodes effectively in your workforce, refer to these related features:

* [Edge Settings](https://relevanceai.com/docs/) - Configure how nodes connect to each other
* [Workforce Task View](https://relevanceai.com/docs/) - Monitor and manage your workforce's activities
* [Approvals and Escalations](https://relevanceai.com/docs/) - Set up human oversight for your automated workflows

By understanding and effectively utilizing nodes, you can create powerful, flexible workflows that automate complex tasks while maintaining appropriate control and oversight.


# Notes
Source: https://relevanceai.com/docs/build/workforces/build-an-ai-workforce/notes

Clarify purpose, instructions, and collaboration using notes for your Workforce.

<img alt="Notes" />

You can now add detailed notes to your workforce, making it easier to organize, explain, and share your AI teams with others!

Workforce Notes provides a dedicated space to document important context about your workforce's purpose, structure, and usage instructions. Whether you're planning your AI team's organization, explaining how agents work together, or preparing to share your workforce in the marketplace, Notes ensures everyone understands how to get the most value.

* Document workforce purpose – Clarify the overall goals and use cases for your AI team
* Provide usage instructions – Help new users understand how to effectively use your workforce
* Organize your planning – Keep track of ideas and improvements for your AI team structure
* Enhance marketplace listings – Make your shared workforces more accessible with clear documentation
* Improve team collaboration – Ensure everyone understands how your workforce is designed to function

With Workforce Notes, you can create more valuable, user-friendly AI teams that are easier to share, understand, and implement across your organization or with the wider community.

## How to add Notes to your Workforce

To add notes to your Workforce, follow these steps:

1. Open your Workforce
2. Drag a Note onto your Workforce
3. Enter the text that you want to add to your Note
4. Save your Workforce to see your changes!


# Create a Workforce
Source: https://relevanceai.com/docs/build/workforces/create-a-workforce

How to create your first Workforce in Relevance AI.

This guide walks you through creating a Workforce. For an overview of what Workforces are and how they work, see [Workforces](/docs/get-started/core-concepts/workforces).

<Frame>
  <iframe title="YouTube video player" />
</Frame>

## Create your first Workforce

1. Click **Workforce** in the left sidebar
2. Click **+ New Workforce** to open the visual canvas
3. Add your Agents to the canvas by dragging them from the sidebar
4. Connect your Agents by drawing lines between them
5. Configure each connection — choose between an **AI connection** (the Agent decides when to hand off) or a **next step** (mandatory transition)
6. Add Tools to your workflow as needed
7. Set up any conditional logic for routing
8. Save your Workforce

<img alt="The Workforce canvas — drag Agents in, draw connections, and configure handoffs." />

<Tip>
  Map out your process before building — know which Agents handle which steps and how they hand off. Simpler Workforces with clearly defined roles tend to perform better than overly complex ones.
</Tip>

## What's next?

Once you've created a Workforce, explore the rest of the build guides to refine it:

* [Add triggers](/docs/build/workforces/build-an-ai-workforce/add-triggers) — configure what starts your Workforce (manual, integration, or scheduled)
* [Introduction to nodes](/docs/build/workforces/build-an-ai-workforce/introduction-to-nodes) — understand the building blocks of your Workforce canvas
* [Add Agents](/docs/build/workforces/build-an-ai-workforce/add-agents) — add and configure Agents on the canvas
* [Add Tools](/docs/build/workforces/build-an-ai-workforce/add-tools) — connect Tools directly within your workflow
* [Add conditions](/docs/build/workforces/build-an-ai-workforce/add-conditions) — create routing rules for smarter workflows
* [Edge settings](/docs/build/workforces/build-an-ai-workforce/edge-settings) — configure how connections between nodes behave
* [Version history](/docs/build/agents/build-your-agent/version-history) — every save creates a snapshot you can restore, so you can experiment without losing working versions


# Share your Workforce
Source: https://relevanceai.com/docs/build/workforces/share-your-workforce

Learn how to share your Workforce with others as a cloneable template

## Overview

You can easily share your Workforce with others by creating a cloneable template. Once you create and share your Workforce's cloneable template, this will allow others to clone your entire Workforce into their own projects.

<Info>
  **Enterprise customers:** When sharing workforces with team members using RBAC, you'll need to grant permissions to both the workforce and all agents within it. Learn more about [Workforce Permissions in RBAC](/docs/enterprise/rbac#workforce-permissions).
</Info>

<Warning>
  Enabling the cloneable template option or any embed option (Chat UI or Chat Widget) makes your Workforce publicly accessible. The resulting link is a public URL that search engines can index and that anyone can share, so the Workforce may be discovered without you sending the link yourself — anyone with it can clone your Workforce or embed it on their site. Disable the option in the Workforce's share settings to revoke access.
</Warning>

## Cloneable template

You can learn how to share your Workforce as a cloneable template by following these steps:

<div>
  <iframe title="Relevanceai Demo" />
</div>

1. Open your Workforce and click 'Run'
2. Then, click 'Share'
3. Toggle on the button to create a cloneable template of your Workforce
4. Click on 'Enable public access' after reading through the message on the screen that pops up
5. Finally, click on 'Copy' to copy the link to your Workforce's cloneable template and share it with others!

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can I share Workforces with free customers?">Unfortunately, the Workforce feature is not available to our free customers, so they will not be able to open the link.</Accordion>
</AccordionGroup>


# Agent to Agent Configuration
Source: https://relevanceai.com/docs/build/workforces/workforce-features/agent-to-agent-configuration

Agent to Agent configuration is a powerful capability within Workforce that enables you to create sophisticated multi-agent systems where agents can collaborate, delegate tasks, and work together to solve complex problems.

## How Agent to Agent Configuration Works

When you connect one agent to another in your workforce, you're establishing a relationship that determines how these agents will interact. This connection creates a pathway for information to flow from one agent to another, allowing for specialized handling of different aspects of a task.

## Setting Up Agent to Agent Connections

To create an Agent to Agent connection in your workforce:

1. Navigate to the "Build" section of your workforce
2. Ensure you have at least two agents added to your workspace
3. Click and drag from the output node of the first agent to the input node of the second agent
4. A connection line will appear, and the edge settings panel will open

## Connection Type Options

When configuring Agent to Agent connections, you have two primary communication style options:

### AI connection

This option gives the first agent autonomy to determine when it should involve the second agent. The agent uses natural language understanding to follow your instructions about when to delegate.

To configure this option:

1. Select "AI connection" in the edge settings
2. Complete the "When to call this agent" field with clear instructions
3. For example: "Call this agent when you need specialized knowledge about financial regulations" or "Delegate to this agent when the customer asks about technical product specifications"

### Next step

This option creates a mandatory transition from one agent to another. When the first agent completes its work, it will always pass control to the second agent.

To configure this option:

1. Select "Forced handover" in the edge settings
2. Choose between "Always start a new task" or "Always continue with one task"

## Task Continuity Options

For both communication styles, you can specify how the task should be handled:

* **Create new task**: The second agent begins a fresh task with its own context
* **Continue same task**: The second agent continues within the same task, maintaining context, conversation history, and short-term memory/metadata. Any metadata set by the first agent will be accessible to the second agent and can be read or updated.

## Best Practices for Agent to Agent Configuration

### When to Use "AI connection"

* For complex workflows where conditional routing is needed
* When you want the first agent to make intelligent decisions about escalation
* In scenarios where not every interaction requires the specialized knowledge of the second agent

### When to Use "Next step"

* For linear workflows with clear handoff points
* When you want to ensure every interaction passes through a specific sequence of agents
* For quality control processes where a second agent should always review the work

### Creating Effective Agent Instructions

When using "AI connection", the quality of your instructions in the "When to call this agent" field is crucial:

* Be specific about the conditions that should trigger delegation
* Include examples of scenarios where the second agent should be involved
* Define clear boundaries of responsibility between agents

## Limitations

* Currently, only one-way communication is supported between agents
* Bidirectional communication (where agents can have back-and-forth conversations) is not yet available

## Example Use Cases

### Customer Support Triage

Connect a general support agent to specialized agents for technical, billing, or account issues. The first agent handles basic inquiries and delegates complex issues to specialists.

### Content Creation Pipeline

Connect a content planning agent to a content writing agent, and then to a content editing agent, creating a streamlined production workflow.

### Research and Analysis

Connect a data collection agent to an analysis agent, where the first agent gathers information and the second performs in-depth analysis of the collected data.

## Troubleshooting

If your Agent to Agent configuration isn't working as expected:

* Verify that the connection line is properly established between agents
* Check that your "When to call this agent" instructions are clear and specific
* Ensure that both agents have the necessary tools and permissions to complete their tasks
* Review the task history in the Workforce Task View to identify where the workflow might be breaking down

By effectively configuring Agent to Agent connections, you can create sophisticated AI workflows that leverage the specialized capabilities of different agents, resulting in more powerful and flexible automation solutions.


# Agent to Tool Configuration
Source: https://relevanceai.com/docs/build/workforces/workforce-features/agent-to-tool-configuration

Agent to Tool configuration is a powerful feature within Workforce that enables you to establish connections between your AI agents and the tools they need to complete tasks. 

## Understanding Agent to Tool Connections

When you connect an agent to a tool in Workforce, you're creating a pathway for your agent to access specific functionality. This connection allows your agent to:

* Make decisions about when to use particular tools
* Pass relevant information to tools for processing
* Receive and interpret results from tool operations
* Continue workflows based on tool outputs

## Setting Up Agent to Tool Configurations

To configure an Agent to Tool connection in your workforce:

1. Open your workforce in the Workforce interface
2. Locate the agent node you want to connect to a tool
3. Click and drag from the agent's output connector to the input connector of your desired tool
4. Once connected, click on the connecting line (edge) to access configuration settings

## Configuration Options

### Connection type

When configuring an Agent to Tool connection, you have two primary communication style options:

#### AI connection

This option gives your agent autonomy to determine when to use the connected tool based on context and need.

* **Communication Settings**: Provide natural language instructions explaining when and how the agent should use this tool. For example: "Use this email tool when you need to send important updates to clients or when follow-up communication is required."
* **Approval mode**: Configure how the agent makes decisions about tool usage:
  * **Auto run**: The agent will use the tool without requiring approval
  * **Approval required**: The agent will draft the tool request but wait for human approval before execution
  * **Let agent decide**: The agent will determine whether to auto-run or request approval based on confidence level
  * **Max auto runs**: Set limits on how many times the agent can automatically use the tool without intervention

#### Next step

This option creates a mandatory connection where the agent must pass information to the tool as part of the workflow.

* **Task Continuity**: Choose between:
  * **Create new task**: Each tool execution begins a fresh task instance
  * **Continue same task**: The tool execution remains part of the original task flow

## Best Practices for Agent to Tool Configuration

### 1. Clear Instructions

When using "Let agent decide" mode, provide specific, unambiguous instructions about when the tool should be used. For example:

> Use this data analysis tool when:
>
> * The user requests statistical insights
> * Raw data needs to be transformed into visualizations
> * Patterns or trends need to be identified in datasets

### 2. Appropriate Permission Levels

Match permission settings to the sensitivity and impact of the tool:

* Use "Approval required" for tools that:
  * Make external communications
  * Modify important data
  * Incur costs or use credits
  * Make decisions with business impact
* Use "Auto run" for tools that:
  * Perform internal data retrieval
  * Execute low-risk, repetitive tasks
  * Enhance the agent's knowledge without external effects

### 3. Workflow Optimization

Consider how tool connections affect the overall workflow:

* **Sequential Processing**: Connect tools in a logical sequence when tasks need to follow a specific order
* **Conditional Branching**: Use agent decision points before tools when different scenarios require different tool selections
* **Feedback Loops**: Create connections from tools back to agents when results need evaluation before proceeding

## Troubleshooting Agent to Tool Connections

### Common Issues and Solutions

1. **Agent not using connected tool**
   * Check that the connection line is properly established
   * Review the communication instructions for clarity
   * Ensure the agent has the necessary context to recognize when to use the tool
2. **Tool execution failing**
   * Verify the tool's configuration and required inputs
   * Check that the agent is passing data in the expected format
   * Review tool logs for specific error messages
3. **Excessive tool usage**
   * Set appropriate "Max auto runs" limits
   * Make instructions more specific about when to use the tool
   * Consider changing to "Approval required" mode temporarily

## Advanced Agent to Tool Configurations

### Chaining Multiple Tools

You can create sophisticated workflows by connecting an agent to multiple tools in sequence:

1. Connect your agent to the first tool
2. Connect the output of the first tool to a second agent or directly to another tool
3. Configure each connection with appropriate settings

This approach allows for complex multi-step processes where each tool performs a specific function in the overall workflow.

### Dynamic Tool Selection

When an agent needs to choose between multiple tools based on context:

1. Connect the agent to each potential tool
2. Use "Let agent decide" mode for all connections
3. Provide clear, distinctive instructions for when to use each tool

The agent will evaluate the situation and select the appropriate tool based on your instructions.

## Related Features

* [Trigger to Agent Configuration](/docs/build/workforces/workforce-features/trigger-to-agent-configuration): Learn how to configure the initial triggers that activate your agents
* [Tool to Tool Configuration](/docs/build/workforces/workforce-features/tool-to-tool-configuration): Discover how to connect tools directly to other tools for automated workflows
* [Condition to Tool Configuration](/docs/build/workforces/workforce-features/condition-to-tool-configuration): Explore how to add conditional logic to your tool executions

## Conclusion

Agent to Tool configuration is a fundamental aspect of building effective AI workforces. By thoughtfully designing these connections, you can create intelligent workflows where agents leverage tools appropriately to accomplish complex tasks. Whether you prefer giving your agents autonomy with "Let agent decide" or creating structured processes with "Forced handover," the configuration options in Workforce Builder provide the flexibility to match your specific business needs.


# Approvals and Escalations
Source: https://relevanceai.com/docs/build/workforces/workforce-features/approvals-and-escalations

Approvals and escalations are essential components of Workforce that ensure your AI agents operate with appropriate oversight and can effectively handle situations requiring human intervention.

## Overview

The approvals and escalations system in Workforce allows you to:

* Configure when and how agents should seek approval before taking actions
* Establish clear escalation paths when agents encounter situations beyond their scope
* Monitor approval requests through the Workforce Task View
* Set up automated alerts for escalations requiring human attention

By implementing proper approval and escalation processes, you can maintain quality control while allowing your AI workforce to operate with appropriate autonomy.

## Setting Up Approval Modes

When connecting agents to tools or other agents in your workforce, you can configure different approval modes that determine how actions are authorized:

### Auto Run

In this mode, the agent will execute actions automatically without requiring approval. This is ideal for:

* Low-risk, routine tasks
* Well-defined processes with predictable outcomes
* Situations where speed is prioritized over verification

To configure auto run:

1. Select the edge connecting an agent to a tool or another agent
2. In the Edge Settings panel, select "Auto run" under Behavior Settings
3. Optionally, set "Max auto runs" to limit the number of automatic executions

### Approval Required

This mode forces the agent to seek explicit approval before executing an action. Use this for:

* High-stakes decisions
* Actions that commit resources or make external communications
* New workflows that need monitoring before being fully automated

To set up approval required:

1. Select the edge connecting an agent to a tool or another agent

2. In the Edge Settings panel, select "Approval required" under Behavior Settings

3. The agent will now draft its intended action and wait for human approval before proceeding

### Let Agent Decide

This balanced approach allows the agent to determine whether approval is needed based on its confidence and context. The agent will:

* Complete tasks autonomously when it has sufficient context and confidence
* Request approval when facing uncertainty or high-risk situations
* Escalate to humans when it needs clarification or additional information

To implement this mode:

1. Select the edge connecting an agent to a tool or another agent
2. In the Edge Settings panel, select "Let agent decide" under Behavior Settings
3. Complete the "When to call this agent" box with natural language instructions explaining the decision logic

## Monitoring Approvals in Workforce Task View

The Workforce Task View serves as your central dashboard for monitoring and managing approval requests:

1. Navigate to the "Task" view to see your workforce
2. Review pending approval requests from your agents
3. Examine the agent's proposed action and supporting reasoning
4. Choose to approve, reject, or provide additional guidance
5. Track the history of approvals and their outcomes

This interface provides complete visibility into your approval workflow, allowing you to maintain oversight while empowering your AI workforce.

## Implementing Escalation Strategies

Escalations occur when an agent encounters a situation it cannot handle independently. Effective escalation strategies ensure these situations are properly addressed:

### Agent-Initiated Escalations

Agents can be configured to recognize their limitations and escalate appropriately:

1. Configure agent settings under Agent Settings > Abilities

2. Define clear criteria for when an agent should escalate issues

3. Specify the escalation path (to specific human roles or other agents)

4. Provide templates for how the agent should format escalation requests

### Escalation Tools

You can connect agents to specific tools designed for escalation:

1. Create dedicated escalation tools (e.g., Slack notifications, email alerts)
2. Connect these tools to your agents in the Workforce Builder
3. Configure the tools to route escalations to the appropriate team members
4. Include relevant context and information in the escalation

### Escalation Monitoring

Track and manage escalations through:

1. The Workforce Task View, which displays all active escalations
2. Notification systems that alert relevant team members
3. Escalation logs that maintain a history of issues and resolutions

## Best Practices for Approvals and Escalations

To maximize the effectiveness of your approval and escalation processes:

### For Approvals:

* Start with more restrictive approval requirements and gradually relax them as confidence in the agent grows
* Use "Max auto runs" settings as guardrails when transitioning to more autonomous operation
* Provide clear criteria for what constitutes an approvable action
* Review approval patterns regularly to identify opportunities for process improvement

### For Escalations:

* Define clear escalation thresholds based on risk, complexity, and agent capabilities
* Create detailed documentation for human responders to handle escalated issues
* Implement a feedback loop where escalation resolutions inform agent improvements
* Establish service level agreements (SLAs) for responding to different types of escalations

## Related Features

* [Edge Settings](https://relevanceai.com/docs/workforce-builder/edge-settings) - Configure the connections between nodes in your workforce, including approval settings

* [Workforce Task View](https://relevanceai.com/docs/workforce-builder/workforce-task-view) - Monitor and manage your workforce's activities, including approvals and escalations

* [Agent Abilities](https://relevanceai.com/docs/build/agents/customise-agent/escalate) - Configure how agents handle situations requiring escalation

## Frequently asked questions (FAQs)

**Q: Can I set different approval requirements for different tools?**\
A: Yes, approval settings are configured on a per-edge basis, allowing you to require approval for sensitive tools while allowing automatic execution for routine ones.

**Q: What happens if an approval request isn't addressed?**\
A: The task will remain in a pending state until approved or rejected. You can configure timeout actions or notifications for approval requests that remain unaddressed for a specified period.

**Q: Can escalations be routed to specific team members?**\
A: Yes, you can configure escalation tools to direct issues to specific individuals or teams based on the nature of the escalation, time of day, or other factors.

**Q: How do I balance autonomy with oversight?**\
A: Start with higher oversight (approval required) for new workflows, then gradually transition to "let agent decide" and eventually "auto run" as you build confidence in the agent's performance for specific tasks.


# Communication
Source: https://relevanceai.com/docs/build/workforces/workforce-features/communication

Communication is a fundamental aspect of Workforce that defines how agents interact with each other within your AI workforce.

## One-way Communication

Currently, Workforce supports one-way communication between agents. This means that when one agent communicates with another, the information flows in a single direction. Bidirectional communication is not yet available in the current version.

### How One-way Communication Works

When you connect two agents in your workforce, the communication follows a unidirectional path:

* The source agent sends information, instructions, or requests to the target agent
* The target agent receives this information and acts upon it
* The target agent cannot automatically send information back to the source agent through the same connection

This one-way communication model creates clear workflows and responsibility chains within your workforce, making it easier to track how information and tasks flow through your system.

## Setting Up Communication Between Agents

When connecting agents in your workforce, you have two connection type options that determine how agents interact with each other:

### AI connection

This option gives the agent autonomy to determine when to communicate with another agent based on natural language instructions.

**Configuration Steps:**

1. Connect two agents in your workforce
2. Select "AI connection" as the connection type
3. Complete the "When to call this agent" field with natural language instructions
4. Configure behavior settings:

**Communication Settings:**
In the "When to call this agent" field, provide clear instructions about when the source agent should delegate to the target agent. For example:

> Call this agent when you need to analyze customer sentiment data or when you need to summarize large volumes of customer feedback. This agent specializes in data analysis and can provide insights that will help with decision-making.

**Behavior Settings:**
Choose one of the following approval modes:

* **Auto run**: The target agent will complete the task without requiring approval
* **Approval required**: The target agent will draft the task but ask for human approval before completing it
* **Let agent decide**: The target agent will complete the task if it has sufficient context and confidence; otherwise, it will ask for human approval or clarification

**Additional Settings:**

* **Max auto runs**: Set a limit on the number of times the agent can auto-run (no limit or a specific number)
* **Task continuation**: Choose between 'always start a new task' or 'always continue with one task'

### Next step

This option creates a mandatory connection where the source agent must pass the task to the target agent.

**Configuration Steps:**

1. Connect two agents in your workforce
2. Select "Next step" as the connection type
3. Configure task continuation:
   * Choose between 'always start a new task' or 'always continue with one task'

With forced handover, the source agent will always transfer control to the target agent when the connection is triggered, regardless of the context or the agent's assessment of the situation.

## Best Practices for Agent Communication

To maximize the effectiveness of agent communication in your workforce:

1. **Define Clear Responsibilities**: Ensure each agent has well-defined roles and responsibilities to avoid confusion about which agent should handle specific tasks.
2. **Use Descriptive Instructions**: When using "AI connection", provide detailed instructions about when and why the source agent should communicate with the target agent.
3. **Create Logical Workflows**: Design your communication paths to follow natural workflows, where information and tasks flow in a logical sequence from one agent to the next.
4. **Balance Autonomy and Control**: Use "AI connection" for situations requiring judgment and "Forced handover" for mandatory steps in a process.
5. **Monitor Communication Patterns**: Regularly review how agents are communicating to identify any bottlenecks or inefficiencies in your workforce.
6. **Implement Appropriate Approval Mechanisms**: Use approval settings to maintain quality control while allowing agents to work autonomously when appropriate.

## Related Features

[Edge Settings](https://relevanceai.com/docs/workforce-builder/edge-settings) - Configure the connections between nodes in your workforce, including communication styles and behavior settings.

[Configurations](https://relevanceai.com/docs/workforce-builder/configurations) - Learn about the different types of connections you can create between nodes in your workforce.

[Workforce Task View](https://relevanceai.com/docs/workforce-builder/task-view) - Monitor your workforce's activities, including communication between agents and approval requests.

## Frequently asked questions (FAQs)

**Q: Can agents communicate bidirectionally?**\
A: Currently, Workforce Builder only supports one-way communication between agents. Bidirectional communication is planned for future updates.

**Q: How do I know if an agent has communicated with another agent?**\
A: You can monitor agent communication in the Workforce Task View, which shows the timeline of activities and interactions between agents.

**Q: Can I set conditions for when an agent should communicate with another agent?**\
A: Yes, when using the "AI connection" communication style, you can provide natural language instructions in the "When to call this agent" field to guide the source agent's decision-making.

**Q: What happens if an agent tries to communicate with another agent but lacks necessary information?**\
A: Depending on your behavior settings, the agent may ask for human approval or clarification before proceeding with the communication.

**Q: Can I create complex communication chains involving multiple agents?**\
A: Yes, you can connect multiple agents in sequence or in parallel to create sophisticated workflows where information passes through several agents.


# Condition to Tool Configuration
Source: https://relevanceai.com/docs/build/workforces/workforce-features/condition-to-tool-configuration

Condition to Tool connections in Workforce allow you to create intelligent workflows that execute tools only when specific conditions are met.

## How Condition to Tool Works

A Condition to Tool connection acts as an IF statement in your workforce flow. When a condition evaluates to true, the connected tool executes; when false, the tool is skipped. This creates branching logic in your workflows, making your AI workforce more adaptable and responsive to different scenarios.

## Setting Up a Condition to Tool Connection

To create a Condition to Tool connection in your workforce:

1. Navigate to Workforce and "Build" in the main navigation
2. Add a condition node to your workspace by dragging it from the node palette
3. Configure your condition (see below for LLM-based vs Rule-based options)
4. Add a tool node to your workspace
5. Connect the condition node to the tool node by clicking and dragging from the condition's output connector to the tool's input connector

## Choosing Between LLM-based and Rule-based Conditions

Workforce offers two types of conditions, each suited for different scenarios:

### LLM-based Conditions

Use natural language to describe complex logic. The AI interprets your description and evaluates whether the condition is met.

**Best for:**

* Complex logic that's easier to express in natural language
* Evaluating unstructured data like text content
* Time-based conditions
* Scenarios requiring interpretation

**Example:** "Go down this path if the email contains an urgent request"

### Rule-based Conditions

Use structured filters to compare specific values with deterministic logic.

**Best for:**

* Precise comparisons of tool outputs
* Numeric threshold checks
* Validating specific field values
* Performance-critical paths

**Example:** Check if `Validate Data / output / status` equals "SUCCESS"

## LLM-based Condition Examples

You can create various types of conditions using simple, natural language instructions:

### Time-Based Conditions

* "Go down this path if today is Monday or Tuesday"
* "Follow this route if the current time is between 9 AM and 5 PM"
* "Use this tool if it's a weekday"

### Data-Based Conditions

* "Go down this path if the customer priority is high"
* "Follow this route if the email contains the word 'urgent'"
* "Use this tool if the sentiment score is greater than 0.7"

### Tool Output Conditions

* "Go down this path if the previous tool completed successfully"
* "Follow this route if the data validation passed"
* "Use this tool if we received a response from the API"

## Rule-based Condition Examples

Rule-based conditions use structured filters for precise, deterministic comparisons:

### Validating Tool Output Status

**Scenario:** Only proceed if a validation tool succeeded

**Configuration:**

* Condition type: Rule-based
* Filter:
  * Target: `Validate Input / output / status`
  * Operator: equals
  * Source: "SUCCESS"

This ensures the next tool only runs when validation explicitly returns "SUCCESS".

### Numeric Threshold Checks

**Scenario:** Route high-value transactions to a special processing tool

**Configuration:**

* Condition type: Rule-based
* Filter:
  * Target: `Calculate Amount / output / total`
  * Operator: >
  * Source: 10000

The tool only executes when the transaction amount exceeds \$10,000.

### Multiple Criteria (AND Logic)

**Scenario:** Process orders that are both complete and paid

**Configuration:**

* Condition type: Rule-based
* Match mode: Match All
* Filter 1:
  * Target: `Check Status / output / order_status`
  * Operator: equals
  * Source: "complete"
* Filter 2:
  * Target: `Check Payment / output / payment_status`
  * Operator: equals
  * Source: "paid"

Both conditions must be true for the tool to execute.

### Multiple Criteria (OR Logic)

**Scenario:** Flag items that need attention (either urgent or high-value)

**Configuration:**

* Condition type: Rule-based
* Match mode: Match Any
* Filter 1:
  * Target: `Get Priority / output / priority`
  * Operator: equals
  * Source: "urgent"
* Filter 2:
  * Target: `Calculate Value / output / amount`
  * Operator: >=
  * Source: 5000

The tool executes if either condition is true.

### Field Existence Check

**Scenario:** Only process records that have an optional field populated

**Configuration:**

* Condition type: Rule-based
* Filter:
  * Target: `Get Data / output / optional_field`
  * Operator: exists

This checks whether the field has any value before proceeding.

### String Pattern Matching

**Scenario:** Route customer IDs with a specific prefix to a specialized tool

**Configuration:**

* Condition type: Rule-based
* Filter:
  * Target: `Get Customer / output / customer_id`
  * Operator: starts with
  * Source: "PREMIUM-"

Only customers with IDs starting with "PREMIUM-" trigger this tool.

## When to Use Each Condition Type

### Use LLM-based Conditions When:

* You need to evaluate natural language or unstructured text
* The logic is complex and easier to express in plain English
* You're working with time-based or contextual conditions
* Exact matching isn't critical
* You want the AI to interpret intent

**Example use case:** Routing customer emails based on content sentiment or urgency

### Use Rule-based Conditions When:

* You need precise, deterministic comparisons
* You're validating tool outputs against known values
* You're working with numeric thresholds
* Performance and reliability are critical
* You want to avoid AI interpretation overhead
* You need to check specific field values or existence

**Example use case:** Validating API responses, checking status codes, or enforcing business rules with specific thresholds

## Best Practices for Condition to Tool Connections

1. **Choose the right condition type**: Use LLM-based for complex interpretation, rule-based for precise comparisons
2. **Keep conditions focused**: Each condition should evaluate a single concept for clarity
3. **Be specific**: Provide clear criteria for when the condition should be true
4. **Test thoroughly**: Verify your conditions work as expected with different inputs
5. **Consider edge cases**: Think about unusual scenarios that might affect your condition
6. **Structure tool outputs appropriately**: For rule-based conditions, ensure tools output flat values rather than nested objects

## Advanced Condition to Tool Patterns

### Chaining Multiple Conditions

You can create sophisticated decision trees by chaining multiple conditions together:

1. Create several condition nodes that evaluate different aspects of your data
2. Connect each condition to the appropriate tool
3. Tools will only execute when their specific conditions are met

### Parallel Condition Evaluation

For more complex scenarios, you can have multiple conditions evaluating the same data simultaneously:

1. Create several condition nodes that check different aspects of the same data
2. Connect each condition to different tools
3. Multiple tools may execute in parallel depending on which conditions are met

### Combining Condition Types

You can use both LLM-based and rule-based conditions in the same workflow:

1. Use rule-based conditions for precise validation checks
2. Use LLM-based conditions for complex interpretation
3. Connect them to different tools based on your workflow needs

## Practical Use Cases

### Email Routing Based on Content

**LLM-based approach:** "Go down this path if the email contains a support request" or "Follow this route if the message mentions billing issues."

**Rule-based approach:** Check if `Classify Email / output / category` equals "support" or "billing"

### Time-Sensitive Workflows

**LLM-based approach:** "Use this tool only during business hours" or "Go down this path if it's the first day of the month."

**Rule-based approach:** Check if `Get Current Hour / output / hour` is  9 and  17

### Data Quality Control

**LLM-based approach:** "Follow this route if all required fields are filled" or "Use this tool if the data passes validation."

**Rule-based approach:** Check if `Validate Data / output / status` equals "valid" and `Check Required Fields / output / all_present` equals true

### Approval Workflows

**LLM-based approach:** "Go down this path if the request has been approved" or "Use this tool if the manager has given permission."

**Rule-based approach:** Check if `Check Approval / output / approved` equals true

## Troubleshooting Condition to Tool Connections

If your condition to tool connections aren't working as expected:

### For LLM-based Conditions:

1. **Review your condition language**: Make sure your natural language condition is clear and specific
2. **Test with simpler conditions**: Try a very basic condition to verify the connection works
3. **Refine your language**: If needed, make your condition more explicit or detailed

### For Rule-based Conditions:

1. **Check the target path**: Verify it points to the correct field from the upstream node
2. **Verify the operator**: Ensure you're using the right comparison type (e.g., numeric operators for numbers)
3. **Check the source value**: Make sure it matches the expected data type and format
4. **Review match mode**: Confirm whether you need "Match All" (AND) or "Match Any" (OR) logic
5. **Test with simple filters**: Start with a single filter to isolate issues

### General Troubleshooting:

1. **Check the flow**: Confirm the connections between nodes are correctly established
2. **Monitor execution**: Watch how the workflow behaves when the condition is evaluated
3. **Review upstream outputs**: Verify that the data you're evaluating actually exists and has the expected format

## Related Features

* [Add Conditions](/docs/build/workforces/build-an-ai-workforce/add-conditions): Complete guide to configuring both LLM-based and rule-based conditions
* [Agent to Tool connections](/docs/build/workforces/workforce-features/agent-to-tool-configuration): Direct connections from agents to tools without conditional logic
* [Tool to Tool connections](/docs/build/workforces/workforce-features/tool-to-tool-configuration): Sequential execution of tools in a workflow
* [Edge Settings](/docs/build/workforces/build-an-ai-workforce/edge-settings): Configure how nodes connect in your workforce

By using Condition to Tool connections with both LLM-based and rule-based conditions in the Workforce Builder, you can create intelligent, adaptive workflows that respond dynamically to changing circumstances, making your AI workforce more efficient and effective.


# Tool to Tool Configuration
Source: https://relevanceai.com/docs/build/workforces/workforce-features/tool-to-tool-configuration

Tool to Tool configuration is a powerful feature in Workforce that allows you to connect tools directly to other tools, creating automated workflows without requiring agent intervention.

## How Tool to Tool Connections Work

When you connect tools directly to each other in Workforce, you're establishing a direct pathway for data to flow from one tool to the next. This creates a streamlined process where:

1. The first tool completes its operation
2. Its output is automatically passed to the connected tool
3. The second tool processes this data without requiring human or agent approval (unless configured otherwise)

This approach is particularly valuable for creating multi-step automations that handle routine processes efficiently.

## Setting Up Tool to Tool Connections

### Prerequisites

* At least two tools added to your Workforce Builder canvas
* Understanding of the input/output requirements for each tool

### Step-by-Step Configuration

1. **Add Tools to Your Workspace**: Drag the tools you want to connect from your project onto the Workforce Builder canvas.

2. **Create the Connection**:
   * Click on the first tool (source)
   * Look for the connection point (usually appears as a dot or node on the edge of the tool)
   * Click and drag to create a line to the second tool (destination)
   * Release to establish the connection

3. **Configure Connection Settings**:
   * After creating the connection, click on the connecting line to access the edge settings
   * Choose between "Auto run" or "Approval required" modes:
     * **Auto run**: The second tool will automatically execute when it receives data from the first tool

     * **Approval required**: The system will wait for human approval before the second tool processes the data

4. **Map Data Fields** (if applicable):
   * Some tools may require you to specify how data from the first tool maps to the input fields of the second tool
   * Use the mapping interface to connect output fields from the first tool to input fields of the second tool

5. **Save Your Configuration**: Click "Save changes" to apply your Tool to Tool configuration

## Use Cases for Tool to Tool Configuration

### Email Processing Workflow

Connect an "Email Receiver" tool to an "Email Classifier" tool, then to either a "Customer Support" tool or "Sales Lead" tool based on classification results.

### Data Enrichment Chain

Link a "Data Extraction" tool to a "Data Enrichment" tool, followed by a "Database Update" tool to automatically enhance and store information.

### Document Processing Pipeline

Connect a "Document Analyzer" tool to a "Summary Generator" tool, then to a "Notification" tool to alert team members about important documents.

### Conditional Workflows

When combined with condition nodes, Tool to Tool connections can create branching workflows that take different paths based on specific criteria.

## Best Practices

1. **Start Simple**: Begin with straightforward two-tool connections before building more complex chains.

2. **Test Thoroughly**: Always test your Tool to Tool configurations with sample data before deploying them in production.

3. **Consider Approval Points**: For critical operations, insert approval requirements at key points in the workflow.

4. **Document Your Flows**: Create documentation for your Tool to Tool configurations to help team members understand the automated processes.

5. **Monitor Performance**: Regularly check the execution logs to ensure your Tool to Tool connections are functioning as expected.

6. **Error Handling**: Plan for potential failures by considering what should happen if a tool in the chain encounters an error.

## Limitations and Considerations

* **Data Compatibility**: Ensure that the output format of the first tool is compatible with the input requirements of the second tool.

* **Execution Time**: Be aware that complex Tool to Tool chains may take longer to execute, especially if tools perform resource-intensive operations.

* **Debugging Challenges**: Troubleshooting issues in Tool to Tool connections can be more complex than single-tool setups.

* **Versioning Impact**: Updating a tool in a chain may affect downstream tools if the output format changes.

## Related Features

Tool to Tool configuration works seamlessly with other Workforce Builder features:

* **Conditions**: Add conditional logic to determine whether a tool connection should be followed
* **Agent to Tool**: Combine agent decision-making with automated tool chains
* **Triggers**: Initiate Tool to Tool workflows based on external events

## Conclusion

Tool to Tool configuration in the Workforce Builder represents a significant advancement in workflow automation capabilities. By connecting tools directly to each other, you can create sophisticated process chains that handle routine tasks efficiently while maintaining the flexibility to incorporate human oversight where needed. This feature empowers you to build more autonomous, intelligent workflows that save time and reduce the need for manual intervention.


# Trigger to Agent Configuration
Source: https://relevanceai.com/docs/build/workforces/workforce-features/trigger-to-agent-configuration

Connecting triggers to agents is a fundamental aspect of building effective AI workforces in Relevance AI.

## Overview

Trigger to Agent configurations define the relationship between a trigger event and the agent that responds to it. This connection is the starting point of any workflow in Workforce, allowing you to specify exactly how your agents are activated and what information they receive.

## Types of Triggers

When connecting triggers to agents, you can utilize several trigger types:

* **Manual Triggers**: Allow you to manually initiate an agent through the Task View interface
* **Recurring Schedule Triggers**: Activate agents at predetermined times or intervals using simple schedules (daily, weekly, monthly) or advanced cron expressions for precise timing control
* **Integration Triggers**: Connect external services (like email, Slack, or CRM systems) to automatically activate agents when specific events occur

<Note>
  For detailed information on configuring recurring schedules and cron expressions, see [Add Triggers](/docs/build/workforces/build-an-ai-workforce/add-triggers).
</Note>

## Setting Up a Trigger to Agent Connection

### Basic Setup

1. Open your workforce in the Build section
2. Locate the trigger node you want to connect (or add a new one by dragging it from the sidebar)
3. Click and drag from the trigger node's output point to connect it to your desired agent
4. Configure the connection settings in the panel that appears

### Configuration Options

When establishing a Trigger to Agent connection, you'll need to configure these important settings:

#### Communication Settings

In the connection settings panel, you can specify:

* **Message Format**: Define how the trigger information is formatted before being sent to the agent
* **Context Passing**: Choose what contextual information from the trigger should be included
* **Variable Mapping**: Map specific trigger data fields to variables the agent can use

#### Behavior Settings

Control how the agent responds to the trigger:

* **Auto Run**: Enable the agent to automatically process the trigger without human approval
* **Approval Required**: Require human approval before the agent acts on the trigger
* **Let Agent Decide**: Allow the agent to determine whether it needs approval based on confidence level

#### Task Management

Configure how tasks are created and managed:

* **Always Start New Task**: Each trigger creates a new, separate task
* **Continue with One Task**: Multiple triggers can feed into a single ongoing task
* **Max Auto Runs**: Set limits on how many times an agent can auto-run to prevent infinite loops

## Best Practices

### Effective Trigger to Agent Design

* **Clear Trigger Definition**: Ensure your triggers have well-defined conditions to avoid unintended activations
* **Appropriate Agent Selection**: Match triggers with agents that have the right skills and tools for the triggered task
* **Contextual Information**: Configure triggers to provide sufficient context for the agent to understand the task
* **Error Handling**: Include fallback paths for when triggers fail or agents cannot process the request
* **Schedule Timing**: For recurring schedules, ensure your agent can complete its work before the next scheduled run to avoid overlap

### Common Patterns

* **Filtering Triggers**: Use conditions before triggers to filter out unnecessary activations
* **Multi-Agent Dispatch**: Connect a trigger to a dispatcher agent that routes tasks to specialized agents
* **Escalation Paths**: Configure triggers to activate different agents based on priority or complexity

## Examples

### Example 1: Email Response Workflow

Here's how a typical email response workflow might be configured:

1. Create an email integration trigger that activates when new emails arrive
2. Connect this trigger to a triage agent
3. Configure the connection to:
   * Pass the email subject, body, and sender information
   * Allow auto-run for initial triage
   * Create a new task for each email
4. The triage agent can then analyze the email and route it to the appropriate specialized agent

### Example 2: Daily Report Generation

Here's how a recurring schedule workflow might be configured:

1. Create a recurring schedule trigger set to run every weekday at 9:00 AM (using cron: `0 9 * * MON-FRI *`)
2. Connect this trigger to a reporting agent
3. Configure the connection to:
   * Always start a new task for each scheduled run
   * Enable auto-run since this is a routine task
   * Pass the current date as context to the agent
4. The reporting agent can then gather data from the previous day and generate a summary report

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Why is my trigger not activating my agent?">
    Verify the trigger conditions and integration settings. For recurring schedules, check that the cron expression or schedule configuration is correct. Ensure the trigger is properly connected to the agent node.
  </Accordion>

  <Accordion title="Why is my agent not receiving context from the trigger?">
    Check the connection configuration to ensure proper data mapping. Review the Communication Settings in the connection panel and verify that variable mapping is set up correctly.
  </Accordion>

  <Accordion title="Why am I getting multiple unintended activations?">
    Review trigger conditions and consider adding filters. For recurring schedules, verify you don't have overlapping schedules. Check the Task Management settings to ensure "Always Start New Task" is configured appropriately.
  </Accordion>

  <Accordion title="How do I test my trigger configuration before enabling auto-run?">
    Use the Task View to monitor trigger activations and agent responses. Test triggers manually first, check connection logs to identify any data transmission issues, and verify the configuration with a few test runs before enabling auto-run for production use.
  </Accordion>

  <Accordion title="Can a recurring schedule trigger activate multiple agents at once?">
    Yes, you can connect a single recurring schedule trigger to multiple agents. Each agent will receive the trigger activation and can be configured independently with different connection settings.
  </Accordion>
</AccordionGroup>

## Related Features

The Trigger to Agent configuration works seamlessly with other Workforce Builder components:

* [Agent to Agent Configuration](/docs/build/workforces/workforce-features/agent-to-agent-configuration): Connect agents to create multi-step workflows
* [Agent to Tool Configuration](/docs/build/workforces/workforce-features/agent-to-tool-configuration): Enable agents to use tools to complete tasks
* [Workforce Task View](/docs/build/workforces/workforce-features/workforce-task-view): Monitor and manage tasks initiated by triggers


# Workforce Task View
Source: https://relevanceai.com/docs/build/workforces/workforce-features/workforce-task-view

The Workforce Task View provides a centralized interface for monitoring, managing, and interacting with your AI workforce.

## Accessing the Workforce Task View

To access the Workforce Task View:

1. Navigate to the "Task" section in your selected Workforce
2. The Task View interface will display all current and recent activities for your selected workforce

## Key Features

### Real-Time Task Monitoring

Track the status and progress of all tasks being handled by your workforce in real time. The Task View provides a comprehensive overview of:

* Active tasks currently being processed
* Completed tasks with timestamps and outcomes
* Tasks awaiting approval or human intervention
* Failed tasks that require attention

### Manual Trigger Activation

Initiate workflows directly from the Task View interface:

1. Locate the manual trigger you want to activate
2. Enter your message or prompt to start the workflow
3. Submit the trigger to activate the associated agent or process
4. Monitor the task's progress as it moves through your workforce

This functionality allows you to test workflows, handle one-off requests, or manually start scheduled processes when needed.

### Approval Management

The Task View serves as your central hub for managing approvals within your workforce:

1. Receive notifications when agents require approval to proceed
2. Review the agent's proposed actions or responses
3. Approve, reject, or modify the agent's work
4. Provide additional context or instructions if needed

This approval system ensures quality control while allowing your workforce to operate autonomously when appropriate.

### Conversation History

Access complete conversation histories and task logs for comprehensive tracking:

* View the full thread of interactions between agents and users
* Examine decision points and tool usage throughout a task
* Reference previous tasks for context or troubleshooting
* Export conversation logs for reporting or analysis

<Note>
  [Visual Data Masking](/docs/build/agents/build-your-agent/agent-settings/visual-data-masking) can be enabled to protect sensitive information like email addresses, phone numbers, and credit card numbers in conversation histories when reviewing tasks with your team or during screen sharing.
</Note>

### Task Timeline

The Task View provides a chronological timeline of all activities, giving you visibility into:

* When tasks were initiated
* How long each step in the process took
* Which agents or tools were involved at each stage
* When human intervention occurred
* Final resolution and completion times

## Working with the Task View

### Monitoring Multiple Tasks

The Task View allows you to monitor multiple concurrent tasks across your workforce:

1. Tasks are organized chronologically with the most recent at the top
2. Use filters to focus on specific task types, agents, or statuses
3. Expand individual tasks to see detailed information
4. Toggle between different workforces if you have multiple configured

### Handling Escalations

When agents escalate issues requiring human attention:

1. You'll receive a notification in the Task View
2. The escalated task will be highlighted for visibility
3. Review the agent's reason for escalation and the current task state
4. Provide guidance, additional information, or take over the task as needed
5. Once resolved, the agent can continue processing or the task can be marked complete

### Approval Workflows

For tasks configured to require human approval:

1. The agent will complete its work and submit it for review
2. The task will appear in your approval queue within the Task View
3. Review the agent's work, including any generated content or proposed actions
4. Approve the work to allow the agent to proceed or continue to the next step
5. Reject or request modifications if changes are needed
6. Provide feedback to help improve future performance

## Best Practices

### Organizing Your Task View

* Use clear naming conventions for your triggers and agents to easily identify tasks
* Configure appropriate approval thresholds to balance autonomy and oversight
* Set up notification preferences to stay informed of critical tasks
* Archive completed tasks regularly to maintain a clean interface

### Effective Monitoring

* Check the Task View regularly for items requiring attention
* Use filters to focus on high-priority tasks or specific workflows
* Review completed tasks periodically to identify optimization opportunities
* Monitor task completion times to identify bottlenecks or inefficiencies

### Troubleshooting

If you encounter issues with tasks in the Task View:

1. Check the task's history to identify where the process stalled
2. Review any error messages or agent notes
3. Verify that all connected tools and resources are functioning properly
4. Test the workflow with a manual trigger to reproduce the issue
5. Adjust agent settings or tool configurations as needed

## Related Features

* [Edge Settings](https://relevanceai.com/docs/build/agents/customise-agent/flowbuilder): Configure how agents communicate and interact with each other in your workforce
* [Approvals and Escalations](https://relevanceai.com/docs/build/agents/customise-agent/escalate): Set up detailed approval workflows and escalation paths for your agents
* [Workforce Builder](https://relevanceai.com/docs/build/agents/customise-agent/subagents): Design and configure your multi-agent workforce with visual tools
* [Visual Data Masking](/docs/build/agents/build-your-agent/agent-settings/visual-data-masking): Protect sensitive information in conversation histories


# Analytics
Source: https://relevanceai.com/docs/enterprise/analytics

Monitor your AI workforce performance, costs, and usage in real-time

<Info>Analytics is an Enterprise feature. Access it from your Relevance AI dashboard under **Account**.</Info>

Analytics gives you complete visibility into how your AI workforce is performing. Track tasks, monitor costs, spot performance issues, and optimize your agents—all from a single dashboard.

<Frame>
  <img alt="Analytics Dashboard showing key metrics, task trends, action trends, and credits over time" />
</Frame>

## Getting Started

<Steps>
  <Step title="Open Analytics">
    Navigate to **Account** in your Relevance AI dashboard and click on **Analytics**
  </Step>

  <Step title="Set Your Filters">
    Select your date range and choose which agents to analyze
  </Step>

  <Step title="Explore Your Data">
    Scroll through the dashboard to explore workforce, agent, and action breakdowns
  </Step>
</Steps>

<Tip>**Pro tip:** Scroll to the bottom of any breakdown table to see summary stats like "Avg: 0.36% error rate" and "Avg: 95 credits/task"—these give you instant performance benchmarks without reviewing individual rows.</Tip>

***

## Key Metrics at a Glance

Your Analytics dashboard displays four critical metrics with trend comparisons:

<CardGroup>
  <Card title="Tasks" icon="list-check">
    Total tasks completed by your agents with percentage change trends
  </Card>

  <Card title="Tasks to Review" icon="eye">
    Tasks awaiting human approval or review
  </Card>

  <Card title="Actions Used" icon="bolt">
    Number of tool executions and API calls made by your agents
  </Card>

  <Card title="Credits Used" icon="coins">
    Total Relevance AI credits consumed across all activities
  </Card>
</CardGroup>

Each metric shows percentage change from the previous period. For example, "+45.78%" on Tasks means you completed 45% more tasks than the previous period—use this to track growth or identify sudden increases in workload.

<Note>**Filter and Focus**: Click the date range picker (top right) to select custom time periods. Use "All agents" dropdown to view individual agent performance. Compare October vs September to track month-over-month growth, or select a specific agent to see only their metrics across all charts and tables.</Note>

***

## Dashboard Sections

<AccordionGroup>
  <Accordion title="Task and Action Trends" icon="chart-line">
    Interactive charts show your daily task volume and action usage over time.

    **Quickly identify:**

    * **Peak usage days** - Spot which days have the highest bars to understand demand patterns
    * **Unusual activity patterns** - Look for unexpected spikes or drops that might indicate issues
    * **Growth trends** - See if bars are generally increasing (growth) or decreasing (reduced usage)

    **How to use it:** Hover over any bar to see exact numbers for that day. Use the date range picker to zoom into specific weeks or months. Compare task volume with credit usage charts to understand cost patterns.
  </Accordion>

  <Accordion title="Credits" icon="coins">
    A dedicated chart showing your daily credit consumption over time.

    **Quickly identify:**

    * **Expensive days** - See which days consumed the most credits (highest bars)
    * **Cost anomalies** - Spot unexpected credit spikes that might indicate workflows using excessive credits
    * **Spending patterns** - Understand if your credit usage is steady, growing, or fluctuating

    **How to use it:** Hover over bars to see exact credit amounts for each day. Compare the Credits chart with the Tasks chart above it—if tasks are flat but credits spike, you have workflows becoming more expensive. Use this to budget and forecast monthly credit needs based on historical patterns.
  </Accordion>

  <Accordion title="Workforce Breakdown" icon="users-gear">
    See performance metrics for each of your AI workforces:

    | Metric                | What It Shows                                    |
    | --------------------- | ------------------------------------------------ |
    | **Workforce Tasks**   | Tasks handled by the workforce                   |
    | **Agent Tasks**       | Individual agent activities within the workforce |
    | **Error Rate**        | Percentage of failed tasks                       |
    | **Actions & Credits** | Resource consumption per workforce               |
    | **Credits/Task**      | Cost efficiency metric                           |

    **How to use it:** Compare Credits/Task across workforces to find your most cost-effective teams. Click into any workforce row to see more details. Check the summary stats at the bottom (e.g., "Avg: 95 credits/task") to gauge overall performance.
  </Accordion>

  <Accordion title="Agent Breakdown" icon="robot">
    Drill down to individual agent performance:

    | Metric           | What It Tells You                                            |
    | ---------------- | ------------------------------------------------------------ |
    | **Tasks**        | Volume of work handled                                       |
    | **Error Rate**   | Reliability and success rate                                 |
    | **Actions/Task** | Complexity indicator (more actions = more complex workflows) |
    | **Credits/Task** | Cost per task for efficiency comparison                      |

    **How to use it:** Sort by Credits/Task to find your most efficient agents (e.g., agent with 5 credits/task vs 3,500 credits/task). Use "Show 35 more agents" to view your full agent list. Investigate agents with high error rates or unusually high Actions/Task ratios.
  </Accordion>

  <Accordion title="Action Breakdown" icon="bolt">
    See which specific tools and integrations are being used:

    | Metric           | Purpose                                   |
    | ---------------- | ----------------------------------------- |
    | **Executions**   | How often each action is called           |
    | **Error Rate**   | Action reliability and integration health |
    | **Tasks**        | Number of tasks using this action         |
    | **Credits/Exec** | Cost per execution                        |

    **How to use it:** Check which actions have high execution counts to see your most-used tools. Actions with 0.00% error rates are working reliably. If an action shows error rates above 0%, investigate that integration. Sort by Credits/Exec to find expensive API calls or tool executions.
  </Accordion>
</AccordionGroup>

***

## Concurrency usage

The Analytics page includes a concurrency usage chart. The chart shows organization-wide concurrency data — not data specific to a single agent — so you can correlate agent activity with overall concurrency patterns and see which agents are most active during peak periods.

<Note>
  For information about concurrency limits by subscription tier and how to monitor usage from the Plan & Billing page, see [Plans and credits](/docs/admin/subscriptions/plans#monitoring-concurrency-usage).
</Note>

***

## Why Use Analytics?

<CardGroup>
  <Card title="Cost Control" icon="chart-line-down">
    Monitor the Credits Used metric and watch for percentage spikes. Compare Credits/Task across agents to find expensive workflows and optimize them
  </Card>

  <Card title="Performance Optimization" icon="gauge-high">
    Sort agents by Credits/Task in the Agent Breakdown table. Agents with lower credits per task are more efficient—study their setup to replicate success
  </Card>

  <Card title="Error Monitoring" icon="triangle-exclamation">
    Check the Error % column in each breakdown table. Sort by error rate to quickly identify which agents or actions are failing and need attention
  </Card>

  <Card title="Capacity Planning" icon="arrow-trend-up">
    Use the date range filter to compare different periods. Watch for consistent upward trends in the task chart and high percentage changes to identify when you need more agents
  </Card>

  <Card title="ROI Visibility" icon="chart-pie">
    Point stakeholders to the Tasks metric showing volume completed and Credits/Task showing cost efficiency. The summary stats at table bottoms provide quick proof of value
  </Card>

  <Card title="Troubleshooting" icon="magnifying-glass-chart">
    Look for unusual spikes in the charts, high error rates in tables, or agents with abnormally high Credits/Task ratios to catch issues early
  </Card>
</CardGroup>


# Asset-level authentication controls
Source: https://relevanceai.com/docs/enterprise/asset-controls

Set authentication controls at an asset level

<Warning>
  Asset-level authentication controls are gradually being rolled out to our Enterprise customers. If you have an Enterprise subscription with Relevance AI and do not have access to this feature yet, please reach out to your sales representative to share your interest in this feature.

  You will not be able to access this feature if you are not on an Enterprise subscription.
</Warning>

Asset-level authentication controls allow users to decide whether they wish to share their OAuth accounts with other users on their assets or not. Users can set their OAuth accounts to be shared (accessible and used by other users of the asset), or hidden from other users.

## How do I set authentication controls on Agents?

<div>
  <iframe title="Set authentication controls on an Agent" />
</div>

1. Head to the Tools section of your Agents
2. For Tools that have OAuth accounts as Inputs, click on the OAuth account you want to configure authentication controls on
3. To make an account a Shared Account, hover over the account and click the globe button
4. To hide the account from other users, hover over the account and click the globe button

## How do I set authentication controls on Tools?

<div>
  <iframe title="Set authentication controls on a Tool" />
</div>

1. Open the Tool you want to configure authentication controls on, and head to the Build screen
2. Select the OAuth account you want to configure
3. To make an account a Shared Account, hover over the account and click the globe button
4. To hide the account from other users, hover over the account and click the globe button

***

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can I access asset-level authentication controls without upgrading to Enterprise?">No. This feature is available for Enterprise subscriptions only.</Accordion>
  <Accordion title="I'm on an Enterprise subscription but do not have access to this feature yet. How do I get access?">This feature is gradually being rolled out to Enterprise customers. Please reach out to your sales representative to express your interest in receiving this feature.</Accordion>
</AccordionGroup>


# Data Retention
Source: https://relevanceai.com/docs/enterprise/data-retention

Control how long sensitive data is stored in your Relevance AI organization

<Warning>
  Data Retention is an Enterprise feature that must be enabled by the Relevance AI team and is not available on other plans. Contact your sales representative or [reach out to support](/docs/get-started/support) to request enablement.
</Warning>

Data Retention is an Enterprise feature that allows organizations to automatically manage the lifecycle of sensitive data stored in Relevance AI. By configuring retention policies, you can ensure that data is automatically deleted after a specified period, supporting compliance requirements, security best practices, and storage optimization.

This feature gives organization admins control over how long different types of data—including tool runs, agent conversations, workforce tasks, and asset versions—are retained before being permanently deleted.

<Frame>
  <img alt="Data Retention configuration interface showing enable toggle and retention period settings" />
</Frame>

## Why use Data Retention?

<CardGroup>
  <Card title="Compliance & governance" icon="scale-balanced">
    Meet regulatory requirements for data retention limits (GDPR, industry-specific regulations) and demonstrate compliance with data minimization principles
  </Card>

  <Card title="Security & risk management" icon="shield-check">
    Reduce your organization's data exposure by automatically removing old, potentially sensitive information that's no longer needed
  </Card>

  <Card title="Storage optimization" icon="database">
    Manage storage costs and platform performance by automatically cleaning up historical data that's no longer actively used
  </Card>

  <Card title="Operational control" icon="gears">
    Establish consistent data lifecycle policies across your entire organization without manual intervention
  </Card>
</CardGroup>

***

## What data is covered?

Data Retention policies apply to four main categories of data in Relevance AI. Each category has specific conditions that determine when data is eligible for deletion.

<AccordionGroup>
  <Accordion title="Tool runs & sync items" icon="wrench">
    **What's included:**

    * Individual tool execution records
    * Sync items from integrated data sources
    * Input and output data from tool runs

    **Deletion condition:**

    * The run or sync item is older than your configured retention period

    **Example:** If you set a 90-day retention period, any tool run older than 90 days will be automatically deleted.
  </Accordion>

  <Accordion title="Agent conversations" icon="messages">
    **What's included:**

    * Complete conversation threads with agents
    * All messages within conversations
    * Conversation metadata and context

    **Deletion conditions (ALL must be met):**

    * The conversation is older than the retention period
    * All messages in the conversation are older than the retention period

    **Example:** A conversation started 100 days ago with a message added 50 days ago will NOT be deleted if your retention period is 90 days. The conversation is only deleted when both the conversation AND all its messages exceed the retention period.
  </Accordion>

  <Accordion title="Workforce tasks" icon="list-check">
    **What's included:**

    * Workforce task records
    * All conversations within workforce tasks
    * Task execution history and outputs

    **Deletion conditions (ALL must be met):**

    * The workforce task is older than the retention period
    * All conversations in the workforce task are older than the retention period (following the conversation conditions above)

    **Example:** A workforce task from 100 days ago with an associated conversation that has a recent message will NOT be deleted until both the task and all its conversations (including all messages) exceed the retention period.
  </Accordion>

  <Accordion title="Agent, Tool & Workforce versions" icon="code-branch">
    **What's included:**

    * Historical versions of agents
    * Historical versions of tools
    * Historical versions of workforces

    **Deletion conditions (ALL must be met):**

    * The version is older than the retention period
    * The version is NOT currently the active version
    * The version is NOT currently the draft version

    **Protection:** Your current active and draft versions are always protected from deletion, regardless of age. Only old, inactive versions are eligible for deletion.

    **Example:** If you have 10 versions of an agent and the retention period is 60 days, only versions older than 60 days that are neither active nor draft will be deleted. Your current working versions remain safe.
  </Accordion>
</AccordionGroup>

<Note>
  Data Retention policies do not affect knowledge bases, integrations, API keys, or organization/project settings. These remain under your manual control.
</Note>

<Warning>
  The Data Retention feature applies only to data stored and managed by Relevance AI, not to data processed or stored by our subprocessors. For questions about subprocessor or integration retention policies, contact your Account Manager or [reach out to support](/docs/get-started/support).
</Warning>

***

## How to configure Data Retention

<Info>
  To configure Data Retention, you must have an Enterprise subscription, the feature must be enabled by Relevance AI, and you must be an organization admin or owner.
</Info>

<Steps>
  <Step title="Request feature enablement" icon="envelope">
    Contact the Relevance AI team to enable Data Retention for your organization. This is not a self-serve feature and must be activated by our team.

    Reach out to your sales representative or [contact support](/docs/get-started/support) to request enablement.
  </Step>

  <Step title="Navigate to data management" icon="gear">
    Once enabled, click **"Settings"** in the sidebar, then select **"Manage organization"** to access organization-level settings.

    Navigate to the **"Data management"** section where you'll find Data Retention controls.
  </Step>

  <Step title="Create configuration" icon="plus">
    Click **"Create configuration"** to set up your organization's data retention policy.

    This opens the configuration interface where you can define your retention settings.
  </Step>

  <Step title="Enable and set retention period" icon="clock">
    Toggle the **"Enable"** slider to activate data retention for your organization.

    Set your desired **retention period in days**. This determines how long data is kept before being automatically deleted.

    **Common retention periods:**

    * 30 days - Short-term retention for highly sensitive data
    * 90 days - Standard retention for most use cases
    * 180 days - Extended retention for compliance requirements
    * 365 days - Annual retention for long-term records
  </Step>

  <Step title="Save configuration" icon="floppy-disk">
    Click **"Save"** to apply your data retention policy.

    The policy takes effect immediately and will begin automatically deleting data that exceeds the retention period.
  </Step>
</Steps>

<Warning>
  Data deletion is permanent and cannot be undone. Once data is deleted through retention policies, it cannot be recovered, so ensure your retention period aligns with your organization's compliance, legal, and operational requirements before enabling this feature.
</Warning>

***

## Use cases & compliance considerations

<Tabs>
  <Tab title="Compliance requirements" icon="scale-balanced">
    ### Meeting regulatory standards

    Data Retention helps organizations comply with various regulatory frameworks that mandate data retention limits:

    **GDPR (General Data Protection Regulation)**

    * Supports data minimization principles
    * Ensures personal data isn't kept longer than necessary
    * Demonstrates compliance with storage limitation requirements

    **Industry-Specific Regulations**

    * Financial services: SOX, PCI-DSS compliance
    * Legal: eDiscovery and litigation hold considerations
    * Government: FISMA and agency-specific requirements

    <Tip>
      Consult with your legal and compliance teams to determine the appropriate retention period for your organization's specific regulatory requirements.
    </Tip>
  </Tab>

  <Tab title="Security best practices" icon="shield-check">
    ### Reducing data exposure

    Implementing data retention policies reduces your organization's security risk profile:

    **Minimize Attack Surface**

    * Less historical data means less potential exposure in a breach
    * Reduces the volume of sensitive information stored
    * Limits the scope of data that could be compromised

    **Data Hygiene**

    * Automatically removes outdated information
    * Prevents accumulation of unnecessary sensitive data
    * Supports "need to know" access principles

    **Incident Response**

    * Smaller data footprint simplifies breach investigations
    * Reduces scope of potential data breach notifications
    * Limits liability exposure

    <Note>
      Balance security benefits with operational needs. Ensure your retention period allows sufficient time for auditing, troubleshooting, and business continuity.
    </Note>
  </Tab>

  <Tab title="Operational efficiency" icon="gauge-high">
    ### Optimizing platform performance

    Data Retention supports operational efficiency and cost management:

    **Storage Optimization**

    * Automatically manages storage consumption
    * Prevents unlimited data accumulation
    * Supports predictable storage costs

    **Performance Benefits**

    * Smaller datasets improve query performance
    * Faster search and retrieval operations
    * Better platform responsiveness

    **Simplified Management**

    * Automated cleanup eliminates manual deletion tasks
    * Consistent policies across the organization
    * Reduces administrative overhead

    **Cost Control**

    * Predictable storage requirements
    * Prevents runaway data growth
    * Aligns storage costs with business value
  </Tab>

  <Tab title="Planning your policy" icon="clipboard-list">
    ### Determining the right retention period

    Consider these factors when setting your retention period:

    **Legal & Compliance Requirements**

    * What are your industry's minimum retention requirements?
    * Do you have specific regulatory obligations?
    * What are your litigation hold considerations?

    **Business Needs**

    * How long do teams need access to historical data?
    * What's your typical audit or review timeframe?
    * Do you need data for trend analysis or reporting?

    **Security Posture**

    * What's your organization's risk tolerance?
    * How sensitive is the data you're processing?
    * What are your data breach notification obligations?

    **Operational Considerations**

    * How often do you reference historical data?
    * What's your troubleshooting and debugging timeframe?
    * Do you need data for training or quality assurance?

    <Tip>
      Start with a longer retention period (e.g., 180 days) and adjust based on actual usage patterns and requirements. It's easier to shorten retention periods than to recover deleted data.
    </Tip>
  </Tab>
</Tabs>

***

## Important considerations

<AccordionGroup>
  <Accordion title="Data deletion is permanent" icon="triangle-exclamation">
    Once data is deleted through retention policies, it **cannot be recovered**. There is no "recycle bin" or backup of deleted data.

    **Before enabling Data Retention:**

    * Review your organization's data retention requirements
    * Consult with legal, compliance, and security teams
    * Consider exporting critical data for long-term archival
    * Test with a longer retention period initially
  </Accordion>

  <Accordion title="Organization-wide impact" icon="building">
    Data Retention policies apply to your **entire organization**, not individual projects or teams.

    All projects within your organization will follow the same retention period. If different teams have different requirements, you'll need to coordinate on a retention period that satisfies all stakeholders.
  </Accordion>

  <Accordion title="Active versions are protected" icon="shield">
    Your current **active and draft versions** of agents, tools, and workforces are always protected from deletion, regardless of age.

    Only old, inactive versions that exceed the retention period are eligible for deletion. This ensures your production assets remain safe while cleaning up historical versions.
  </Accordion>

  <Accordion title="No automatic notifications" icon="bell">
    The system does not send notifications before deleting data. Deletion happens automatically based on your configured retention period.

    Ensure your teams understand the retention policy and plan accordingly for any data they need to preserve long-term.
  </Accordion>

  <Accordion title="Export data before deletion" icon="download">
    If you need to preserve specific data beyond the retention period, export it before it's deleted.

    Use Relevance AI's export features to download conversation logs, tool run results, or other data you need to retain for longer periods.
  </Accordion>

  <Accordion title="Audit and compliance records" icon="file-lines">
    Consider your audit and compliance requirements when setting retention periods. Some regulations require specific retention periods for audit trails.

    If you use [S3 Event Streaming](/docs/enterprise/streaming-events), those audit logs and traces are delivered to your own S3 bucket and are not affected by Data Retention policies in Relevance AI. Their retention is governed by your bucket's own lifecycle policy. On the Relevance side, telemetry only passes through a temporary staging bucket that is automatically purged after 30 days.
  </Accordion>
</AccordionGroup>

***

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can I access Data Retention without upgrading to Enterprise?">
    No. Data Retention is available exclusively for Enterprise subscriptions. Additionally, the feature must be enabled by the Relevance AI team—it's not automatically available even for Enterprise customers.
  </Accordion>

  <Accordion title="Can I set different retention periods for different projects?">
    No. Data Retention policies are set at the organization level and apply uniformly across all projects within your organization. All teams must follow the same retention period.
  </Accordion>

  <Accordion title="What happens to data that's currently older than my retention period?">
    When you first enable Data Retention, any existing data that already exceeds your configured retention period will be deleted according to the deletion conditions for each data type. This cleanup happens automatically after you save your configuration.
  </Accordion>

  <Accordion title="Can I recover data after it's been deleted?">
    No. Data deletion through retention policies is permanent and cannot be undone. There is no backup or recovery mechanism for deleted data. Export any data you need to preserve before it's deleted.
  </Accordion>

  <Accordion title="Does Data Retention affect my knowledge bases?">
    No. Data Retention policies do not affect knowledge bases — they remain under your manual control and must be deleted explicitly if needed.
  </Accordion>

  <Accordion title="Will my active agent versions be deleted?">
    No. Your current active and draft versions of agents, tools, and workforces are always protected from deletion, regardless of age. Only old, inactive versions are eligible for deletion.
  </Accordion>

  <Accordion title="How do I know what data will be deleted?">
    Data is deleted based on the conditions outlined in the "What data is covered?" section above. Review those conditions carefully to understand what data is eligible for deletion based on your retention period.
  </Accordion>

  <Accordion title="Can I temporarily disable Data Retention?">
    Yes. You can toggle the "Enable" slider off at any time to pause automatic data deletion. Your retention period configuration is saved, so you can re-enable it later without reconfiguring.
  </Accordion>

  <Accordion title="Does this affect my S3 Event Streaming data?">
    No. If you use [S3 Event Streaming](/docs/enterprise/streaming-events), audit logs and events delivered to your own S3 bucket are not affected by Data Retention policies in Relevance AI. You control the lifecycle of data in your own S3 bucket through its lifecycle policy. On the Relevance side, telemetry only passes through a temporary staging bucket that is automatically purged after 30 days.
  </Accordion>

  <Accordion title="Who can configure Data Retention settings?">
    Only organization admins and owners can configure Data Retention settings. Members, viewers, and other roles cannot access or modify these settings.
  </Accordion>
</AccordionGroup>

***

## Related Enterprise features

<CardGroup>
  <Card title="S3 Event Streaming" icon="database" href="/enterprise/streaming-events">
    Stream audit logs and compliance data to your own S3 bucket for long-term retention
  </Card>

  <Card title="Role-based access controls" icon="shield-halved" href="/enterprise/rbac">
    Control who can access and manage data with granular permissions
  </Card>

  <Card title="Analytics" icon="chart-line" href="/enterprise/analytics">
    Monitor usage patterns to inform your retention policy decisions
  </Card>

  <Card title="Integration controls" icon="plug" href="/enterprise/integration-controls">
    Govern which integrations can access your organization's data
  </Card>
</CardGroup>

***

<Note>
  **Need help?** Enterprise customers have access to dedicated support channels. Contact your sales representative or [reach out to our support team](/docs/get-started/support) for assistance with Data Retention configuration.
</Note>


# Default organization and project
Source: https://relevanceai.com/docs/enterprise/default-org

Auto-assign users to the right organization and project

<img alt="Default Organization & Project" />

Default Organization & Project simplifies enterprise onboarding by automatically placing new users in the correct organization and project when they sign up with their company email. This eliminates manual assignments and ensures immediate access to the right resources from day one.

## Prerequisites

<Warning>
  Before you can set a default organization and project, your company email domain must be configured by the Relevance AI team. This is a required first step.
</Warning>

Setting up a default organization and project is a **two-step process**:

<Steps>
  <Step title="Email Domain Configuration" icon="envelope">
    **Contact the Relevance AI team** to configure your company email domain (e.g., @betacorp.com, @acmecorp.io, @techstartup.com). This step is typically completed during your [SSO setup](/docs/enterprise/sso-setup) or can be configured separately.

    <Info>
      This configuration allows Relevance AI to recognize users signing up with your company email domain and automatically route them to your organization instead of creating separate personal organizations.
    </Info>
  </Step>

  <Step title="Set Default Project" icon="folder">
    Once your email domain is configured, organization admins can set a default project following the instructions below.
  </Step>
</Steps>

### What You'll Need

To complete the setup after email domain configuration:

* **Enterprise subscription** with Relevance AI
* **Organization Admin** role or higher
* Your company email domain already configured by Relevance AI (Step 1 above)

***

## How do I set a default organization and project?

<Info>To set a default organization and project, you must have an Enterprise subscription and be an organization admin. Your company email domain must also be configured by the Relevance AI team (see Prerequisites above).</Info>

<div>
  <iframe title="How to set default organization and project" />
</div>

1. Click 'Settings' in the sidebar of Relevance AI
2. Then, select the project you want to set as default
3. Click on the three dots button in the top right of the project modal
4. Then, select 'Make Default'

Once you set a default project, new users with your company email domain(s) will be automatically added to this project in your organization when they create a new account.

### User Permissions for New Users

When new users are automatically added to your organization and default project, they receive the following permissions:

* **Organization-level**: Organization Viewer
* **Project-level**: Project Viewer

<Tip>
  Organization and project admins can adjust these permissions after users join. Learn more about managing user permissions in our [Role-Based Access Controls (RBAC)](/docs/enterprise/rbac) documentation.
</Tip>

***

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Why can't I set a default organization?">
    If you're unable to set a default organization and project, the most common reason is that your company email domain hasn't been configured yet. Contact the Relevance AI team to have your email domain configured. This is typically done during [SSO setup](/docs/enterprise/sso-setup) but can also be configured separately.

    You must also:

    * Have an Enterprise subscription
    * Be an organization admin or owner
  </Accordion>

  <Accordion title="What permissions do new users get?">
    New users who are automatically added through the default organization feature receive:

    * **Organization Viewer** permissions at the organization level
    * **Project Viewer** permissions at the project level

    These are the most restrictive permissions, ensuring security by default. Organization and project admins can upgrade permissions as needed. See our [RBAC documentation](/docs/enterprise/rbac) for details on all available roles and permissions.
  </Accordion>

  <Accordion title="What if I invite a new user to a project that isn't the default project?">
    Once they create their account, they'll be added to both the default project and the project they were invited to.
  </Accordion>

  <Accordion title="Will new users have new organizations created for them?">
    No, they'll be added to your organization and the default project in that organization. This is only possible after your company email domain has been configured by the Relevance AI team.
  </Accordion>

  <Accordion title="Can I configure my email domain myself?">
    No, email domain configuration must be done by the Relevance AI team for security reasons. Contact your Relevance AI representative to request this configuration. If you're setting up SSO, this will typically be handled as part of that process.
  </Accordion>
</AccordionGroup>


# Organization-level integration controls
Source: https://relevanceai.com/docs/enterprise/integration-controls

Enable integrations that your team can use at the organization level

Integration controls is an Enterprise feature that allows organization admins and owners to enable integrations that your team can use at the organization level.

This feature will hide integrations and LLMs that aren't enabled in-app for most users, while organization and project admins can still see all integration options in their respective control pages.

## How do I set organization-level integration controls?

<Info>To set organization-level integration controls, you must have an Enterprise subscription and be an organization admin or owner.</Info>

<div>
  <iframe title="How to set organization level integration controls" />
</div>

1. Click 'Settings' in the sidebar of Relevance AI
2. Then, click 'Integration controls'
3. From this page, you can quickly toggle all integrations on or off
4. To toggle a specific integration on or off, click the toggle next to the integration
5. You can also use the search bar to find an integration

## Understanding integration dependencies

The Integration Controls panel displays a **"Used by"** annotation beneath each integration toggle. This shows which native tool steps depend on that integration, so you can see the potential impact before enabling or disabling it.

For example:

* **Serper** shows "Used by: Google Search"
* **Browserless** shows "Used by: Web Scrape, Extract Content"

Use these annotations to audit your organization's tool dependencies and make informed decisions when managing integrations.

***

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can I access integration controls without upgrading to Enterprise?">No. This feature is available for Enterprise subscriptions only.</Accordion>
</AccordionGroup>


# Managing projects and users with the API
Source: https://relevanceai.com/docs/enterprise/manage-projects-and-users-api

Use the API to create projects, manage invites, and control user access

Relevance AI has APIs for teams to automate their project controls and governance. For up-to-date API docs, please visit the API documentation for the region your organization is hosted in:

* **AU**: [https://api-f1db6c.stack.tryrelevance.com/latest/documentation](https://api-f1db6c.stack.tryrelevance.com/latest/documentation)
* **EU**: [https://api-d7b62b.stack.tryrelevance.com/latest/documentation](https://api-d7b62b.stack.tryrelevance.com/latest/documentation)
* **US**: [https://api-bcbe5a.stack.tryrelevance.com/latest/documentation](https://api-bcbe5a.stack.tryrelevance.com/latest/documentation)

In order to use these APIs, you must create an Relevance AI API Key which will have access to your full account. You can generate this from 'Integrations & API Keys'.

<Info>All API references on this page will reference the AU endpoint. Please change the endpoint if your organization is hosted in EU or US.</Info>

## Projects

### Create projects

Creates a project in the specified organization.

```bash theme={null}
curl https://api-f1db6c.stack.tryrelevance.com/latest/projects/create \
  --request POST \
  --header 'Content-Type: application/json' \
  --header 'Authorization: Bearer :<api_key>' \
  --data '{
  "name": "",
  "description": "",
  "organization_id": ""
}'
```

#### Response

```bash theme={null}
{
  "project_id": "string"
}
```

***

### Delete projects

This API endpoint is currently being worked on and is not yet available.

***

## Invites

### Invite a user to a project

**Permissions Required**: Admin

The available actions are:

* `role:editor` for editor permissions
* `role:viewer` for viewer permissions
* `admin` for admin permissions

#### Notes

* In order to update the roles on the user invite, it must first be deleted and then recreated.
* If a user is already part of the organization, the invite will be automatically accepted.

```bash theme={null}
curl https://api-f1db6c.stack.tryrelevance.com/latest/auth/invite/create \
  --request POST \
  --header 'Content-Type: application/json' \
  --header 'Authorization: Bearer <project_id>:<api_key>' \
  --data '{
  "permissions": {
    "items": {
      "relevanceDefault": {
        "actions": {
          "role:editor": true
        }
      }
    },
  },
  "email": ""
}'
```

#### Response

```bash theme={null}
{
  "invite_code": "string"
}
```

***

### Update a user's role on a project

**Permissions Required**: Admin

```bash theme={null}
curl 'https://api-f1db6c.stack.tryrelevance.com/latest/auth/users/<user_id>/update' \
  --request POST \
  --header 'Content-Type: application/json' \
  --header 'Authorization: <project_id>:<api_key>' \
  --data '{
  "permissions": {
    "items": {
	    "relevanceDefault": {
        "actions": {
          "role:editor": true
        }
      }
    }
  }
}'
```

***

### Delete a project invite

**Permissions Required**: Admin

```bash theme={null}
curl https://api-f1db6c.stack.tryrelevance.com/latest/auth/invite/delete \
  --request POST \
  --header 'Content-Type: application/json' \
  --header 'Authorization: Bearer <project id>:<api_key>' \
  --data '{
  "invite_code": ""
}'
```

***

### Remove a user from a project

**Permissions Required**: Admin

```bash theme={null}
curl 'https://api-f1db6c.stack.tryrelevance.com/latest/auth/users/{user_id}/delete' \
  --request POST \
  --header 'Content-Type: application/json' \
  --header 'Authorization: Bearer <project id>:<api_key>' \
  --data '{}'
```

***

## SCIM - enterprise user management

<Warning>This feature is coming soon!</Warning>

SCIM support is coming soon to Relevance AI. Currently, enterprises can deactivate users using the WorkOS Dashboard or SCIM API. If a user is marked as deactivated in WorkOS, their accounts and associated API keys will be deactivated immediately. Please see the SCIM docs for WorkOS for documentation on how to use the API.


# Organization- and project-level API keys
Source: https://relevanceai.com/docs/enterprise/org-project-level-api-keys

Set API keys at the organization and project level in Relevance AI

Organization and project-level API keys are an Enterprise feature. Enterprise subscriptions allow you to set API keys at both a project level and organization level.

## How do I set organization-level API keys?

<Info>To set organization-level API keys, you must have an Enterprise subscription and be an organization admin or owner.</Info>

<Note>
  If you set an API key at the organization level, any projects in this organization will use this API key *unless* an API key is added at the project level.
</Note>

<div>
  <iframe title="How to set organization level API keys" />
</div>

1. First, click 'Settings' in the sidebar of Relevance AI
2. Then, click 'Integrations control'
3. Navigate to the integration you want to set an API key for and toggle it on
4. Enter the API key on the pop-up that appears, and click 'Save'

## How do I set project-level API keys?

<Info>To set project-level API keys, you must have an Enterprise subscription and be an organization admin or owner, or project admin or editor.</Info>

<Note>
  If you set an API key at the project level, ***this will override the organization level API key***.
</Note>

<div>
  <iframe title="How to set project level API keys" />
</div>

1. First, navigate to your project in Relevance AI and click 'Integrations & API Keys'
2. Then, navigate to the integration you want to set an API key for
3. Enter the API key on the pop-up that appears, and click 'Save'

***

## Custom API Keys

Custom API Keys allow you to securely store and use API credentials from external services in your agents and tools. Understanding how these keys are accessed and shared within your projects is important for maintaining security.

### Who Can Access Custom API Keys

Custom API Keys are accessed in [JavaScript code steps](/docs/build/tools/tool-steps/code-javascript) and [Python code steps](/docs/build/tools/tool-steps/python-code/code-python) using the syntax `{{secrets.key_name}}`. These keys are available to **all agent builders** within a project.

Key points about access:

* Any builder with access to the project can use any Custom API Key in their code execution steps
* Custom API Key access is project-wide, regardless of who created the key
* All project members with builder permissions (Admin, Editor, or Member who can create assets) can reference Custom API Keys in their tools and agents

<Info>
  Learn more about managing project membership and permissions in the [RBAC documentation](/docs/enterprise/rbac).
</Info>

### Security Considerations

<Warning>
  While Custom API Keys are automatically redacted with `<redacted>` when code steps attempt to print them out, members who can create Custom API Keys should exercise caution. Other builders in the project can use those keys to gain elevated access to downstream resources.
</Warning>

Important security points:

* **Automatic redaction**: Custom API Keys are protected from accidental exposure in logs and outputs
* **Shared access risk**: Any builder in the project can use Custom API Keys to access external systems with the permissions granted to those credentials
* **Project membership matters**: Carefully consider who has builder access to projects containing sensitive API credentials

***

## Member self-service API keys

On enterprise plans, users with the **Member** role can create and manage their own personal Relevance API keys. This lets Members set up programmatic access independently, without needing an administrator to provision credentials.

<Note>
  Members can only create and manage API keys for their own account. They cannot manage API keys for other users, or set organization- or project-level integration keys.
</Note>

The primary use case is connecting MCP clients to Relevance AI. A Member who needs to configure a client like Claude Desktop, Cursor, or VS Code can generate their own API key without waiting for admin intervention.

For more on generating Relevance API keys, see [API integration](/docs/get-started/core-concepts/api-integration#relevance-api-key).

***

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can I access organization- and project-level API keys without upgrading to Enterprise?">
    No. This feature is available for Enterprise subscriptions only.
  </Accordion>

  <Accordion title="I've set an organization- and project-level API key. Which one takes precedence?">The *project-level API key does*. If you want the organization-level API key to be used, the project-level API key must be removed.</Accordion>
  <Accordion title="How does this apply to OAuth accounts?">Integrations that rely on OAuth or third-party services don't support shared credentials, so users will need to connect their own, or service accounts.</Accordion>

  <Accordion title="What are the best practices for using Custom API Keys?">
    * Provide Custom API Keys that are limited in scope and permissions (e.g., read-only access, specific resource access)
    * Create separate projects with only trusted team members when working with sensitive credentials
    * Use OAuth credentials via [User Level Authentication](/docs/enterprise/user-level-authentication) instead of Custom API Keys where possible
  </Accordion>

  <Accordion title="Can I restrict which builders can use specific Custom API Keys?">Currently, all Custom API Keys in a project are accessible to all builders in that project. Use project isolation to control access to sensitive credentials by creating separate projects with limited membership.</Accordion>
  <Accordion title="What's the difference between Custom API Keys and OAuth authentication?">Custom API Keys are shared credentials stored at the project level that all builders can use. OAuth authentication via [User Level Authentication](/docs/enterprise/user-level-authentication) allows each user to authenticate with their own credentials, providing better security and individual accountability.</Accordion>
</AccordionGroup>


# Organization Structure Best Practices
Source: https://relevanceai.com/docs/enterprise/org-structure-best-practices

Design an effective organizational structure for your Enterprise deployment

<Note>
  This guide is designed for Enterprise customers planning their Relevance AI deployment. These best practices help you scale efficiently and maintain governance as your AI workforce grows.
</Note>

A well-planned organizational structure is critical for Enterprise success with Relevance AI. The right structure makes it easier to manage permissions, track usage, maintain governance, and scale your AI workforce as your team grows.

## Understanding the hierarchy

Relevance AI uses a three-level hierarchy for organizing resources:

<Steps>
  <Step title="Organization" icon="building">
    The top-level container for your entire company. All users, projects, and assets belong to one organization.

    **Key characteristics**:

    * One organization per company
    * Organization-level billing and credits
    * Organization-level SSO and security settings
    * Organization-level admins manage everything
  </Step>

  <Step title="Projects" icon="folder-tree">
    Containers for related agents, tools, knowledge bases, and workforces. Projects are the primary way to organize work and control access.

    **Key characteristics**:

    * Multiple projects per organization
    * Project-level permissions and access control
    * Project-level usage tracking and limits
    * Project-level integrations and API keys
  </Step>

  <Step title="Assets" icon="cube">
    Individual agents, tools, knowledge bases, and workforces. Assets are the building blocks of your AI workforce.

    **Key characteristics**:

    * Assets belong to one project
    * Asset-level permissions for granular control
    * Assets can be shared across projects (with proper permissions)
    * Assets can be organized into folders within projects
  </Step>
</Steps>

***

## Project organization strategies

There are three main strategies for organizing projects. Most organizations use a combination of these approaches.

<Tabs>
  <Tab title="By Team">
    ### Organize by Team or Department

    Organize projects around your organizational structure (teams, departments, or business units).

    **Best for:** Companies with distinct teams that work independently

    **Example structure:**

    ```
    Organization: Acme Corp
    ├── Project: Marketing Team
    ├── Project: Sales Team
    ├── Project: Customer Support
    ├── Project: Product Team
    ├── Project: Finance & Operations
    └── Project: Executive Team
    ```

    **Advantages:**

    * Aligns with existing organizational structure
    * Easy to assign permissions based on team membership
    * Clear ownership and accountability
    * Simple usage tracking per team

    **Considerations:**

    * May lead to duplicate agents across teams
    * Harder to share resources across teams
    * Can create silos if not managed carefully

    **When to use:**

    * You have 3+ distinct teams using Relevance AI
    * Teams work independently with different use cases
    * You want to track usage and costs per team
    * Teams need different integration access or security policies
  </Tab>

  <Tab title="By Use Case">
    ### Organize by Use Case or Function

    Organize projects around specific use cases, workflows, or business functions.

    **Best for:** Companies with cross-functional teams working on shared use cases

    **Example structure:**

    ```
    Organization: Acme Corp
    ├── Project: Lead Generation & Qualification
    ├── Project: Customer Onboarding
    ├── Project: Support Automation
    ├── Project: Content Creation
    ├── Project: Data Enrichment
    └── Project: Internal Operations
    ```

    **Advantages:**

    * Encourages collaboration across teams
    * Reduces duplication of similar agents
    * Easier to share knowledge and best practices
    * Clear purpose for each project

    **Considerations:**

    * Requires cross-team coordination
    * May need more complex permission management
    * Harder to track usage per team

    **When to use:**

    * Multiple teams collaborate on the same use cases
    * You want to centralize similar workflows
    * You have dedicated AI/automation team managing projects
    * Use cases span multiple departments
  </Tab>

  <Tab title="By Environment">
    ### Organize by Environment or Stage

    Organize projects by development stage (development, staging, production).

    **Best for:** Companies with formal development and deployment processes

    **Example structure:**

    ```
    Organization: Acme Corp
    ├── Project: Development Sandbox
    ├── Project: Testing & QA
    ├── Project: Staging
    └── Project: Production
    ```

    **Advantages:**

    * Clear separation between development and production
    * Safer testing without affecting live agents
    * Easier to manage deployment pipelines
    * Better governance and change control

    **Considerations:**

    * Requires discipline to maintain separation
    * May need to duplicate assets across environments
    * More complex for non-technical users

    **When to use:**

    * You have strict change management requirements
    * You need to test agents before deploying to production
    * You have compliance or regulatory requirements
    * You have a dedicated DevOps or platform team
  </Tab>

  <Tab title="Hybrid (Recommended)">
    ### Hybrid Approach

    Most successful Enterprise deployments use a **hybrid approach** combining multiple strategies.

    **Example structure:**

    ```
    Organization: Acme Corp
    ├── Project: Marketing - Production
    ├── Project: Marketing - Development
    ├── Project: Sales - Production
    ├── Project: Sales - Development
    ├── Project: Support - Production
    ├── Project: Support - Development
    ├── Project: Shared Resources
    └── Project: Sandbox (for experimentation)
    ```

    **This approach provides:**

    * Team-based organization for clear ownership
    * Environment separation for safe development
    * Shared resources project for common tools and knowledge
    * Sandbox for experimentation without affecting production

    **Why it works:**

    The hybrid approach combines the best aspects of all strategies. Teams get their own projects for clear ownership and usage tracking, while environment separation (production/development) provides safety and governance. Shared resources prevent duplication, and a sandbox encourages innovation without risk.

    This is the most scalable approach and works well for organizations from 20 to 1000+ users.
  </Tab>
</Tabs>

***

## Structure by company size

Choose your team size to see recommended structures and best practices:

<Tabs>
  <Tab title="Small (5-20)">
    ### Small Teams (5-20 users)

    **Recommended structure:** Simple, minimal projects

    ```
    Organization: Acme Corp
    ├── Project: Production
    ├── Project: Development
    └── Project: Sandbox
    ```

    **Key recommendations:**

    Start simple with 2-3 projects. Use folders within projects to organize assets. Assign most users as Project Members or Editors. Use asset-level permissions for sensitive agents. Focus on getting value quickly, not perfect structure.

    <Warning>
      **Common mistakes to avoid:**

      * Creating too many projects too early
      * Over-engineering permissions
      * Spending too much time on structure instead of building
    </Warning>
  </Tab>

  <Tab title="Medium (20-100)">
    ### Medium Teams (20-100 users)

    **Recommended structure:** Team-based with shared resources

    ```
    Organization: Acme Corp
    ├── Project: Marketing
    ├── Project: Sales
    ├── Project: Customer Success
    ├── Project: Product & Engineering
    ├── Project: Shared Resources
    ├── Project: Development Sandbox
    └── Project: Executive & Leadership
    ```

    **Key recommendations:**

    Organize by team or department. Create a "Shared Resources" project for common tools and knowledge. Use RBAC Groups to assign teams to projects. Implement project-level usage limits. Designate project admins for each team. Use folders extensively within projects.

    <Warning>
      **Common mistakes to avoid:**

      * Creating a project for every small team or sub-team
      * Not using groups for bulk permission management
      * Allowing unlimited project creation by all users
    </Warning>
  </Tab>

  <Tab title="Large (100+)">
    ### Large Teams (100+ users)

    **Recommended structure:** Hybrid with governance

    ```
    Organization: Acme Corp
    ├── Project: Marketing - Production
    ├── Project: Marketing - Development
    ├── Project: Sales - Production
    ├── Project: Sales - Development
    ├── Project: Customer Success - Production
    ├── Project: Customer Success - Development
    ├── Project: Product - Production
    ├── Project: Product - Development
    ├── Project: Shared Libraries
    ├── Project: Shared Knowledge Bases
    ├── Project: Compliance & Audit
    └── Project: Innovation Sandbox
    ```

    **Key recommendations:**

    Separate production and development environments. Create dedicated projects for shared resources. Implement strict governance and approval processes. Use RBAC Groups extensively. Set up usage limits and monitoring. Designate a platform team to manage structure. Create naming conventions and documentation. Regular audits of projects and permissions.

    <Warning>
      **Common mistakes to avoid:**

      * Allowing uncontrolled project proliferation
      * Not documenting structure and policies
      * Inconsistent naming conventions
      * Not having a platform team or governance process
    </Warning>
  </Tab>
</Tabs>

***

## Asset organization within projects

Once you've structured your projects, organize assets within them using folders and naming conventions.

<AccordionGroup>
  <Accordion title="Folder structure best practices">
    Use folders to organize assets in ways that make sense for your workflow:

    **By asset type:**

    ```
    Project: Marketing
    ├── 📁 Agents
    │   ├── Content Writer Agent
    │   ├── SEO Optimizer Agent
    │   └── Social Media Agent
    ├── 📁 Tools
    │   ├── Content Analysis Tool
    │   └── Keyword Research Tool
    ├── 📁 Knowledge
    │   ├── Brand Guidelines
    │   └── Product Information
    └── 📁 Workforces
        └── Content Production Workforce
    ```

    **By use case or workflow:**

    ```
    Project: Sales
    ├── 📁 Lead Generation
    │   ├── Lead Finder Agent
    │   ├── Lead Enrichment Tool
    │   └── Company Database (Knowledge)
    ├── 📁 Lead Qualification
    │   ├── Lead Scorer Agent
    │   └── Qualification Criteria (Knowledge)
    └── 📁 Outreach
        ├── Email Writer Agent
        └── Outreach Workforce
    ```

    **By status or stage:**

    ```
    Project: Customer Success
    ├── 📁 Active / Production
    ├── 📁 In Development
    ├── 📁 Testing
    └── 📁 Archived
    ```
  </Accordion>

  <Accordion title="Naming conventions">
    Establish clear naming conventions for consistency across your organization:

    **For projects:**

    * Use descriptive names: `Marketing - Production` not `Mktg Prod`
    * Include environment if applicable: `Sales - Development`
    * Use consistent separators: `-` or `|`

    **For assets:**

    * Start with asset type: `Agent: Lead Qualifier` or `Tool: Data Enrichment`
    * Include purpose: `Agent: Customer Support - Tier 1`
    * Use version numbers if needed: `Agent: Content Writer v2`
    * Avoid special characters that may cause issues

    **For folders:**

    * Use emojis for visual clarity: `📁 Active Agents`, `🔧 Tools`, `📚 Knowledge`
    * Keep names short and clear
    * Use consistent capitalization
  </Accordion>
</AccordionGroup>

***

## Permission management

<AccordionGroup>
  <Accordion title="Organization-level roles">
    **Owner role:**

    Assign to 1-2 people maximum (CEO, CTO, or Head of Operations). These users have full control including billing.

    **Admin role:**

    Assign to IT leads, platform team, or workspace admins. Typically 2-5 people depending on company size. These users manage projects, users, and security.

    **Member role:**

    Default role for most users. Can create their own projects and assets. Good for power users and builders.

    **Viewer role:**

    For external collaborators or users in onboarding. Cannot create anything until assigned to a project. Good for stakeholders who need visibility only.
  </Accordion>

  <Accordion title="Project-level roles">
    **Admin role:**

    Assign to project owners (1-2 per project). These users manage project permissions and settings.

    **Editor role:**

    Assign to core contributors and builders. Can create and edit all assets in the project. Typically 20-40% of project users.

    **Member role:**

    Default role for most project users. Can create their own assets but don't see others' by default. Good for independent contributors.

    **Chat role:**

    For users who only need to interact with agents via Chat. Cannot access the web app. Perfect for end users and external users.

    **Viewer role:**

    For stakeholders who need visibility only. Cannot run or edit anything. Add only to specific assets they need to see.
  </Accordion>

  <Accordion title="Asset-level permissions">
    **Default approach:** Let project-level permissions handle most access control.

    **When to use asset-level permissions:**

    * Sensitive agents with confidential data
    * Agents that should only be run by specific users
    * Shared agents that need different permissions than project default
    * Agents used by external collaborators

    **Best practice:** Start with project-level permissions and add asset-level permissions only when needed. Over-complicating permissions makes administration harder and can cause confusion.
  </Accordion>
</AccordionGroup>

***

## Common organizational patterns

Choose the pattern that best fits your organization's governance model:

<AccordionGroup>
  <Accordion title="Center of Excellence (CoE)" icon="building-columns">
    A centralized team manages the platform and creates shared resources.

    **Structure:**

    ```
    Organization: Acme Corp
    ├── Project: CoE - Shared Agents (managed by platform team)
    ├── Project: CoE - Shared Tools (managed by platform team)
    ├── Project: CoE - Knowledge Libraries (managed by platform team)
    ├── Project: Marketing (uses shared resources)
    ├── Project: Sales (uses shared resources)
    └── Project: Support (uses shared resources)
    ```

    **Best for:**

    * Companies with dedicated AI/automation team
    * Organizations wanting centralized governance
    * Companies with compliance requirements

    **Why this works:**

    The CoE model ensures consistency and quality across the organization. A centralized team becomes experts in the platform, creates reusable resources, and maintains governance standards. Teams benefit from professionally-built shared resources without needing deep platform expertise.
  </Accordion>

  <Accordion title="Federated Model" icon="network-wired">
    Each team manages their own projects with light central governance.

    **Structure:**

    ```
    Organization: Acme Corp
    ├── Project: Marketing (self-managed)
    ├── Project: Sales (self-managed)
    ├── Project: Support (self-managed)
    ├── Project: Product (self-managed)
    └── Project: Shared Resources (light governance)
    ```

    **Best for:**

    * Companies with autonomous teams
    * Organizations with distributed decision-making
    * Companies prioritizing speed over standardization

    **Why this works:**

    The federated model empowers teams to move quickly and innovate independently. Each team controls their own destiny while a lightweight shared resources project prevents complete duplication. This works well for organizations that value team autonomy over centralized control.
  </Accordion>

  <Accordion title="Hybrid Model (Recommended)" icon="layer-group">
    Combines centralized shared resources with team autonomy.

    **Structure:**

    ```
    Organization: Acme Corp
    ├── Project: Platform - Shared Agents (centrally managed)
    ├── Project: Platform - Shared Tools (centrally managed)
    ├── Project: Platform - Knowledge Libraries (centrally managed)
    ├── Project: Marketing - Production (team-managed)
    ├── Project: Marketing - Development (team-managed)
    ├── Project: Sales - Production (team-managed)
    ├── Project: Sales - Development (team-managed)
    └── Project: Innovation Sandbox (open to all)
    ```

    **Best for:**

    * Most Enterprise organizations
    * Companies wanting balance between governance and autonomy
    * Organizations with both shared and team-specific use cases

    **Why this works:**

    The hybrid model combines the best of both worlds. Teams get autonomy for their specific needs while benefiting from centrally-managed shared resources. This prevents duplication, maintains quality standards, and still allows teams to innovate. It's the most scalable and flexible approach for growing organizations.
  </Accordion>
</AccordionGroup>

***

## Migration and restructuring

### When to restructure

Consider restructuring when:

* Projects have become too large or complex
* Permission management is becoming difficult
* Usage tracking doesn't align with business needs
* Teams are duplicating work across projects
* Compliance or security requirements change

### How to restructure safely

<Steps>
  <Step title="Plan the new structure">
    Document your current structure and design the new structure. Get buy-in from stakeholders.
  </Step>

  <Step title="Create new projects">
    Set up new projects with proper permissions before moving assets.
  </Step>

  <Step title="Move assets gradually">
    Move assets in phases, starting with non-critical ones. Test after each phase.
  </Step>

  <Step title="Update permissions">
    Ensure all users have appropriate access in the new structure.
  </Step>

  <Step title="Communicate changes">
    Notify all users about the new structure and any changes to their access.
  </Step>

  <Step title="Archive old projects">
    Once migration is complete, archive (don't delete) old projects for reference.
  </Step>
</Steps>

<Warning>
  Restructuring can temporarily disrupt access. Plan migrations during low-usage periods and communicate clearly with all users.
</Warning>

***

## Governance and maintenance

<AccordionGroup>
  <Accordion title="Establish clear policies">
    Document policies for:

    * Who can create new projects
    * Naming conventions for projects and assets
    * Approval process for new integrations
    * Usage limits and escalation procedures
    * Asset archival and deletion policies

    Share these policies in a central location that's easy to find and keep them updated as your deployment evolves.
  </Accordion>

  <Accordion title="Conduct regular audits">
    Conduct quarterly audits to:

    * Review project structure and usage
    * Remove inactive users and projects
    * Verify permissions are still appropriate
    * Identify duplicate or unused assets
    * Check compliance with policies

    Regular audits prevent organizational drift and keep your structure aligned with your business needs.
  </Accordion>

  <Accordion title="Designate clear ownership">
    Every project should have:

    * **Primary owner** (Project Admin) — responsible for day-to-day management
    * **Secondary owner** (Project Admin) — backup for primary owner
    * **Executive sponsor** — business stakeholder who approves major changes

    Clear ownership ensures accountability and makes it easy for users to know who to contact for access or questions.
  </Accordion>

  <Accordion title="Monitor and optimize">
    **Track these metrics:**

    * Number of active projects and assets
    * Users per project
    * Usage per project (credits consumed)
    * Permission changes and access requests
    * Asset creation and deletion rates

    **Use these metrics to identify:**

    * Projects that should be merged or split
    * Users who need training or support
    * Opportunities to create shared resources
    * Potential security or compliance issues

    Data-driven optimization keeps your structure efficient as your organization grows.
  </Accordion>
</AccordionGroup>

***

## Industry examples

See how organizations in different industries structure their Relevance AI deployments:

<Tabs>
  <Tab title="Technology / SaaS">
    ```
    Organization: TechCorp
    ├── Project: Product - Production
    ├── Project: Product - Development
    ├── Project: Engineering - Production
    ├── Project: Engineering - Development
    ├── Project: Customer Success
    ├── Project: Sales & Marketing
    ├── Project: Shared - Documentation & Knowledge
    └── Project: Sandbox
    ```

    **Key characteristics:**

    * Separate production and development environments
    * Engineering and Product have their own projects
    * Shared documentation project for company knowledge
  </Tab>

  <Tab title="Professional Services">
    ```
    Organization: ConsultingCo
    ├── Project: Client Services - Active Projects
    ├── Project: Client Services - Templates
    ├── Project: Business Development
    ├── Project: Internal Operations
    ├── Project: Knowledge Management
    └── Project: Training & Onboarding
    ```

    **Key characteristics:**

    * Client-facing work separated from internal operations
    * Templates project for reusable client deliverables
    * Dedicated knowledge management project
  </Tab>

  <Tab title="Financial Services">
    ```
    Organization: FinanceCorp
    ├── Project: Compliance & Audit (restricted access)
    ├── Project: Risk Management (restricted access)
    ├── Project: Client Services - Production
    ├── Project: Client Services - Development
    ├── Project: Operations - Production
    ├── Project: Operations - Development
    └── Project: Shared Resources
    ```

    **Key characteristics:**

    * Compliance and risk projects with restricted access
    * Strict production/development separation
    * Emphasis on governance and audit trails
  </Tab>

  <Tab title="E-commerce / Retail">
    ```
    Organization: RetailCo
    ├── Project: Customer Experience
    ├── Project: Marketing & Growth
    ├── Project: Operations & Fulfillment
    ├── Project: Merchandising
    ├── Project: Customer Support
    └── Project: Analytics & Insights
    ```

    **Key characteristics:**

    * Customer-centric project organization
    * Separate projects for different customer touchpoints
    * Analytics project for data-driven insights
  </Tab>
</Tabs>

***

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="How many projects should I create?">
    Start with 3-5 projects and add more as needed. Most organizations have 5-15 projects. If you have more than 20 projects, consider consolidating.
  </Accordion>

  <Accordion title="Can I move assets between projects?">
    Yes, but this requires proper permissions. Contact support for assistance with bulk asset moves.
  </Accordion>

  <Accordion title="Should I create separate projects for production and development?">
    For Enterprise deployments, yes. This provides safer testing and better governance. For smaller teams, you can use folders within a single project.
  </Accordion>

  <Accordion title="How do I handle shared resources across teams?">
    Create a "Shared Resources" or "Platform" project that all teams can access. Use asset-level permissions to control who can edit vs. use shared resources.
  </Accordion>

  <Accordion title="What's the best way to organize a project with 50+ agents?">
    Use folders extensively to organize by use case, status, or team. Consider splitting into multiple projects if agents serve very different purposes.
  </Accordion>

  <Accordion title="Can users be in multiple projects?">
    Yes, users can be members of multiple projects with different roles in each project.
  </Accordion>

  <Accordion title="How do I prevent project sprawl?">
    Require approval for new projects, conduct regular audits, and establish clear criteria for when a new project is needed vs. using an existing one.
  </Accordion>

  <Accordion title="Should I use asset-level permissions or project-level permissions?">
    Start with project-level permissions for simplicity. Add asset-level permissions only for sensitive or special-case assets.
  </Accordion>
</AccordionGroup>


# Quick start guide for Enterprise
Source: https://relevanceai.com/docs/enterprise/quick-start-guide

Your complete guide to setting up and configuring Relevance AI for Enterprise teams

<Note>
  This guide is designed for Enterprise customers. If you're not on an Enterprise subscription, [contact our sales team](https://relevanceai.com/contact) to learn more about Enterprise features.
</Note>

Welcome to Relevance AI for Enterprise! This guide will walk you through the recommended setup process to get your organization up and running with enterprise-grade security, governance, and collaboration features.

Enterprise features in Relevance AI provide your organization with advanced security controls, bulk user management, compliance capabilities, and granular access controls across your AI workforce.

## What you'll set up

<Columns>
  <Card title="Single Sign-On (SSO)" icon="key">
    Secure authentication through your identity provider (Okta, Azure AD/Entra ID, or Google Workspace)
  </Card>

  <Card title="Directory sync & groups" icon="users">
    Automatic user provisioning and bulk access management from your IDP
  </Card>

  <Card title="Organization structure" icon="sitemap">
    Projects, teams, and workspace organization best practices
  </Card>

  <Card title="Role-based access controls" icon="shield-halved">
    Granular permissions at organization, project, and asset levels
  </Card>

  <Card title="Usage controls" icon="gauge-high">
    Credit limits, monitoring, and usage alerts across your organization
  </Card>

  <Card title="Event streaming" icon="database">
    Audit logs and compliance data streamed to your S3 bucket
  </Card>

  <Card title="Integration governance" icon="plug">
    Control which third-party integrations your teams can use
  </Card>

  <Card title="API key management" icon="code">
    Organization and project-level API keys for secure automation
  </Card>

  <Card title="Data retention" icon="clock-rotate-left">
    Automatically manage data lifecycle and retention policies
  </Card>

  <Card title="User Level Authentication" icon="user-lock">
    Enable agents to use each user's own credentials for integrations
  </Card>

  <Card title="Visual Data Masking" icon="eye-slash">
    Mask PII in agent conversation histories and task views for UI privacy (limited beta)
  </Card>
</Columns>

## Prerequisites

<Warning>
  Before you begin, ensure you have all of these prerequisites. Missing any of these may prevent you from completing the setup process.
</Warning>

**Required for setup:**

* An active Enterprise subscription with Relevance AI
* Organization Owner or Admin access in Relevance AI
* Access to your identity provider (Okta, Azure AD/Entra ID, or Google Workspace)
* Admin credentials for your identity provider

<Info>
  If you don't have these prerequisites, contact your Relevance AI sales representative or [reach out to support](/docs/get-started/support).
</Info>

***

## Recommended setup order

Follow these steps in order for the smoothest Enterprise onboarding experience. Each step builds on the previous one to create a secure, well-governed AI platform for your organization.

<Steps>
  <Step title="Connect single sign-on (SSO)" icon="key">
    **Why this matters**: SSO is the foundation of Enterprise security. It ensures all users authenticate through your identity provider, enabling centralized access control and compliance.

    **What you'll do**:

    * Configure SSO with your identity provider (Okta, Azure AD/Entra ID, or Google Workspace)
    * Test SSO login flow
    * Enforce SSO for all organization members

    **Time required**: 30-45 minutes

    [View SSO setup guide →](/docs/enterprise/sso-setup)
  </Step>

  <Step title="Enable directory sync & groups" icon="users">
    **Why this matters**: Directory Sync automatically provisions users and syncs groups from your identity provider, eliminating manual user management and enabling bulk access control.

    **What you'll do**:

    * Connect your identity provider directory to Relevance AI
    * Configure which groups to sync
    * Verify group synchronization

    **Time required**: 20-30 minutes

    <Info>
      **Feature Status**: Directory Sync & RBAC Groups are in Beta with full rollout by **November 20, 2025**. Contact your sales representative to enable this feature.
    </Info>

    [View Directory Sync setup guide →](/docs/enterprise/rbac-groups)
  </Step>

  <Step title="Plan your organization & project structure" icon="sitemap">
    **Why this matters**: A well-planned organizational structure makes it easier to manage permissions, track usage, and scale your AI workforce as your team grows.

    **What you'll do**:

    * Decide how to organize projects (by team, department, or use case)
    * Plan folder structure for agents and assets
    * Determine project ownership and admin assignments

    **Time required**: 15-30 minutes (planning)

    [View organization structure best practices →](/docs/enterprise/org-structure-best-practices)
  </Step>

  <Step title="Configure role-based access controls (RBAC)" icon="shield-halved">
    **Why this matters**: RBAC ensures users only have access to what they need, supporting security, compliance, and operational control.

    **What you'll do**:

    * Assign organization-level roles (Owner, Admin, Member, Viewer)
    * Set project-level permissions (Admin, Editor, Member, Chat, Viewer)
    * Configure asset-level access for sensitive agents and tools
    * Assign groups to projects and assets (if using Directory Sync)

    **Time required**: 30-60 minutes

    [View RBAC configuration guide →](/docs/enterprise/rbac)
  </Step>

  <Step title="Set usage limits & monitoring" icon="gauge-high">
    **Why this matters**: Usage controls help you manage costs, prevent overuse, and ensure fair resource allocation across teams.

    **What you'll do**:

    * Set organization-wide credit limits
    * Configure project-level usage caps
    * Set up usage alerts and notifications
    * Review usage dashboards

    **Time required**: 15-20 minutes

    [View usage limits setup guide →](/docs/admin/project-management/usage-limits)
  </Step>

  <Step title="Connect S3 bucket for event streaming (optional)" icon="database">
    **Why this matters**: Streaming events to your own S3 bucket provides audit logs, compliance data, and usage analytics for your security and data teams.

    **What you'll do**:

    * Create and configure an S3 bucket
    * Set up IAM permissions
    * Connect your bucket to Relevance AI
    * Verify event streaming

    **Time required**: 20-30 minutes

    <Info>
      **Feature Status**: S3 Event Streaming will be fully available by **November 20, 2025**, with self-serve setup by end of year. Contact your sales representative for early access.
    </Info>

    [View event streaming setup guide →](/docs/enterprise/streaming-events)
  </Step>

  <Step title="Configure integration controls" icon="plug">
    **Why this matters**: Integration controls let you govern which third-party services your teams can connect, supporting security policies and compliance requirements.

    **What you'll do**:

    * Review available integration controls
    * Set organization-level integration policies
    * Configure project-level integration restrictions

    **Time required**: 15-20 minutes

    [View integration controls guide →](/docs/enterprise/integration-controls)
  </Step>

  <Step title="Set up API key management (optional)" icon="code">
    **Why this matters**: Organization and project-level API keys allow you to centrally manage credentials for integrations, ensuring consistent authentication across your teams and simplifying key rotation.

    **What you'll do**:

    * Configure organization-level API keys for company-wide integrations
    * Set project-level API keys to override organization defaults when needed
    * Understand key precedence (project-level overrides organization-level)

    **Time required**: 10-15 minutes

    <Note>
      If you set an API key at the organization level, all projects will use it unless a project-level key is specified. Project-level keys always take precedence over organization-level keys.
    </Note>

    [View API key management guide →](/docs/enterprise/org-project-level-api-keys)
  </Step>

  <Step title="Configure data retention (optional)" icon="clock-rotate-left">
    **Why this matters**: Data Retention policies help you meet compliance requirements, reduce security risks, and optimize storage by automatically deleting old data.

    **What you'll do**:

    * Request Data Retention feature enablement from Relevance AI
    * Set your organization's retention period (in days)
    * Configure automatic deletion policies for tool runs, conversations, and versions
    * Understand what data is protected from deletion

    **Time required**: 10-15 minutes

    <Info>
      **Feature Status**: Data Retention must be enabled by the Relevance AI team. Contact your sales representative to request this feature.
    </Info>

    [View Data Retention setup guide →](/docs/enterprise/data-retention)
  </Step>

  <Step title="Enable User Level Authentication (Optional)" icon="user-lock">
    **Why this matters**: User Level Authentication allows each user to authenticate with their own credentials when using agents, ensuring privacy and proper access control across integrations.

    **What you'll do**:

    * Enable User Level Authentication on agent tools that use OAuth integrations
    * Users will be prompted to connect their own accounts when using agents
    * Credentials are saved and reused automatically for future runs

    **Time required**: 5-10 minutes per agent

    [View User Level Authentication guide →](/docs/enterprise/user-level-authentication)
  </Step>

  <Step title="Enable Visual Data Masking on sensitive agents (Optional)" icon="eye-slash">
    **Why this matters**: Visual Data Masking automatically masks PII (email addresses, phone numbers, credit card numbers, and names) in agent conversation histories and task views. This protects sensitive information during screen sharing, team reviews, and demos — while agents continue to process the actual unmasked data.

    **What you'll do**:

    * Request Visual Data Masking enablement from your account manager
    * Enable the toggle in agent or workforce Advanced settings
    * Verify masking behavior with sample data before rolling out to production agents

    **Time required**: 5-10 minutes per agent (after feature is enabled)

    <Info>
      **Feature Status**: Visual Data Masking is in limited beta. Contact your account manager or [reach out to support](/docs/get-started/support) to have it enabled for your organization.
    </Info>

    <Warning>
      If any agent in a workforce has Visual Data Masking enabled, PII will be masked across the entire workforce. Plan your enablement accordingly.
    </Warning>

    [View Visual Data Masking guide →](/docs/build/agents/build-your-agent/agent-settings/visual-data-masking)
  </Step>
</Steps>

***

## Feature rollout timeline

Some Enterprise features are being gradually rolled out. Here's what to expect:

| Feature                               | Status          | Availability                      |
| ------------------------------------- | --------------- | --------------------------------- |
| SSO (Okta, Azure AD/Entra ID)         | ✅ Available     | Now                               |
| RBAC (Organization, Project, Asset)   | ✅ Available     | Now                               |
| Directory Sync & Groups               | ✅ Available     | Now                               |
| Integration Controls                  | ✅ Available     | Now                               |
| Data Retention                        | ✅ Available     | Contact sales to enable           |
| S3 Event Streaming                    | ✅ Available     | Now                               |
| Organization & Project-level API Keys | ✅ Available     | Now                               |
| Asset-level Authentication Controls   | ✅ Available     | Now                               |
| User Level Authentication             | 🔄 Rolling out  | Contact sales to enable           |
| Visual Data Masking                   | 🔄 Limited beta | Contact account manager to enable |

<Info>
  If you have an Enterprise subscription but don't have access to a feature marked as "Beta" or "Gradual rollout," contact your sales representative to express interest and get early access.
</Info>

***

## Best practices for Enterprise deployment

<Tabs>
  <Tab title="Security first">
    ### Configure security before inviting users

    Set up SSO and RBAC before inviting your entire organization. This ensures users have appropriate access from day one and prevents security gaps during onboarding.

    **Initial security setup**

    Complete SSO setup and test with a pilot group before enforcing it organization-wide. Configure organization and project-level roles thoughtfully, ensuring that sensitive resources have additional asset-level permissions.

    **Multi-factor authentication**

    Work with your identity provider team to enable multi-factor authentication (MFA) for an additional security layer. This is especially important for Admin and Owner roles.

    **The cost of shortcuts**

    The security foundation you build now will scale with your organization. Taking shortcuts during initial setup often leads to security remediation work later, which is more disruptive and time-consuming than getting it right from the start.
  </Tab>

  <Tab title="Scale with groups">
    ### Use directory sync for teams of 20+

    If you have more than 20 users, use Directory Sync and Groups instead of managing individual user permissions. This approach saves time, reduces errors, and scales effortlessly as your organization grows.

    **Leverage existing structure**

    Enable Directory Sync with your identity provider and organize users into logical groups based on your existing structure—whether by team, department, or function. Assign entire groups to projects instead of managing hundreds of individual users.

    **Automatic synchronization**

    This approach leverages the organizational structure you've already built in your identity provider and keeps access control synchronized automatically. When someone joins the Marketing team in your IDP, they automatically get access to Marketing projects in Relevance AI.

    **Naming conventions matter**

    Document your group naming conventions early. As your deployment grows, consistent naming makes it easy for project admins to find and assign the right groups. The time invested in setting up Directory Sync properly pays dividends as your organization scales.
  </Tab>

  <Tab title="Plan structure">
    ### Design before you build

    Take time to plan your organization structure before creating projects and assets. Restructuring later is more difficult than getting it right from the start.

    **Start with strategy**

    Map out your project structure before creating anything in Relevance AI. Decide whether to organize by team, use case, environment, or a hybrid approach. Establish clear naming conventions for projects and assets, and define ownership models.

    **Think ahead**

    Consider both your current needs and how you expect the platform to grow over the next 12-24 months. Will you need separate production and development environments? How will you handle shared resources? Where will experimental work live?

    **Organize within projects**

    Plan folder structures within projects to keep assets organized. A well-designed structure makes it easier to manage permissions, track usage, and scale operations as adoption grows.
  </Tab>

  <Tab title="Access control">
    ### Follow least privilege principle

    Grant users the minimum permissions they need to do their work. You can always expand access later, but restricting access after the fact is harder and may disrupt workflows.

    **Start restrictive, expand as needed**

    Start by assigning users to the Viewer role and escalate permissions based on demonstrated need. Use project-level permissions as your default access control mechanism, adding asset-level permissions only for sensitive resources that require additional restrictions.

    **Careful with admin access**

    Assign project Admin roles carefully—typically just one or two people per project who are responsible for managing that project's resources and access. Too many admins create confusion about ownership and accountability.

    **Avoid a single org admin**

    At the organization level, assign the Admin role to more than one person. Org Admins manage organization users, organization-level settings, org-level API keys and OAuth connections, and project roles—so if your only Admin is unavailable, no one else can manage members or organization-wide settings. Keeping two or more org Admins preserves continuity while still maintaining clear accountability.

    **Regular audits**

    Regularly audit permissions during growth phases, especially when teams reorganize or people change roles. The principle of least privilege isn't just about security—it also reduces confusion by ensuring users only see the resources relevant to their work.
  </Tab>

  <Tab title="Governance">
    ### Establish clear policies

    Create and document governance policies before scaling your deployment. This prevents confusion, ensures consistency, and helps maintain security standards.

    **Document your rules**

    Document who can create projects, what approval processes exist for new integrations, and how teams should name their projects and assets. Establish regular permission audits—quarterly is a good starting point—and create clear workflows for requesting access.

    **Enable, don't block**

    Good governance doesn't mean slowing teams down. It means providing clear guidelines that help teams move quickly while maintaining security and compliance. When people know the rules and understand why they exist, they're more likely to follow them.

    **Make it accessible**

    Share your policies in a central location that's easy to find and update them as your deployment evolves. Consider creating a dedicated project or knowledge base for governance documentation.
  </Tab>

  <Tab title="Pilot & monitor">
    ### Test with a pilot team

    Before rolling out to your entire organization, test your setup with a small pilot team (5-10 users). This helps you identify issues and refine your configuration without impacting everyone.

    **Choose diverse pilot users**

    Select pilot users from different teams and use cases to get diverse feedback. Test SSO login flows thoroughly, verify that permissions work as expected, and ensure usage monitoring is capturing the data you need.

    **Set up monitoring early**

    Set up usage alerts during the pilot so you can understand consumption patterns before opening access widely. This helps you set realistic usage limits and identify potential cost issues before they become problems.

    **Learn and iterate**

    Gather feedback from pilot users about what's working and what's confusing. Use this input to refine your setup, update documentation, and improve onboarding materials. Monitor usage patterns closely for the first 30 days after full rollout—this is when you'll learn the most about how your organization uses the platform.
  </Tab>
</Tabs>

***

## Next steps

Once you've completed the setup steps above, you're ready to start building your AI workforce:

<CardGroup>
  <Card title="Build your first Agent" icon="robot" href="/get-started/quick-start-guide">
    Create an AI agent to automate tasks and workflows
  </Card>

  <Card title="Create a Workforce" icon="users-gear" href="/get-started/core-concepts/workforces">
    Build multi-agent teams for complex workflows
  </Card>

  <Card title="Start using Chat" icon="messages" href="/get-started/chat/introduction">
    Interact with AI agents through conversational chat
  </Card>

  <Card title="Browse the Marketplace" icon="store" href="/get-started/marketplace/introduction">
    Discover pre-built agents, tools, and templates
  </Card>

  <Card title="Add Knowledge" icon="brain" href="/get-started/core-concepts/knowledge">
    Connect your data sources to power your agents
  </Card>

  <Card title="Explore integrations" icon="puzzle-piece" href="/integrations/introduction">
    Connect your favorite tools and services
  </Card>
</CardGroup>

***

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="How long does Enterprise setup take?">
    Most organizations complete the core setup (Steps 1-5) in 2-3 hours. The optional steps (Data Retention, S3 streaming, integration controls) add another 30-60 minutes. If you're planning your organization structure carefully, allow additional time for that planning phase.
  </Accordion>

  <Accordion title="Can I skip steps in the recommended order?">
    While you can technically skip steps, we strongly recommend following the order. SSO and Directory Sync should always come first, as they affect how users access the platform. Skipping these steps may require you to reconfigure access later, which can disrupt your users.
  </Accordion>

  <Accordion title="What if I don't have an identity provider?">
    Enterprise features require an identity provider (Okta, Azure AD/Entra ID, or Google Workspace). If you don't have one, you'll need to set one up before using SSO and Directory Sync features. Most organizations already have an IDP for email and other services.
  </Accordion>

  <Accordion title="How do I get help during setup?">
    Enterprise customers have access to dedicated support. Contact your sales representative or [reach out to our support team](/docs/get-started/support) via your Slack channel or other support methods. We can also provide setup assistance for complex deployments.
  </Accordion>

  <Accordion title="What happens if I don't complete all setup steps?">
    You can use Relevance AI without completing all steps, but you won't have full Enterprise security and governance capabilities. We recommend completing at least Steps 1-4 (SSO, Directory Sync, Organization Structure, and RBAC) for a secure deployment.
  </Accordion>

  <Accordion title="Do I need Directory Sync if I only have a small team?">
    For teams under 20 users, Directory Sync is optional but still recommended. Manual user management works fine for small teams, but Directory Sync makes it easier to scale and ensures your access control stays in sync with your organization's identity provider.
  </Accordion>

  <Accordion title="Can I use Relevance AI without SSO?">
    Yes, but it's not recommended for Enterprise customers. SSO provides centralized access control, enhanced security, and is required for features like Directory Sync. For Enterprise deployments, SSO should be considered essential.
  </Accordion>
</AccordionGroup>

***

<Note>
  **Need help?** Enterprise customers have access to dedicated support channels. Contact your sales representative or [reach out to our support team](/docs/get-started/support) for assistance with your Enterprise setup.
</Note>


# Role-based access controls (RBAC)
Source: https://relevanceai.com/docs/enterprise/rbac

Enhancing security, collaboration, and control for Enterprise teams

<Warning>
  RBAC is gradually being rolled out to our Enterprise customers. If you have an Enterprise subscription with Relevance AI and do not have access to this feature yet, please reach out to your sales representative to share your interest in this feature.

  You will not be able to access this feature if you are not on an Enterprise subscription.
</Warning>

Role-based access controls (RBAC) in Relevance AI is a security and governance feature that allows organizations to manage user access based on predefined roles and responsibilities. It ensures that individuals only have access to the data, tools, and platform functions necessary for their role, supporting operational control, collaboration, and compliance across teams.

These new controls are designed to enhance security for our Enterprise clients by providing granular permissions at different levels.

There are three main levels: org-level roles (Owner, Admin, Member, Viewer), project-level roles (Admin, Editor, Member, Chat, Viewer), and asset-level roles (Admin, Member, Viewer).

Each role has specific permissions that dictate what users can do within the platform, such as managing credits, using agents, accessing integrations, and using Chat.

This means you can tailor access based on the needs of your team, ensuring that sensitive information and functionalities are only available to those who need them.

## Best Practices

Keep permissions simple. Assign roles based on who needs to build, who needs to view, who needs to chat, and who needs to govern.

### Organization Level

* **Owners** — have full control including billing and can delete the organization. Assign to 1-2 people maximum (CEO, CTO, or Head of Operations). Owners have all Admin permissions plus billing control.
* **Admins** — set up and manage teams, projects, and authentication. Usually your workspace or IT leads.
* **Members** — can only access projects they're assigned to. Cannot create projects at the organization level. Asset creation within projects is controlled by project-level permissions. Best for users who need access to specific projects but shouldn't manage organization-wide settings.
* **Viewers** — view-only access to audit logs, usage data, and compliance reports. Use this for stakeholders who need visibility without operational access.

### Project Level

| Role        | When to use it                                                                                                                           |
| ----------- | ---------------------------------------------------------------------------------------------------------------------------------------- |
| **Admins**  | Own the project setup and governance. They control permissions and authentication accounts.                                              |
| **Editors** | Build, edit, and run all assets in the project. Treat them as your core contributors.                                                    |
| **Members** | Can build their own assets but don't automatically see or edit others'. Best for independent work within shared projects.                |
| **Chat**    | Access [Relevance Chat](/docs/get-started/chat/introduction) only, cannot access the web app. Requires asset-level permissions to run agents. |
| **Viewers** | Can't build or edit. Add them only to the specific assets they need to see.                                                              |

<Tip>Keep admin/editor access limited to people actively managing or building. Everyone else should be a member or viewer by default.</Tip>

### Asset Level

* Project admins and editors automatically have full control.
* Members can run assets but can't modify them unless granted edit rights.
* Viewers can only see results or outputs — they can't run or trigger anything.

<Tip>Default to least privilege. Grant edit rights intentionally, not by habit.</Tip>

## Organization level controls

New organization members will be given viewer permissions by default. If invited, their role will be selected during invite.

### Roles

| **Role**   | **Capabilities**                                                                                                                                        |
| ---------- | ------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Owner**  | Full control of organization, billing, security, users and all projects                                                                                 |
| **Admin**  | Manage users, projects, organization-level API keys and OAuths                                                                                          |
| **Member** | Access only assigned projects. Cannot create projects at organization level. Asset creation within projects is controlled by project-level permissions. |
| **Viewer** | View-only access to agent and tool audit logs, usage data and compliance reports                                                                        |

### Permissions

| **Permission**                                         | **Owner** | **Admin** | **Member** | **Viewer** |
| :----------------------------------------------------- | :-------- | :-------- | :--------- | :--------- |
| Manage billing                                         | ✅         | ❌         | ❌          | ❌          |
| Manage organization settings (name, logo, domain etc.) | ✅         | ✅         | ❌          | ❌          |
| Manage organization users                              | ✅         | ✅         | ❌          | ❌          |
| Manage API keys & OAuths (Org-level connections)       | ✅         | ✅         | ❌          | ❌          |
| View global audit logs                                 | ✅         | ✅         | ❌          | ❌          |
| View all projects and agents                           | ✅         | ✅         | ❌          | ❌          |
| Delete any asset                                       | ✅         | ✅         | ❌          | ❌          |
| Edit project roles                                     | ✅         | ✅         | ❌          | ❌          |
| View credit information                                | ✅         | ✅         | ❌          | ❌          |
| Create projects                                        | ✅         | ✅         | ❌          | ❌          |
| View organization members / admins                     | ✅         | ✅         | ✅          | ✅          |

<Tip>
  If your only org Admin becomes unavailable, no one else can manage members, organization settings, org-level API keys and OAuth connections, or project roles. Assign the Admin role to more than one person so access isn't lost if someone leaves or needs to hand over control.
</Tip>

***

## Project level controls

Project admins will be able to set a users role upon invite. Organization admins can also set this for any project.

### Roles

| **Role**   | **Capabilities**                                                                                                                          |
| ---------- | ----------------------------------------------------------------------------------------------------------------------------------------- |
| **Admin**  | Manage users, agents, tools, knowledge and project-level integrations. Can create assets                                                  |
| **Editor** | Can edit and create assets, does not manage users                                                                                         |
| **Member** | Use shared assets, provide inputs and view outputs. Can create assets, private by default.                                                |
| **Chat**   | Access [Relevance Chat](/docs/get-started/chat/introduction) only - cannot access the web app. Requires asset-level permissions to run agents. |
| **Viewer** | View agents, tools, and knowledge outputs only, cannot run or edit anything                                                               |

<Info>
  Editor is a project-level role only and does not exist at organization or asset levels. Project Editors automatically have Admin permissions on all assets within the project.
</Info>

### Permissions

<Tip>
  Scroll horizontally to view all columns, including the Chat role permissions.
</Tip>

| **Permission**                             | **Admin** | **Editor** | **Member**       | **Viewer** | **Chat**  |
| :----------------------------------------- | :-------- | :--------- | :--------------- | :--------- | :-------- |
| Delete project                             | ✅         | ❌          | ❌                | ❌          | ❌         |
| Assign project roles to users              | ✅         | ❌          | ❌                | ❌          | ❌         |
| Manage project-level API keys & OAuths     | ✅         | ❌          | ❌                | ❌          | ❌         |
| Add personal OAuth accounts (dynamic auth) | ✅         | ✅          | ✅                | ✅          | ✅         |
| Delete agents                              | ✅         | ✅          | ❌                | ❌          | ❌         |
| View all assets by default                 | ✅         | ✅          | ❌                | ❌          | ❌         |
| Edit/run assets they did not create        | ✅         | ✅          | ❌                | ❌          | ❌         |
| View project activity logs                 | ✅         | ✅          | ❌                | ❌          | ❌         |
| Manage personal Relevance API key          | ✅         | ✅          | ✅                | ❌          | ❌         |
| Create assets                              | ✅         | ✅          | ✅                | ❌          | ❌         |
| View Project                               | ✅         | ✅          | ✅                | ✅          | ❌         |
| Access Web App                             | ✅         | ✅          | ✅                | ✅          | ❌         |
| Run a chat (LLM)                           | ✅         | ✅          | ✅                | ✅          | ✅         |
| Authenticate MCP                           | ✅         | ✅          | ✅                | ✅          | ✅         |
| MCP access mode                            | Full      | Full       | Full or run-only | Run-only   | Chat mode |

<Info>
  Project Viewer access grants read-only visibility to full asset configurations — including prompts, tools, and steps. There is no field-level redaction. If a Viewer cannot see an agent's internals, it's because they lack asset-level access entirely, not because their read access is limited to metadata.
</Info>

<Note>
  "Manage project-level API keys & OAuths" refers to shared, project-wide accounts only. All team members can add their own personal OAuth accounts when [dynamic authentication](/docs/enterprise/user-level-authentication) is enabled on a shared agent — this is not restricted to admins.
</Note>

### Chat Role Details

<Warning>
  The Chat role is a new feature being gradually rolled out to Enterprise customers. If you have an Enterprise subscription with Relevance AI and do not have access to this feature yet, please reach out to your sales representative to share your interest in this feature.
</Warning>

The **Chat** role is designed for users who only need to interact with AI agents through [Relevance Chat](/docs/get-started/chat/introduction) without accessing the main web application. Key characteristics:

<AccordionGroup>
  <Accordion title="Controlled agent visibility">
    Chat role users can only see and run agents they have Member or higher permissions on. They won't be able to see any other agents in Chat. This allows you to control exactly which agents each Chat user can access by assigning asset-level permissions.
  </Accordion>

  <Accordion title="Chat-only access">
    Users with this role are automatically redirected to Chat and cannot access the web app.
  </Accordion>

  <Accordion title="No asset creation">
    Cannot create new agents, tools, or knowledge bases.
  </Accordion>

  <Accordion title="LLM conversations">
    Can have conversations with LLMs and in-built Chat Agents directly without agents.
  </Accordion>

  <Accordion title="More powerful than Viewer">
    Can trigger Chat and interact with agents (with proper asset permissions).
  </Accordion>

  <Accordion title="Less powerful than Member">
    Cannot create assets or access the web app.
  </Accordion>

  <Accordion title="MCP access">
    Chat role users can authenticate the Relevance AI MCP server but are automatically placed into Chat mode — a restricted subset of tools limited to read access and running agents. Write and destructive tools are not available in the session. Dynamic (user-level) OAuth connections are not supported for Chat role users; if an agent requires dynamic authentication, assign the user a Member role in a dedicated project instead.
  </Accordion>
</AccordionGroup>

This role is ideal for end users, external collaborators, or team members who only need to use pre-built agents through the chat interface.

***

## Asset level controls

An asset is an Agent, Tool, Knowledge or Workforce.

Upon asset creation, the creator becomes the admin. An asset can have multiple admins (project admin is by default).

### Roles

| **Role**   | **Capabilities**                                                          |
| ---------- | ------------------------------------------------------------------------- |
| **Admin**  | Manage asset configuration, tools, knowledge and assign asset-level users |
| **Member** | Use asset only, provide inputs and view outputs                           |
| **Viewer** | View asset configuration and outputs only, cannot run or edit anything    |

### Permissions

| **Permission**                  | **Admin** | **Member** | **Viewer** |
| :------------------------------ | :-------- | :--------- | :--------- |
| Edit asset                      | ✅         | ❌          | ❌          |
| Delete asset                    | ✅         | ❌          | ❌          |
| Assign roles on asset           | ✅         | ❌          | ❌          |
| Assign auth per tool            | ✅         | ❌          | ❌          |
| Enable cloning/sharing of asset | ✅         | ❌          | ❌          |
| Create tasks on asset           | ✅         | ✅          | ❌          |
| View asset configuration        | ✅         | ✅          | ✅          |
| View asset outputs              | ✅         | ✅          | ✅          |
| View asset audit logs           | ✅         | ✅          | ✅          |

<Info>
  Asset Viewer access grants read-only visibility to full asset configurations — including prompts, tools, and steps. There is no field-level redaction. If a Viewer cannot see an agent's internals, it's because they lack asset-level access entirely, not because their read access is limited to metadata.
</Info>

<Warning>
  Adding someone as a Member on an agent does not give them access to the agent's tools. If a tool is set to **Only invited**, you must also add that person as a Member on each of those tools, otherwise the tool is silently dropped when they run the agent.
</Warning>

***

## Workforce Permissions

Workforces are treated as first-class assets in the RBAC system. However, running a workforce requires permissions at two levels:

1. **Workforce permission** - The user must have at least Member (run) access to the workforce itself
2. **Agent permissions** - The user must have at least Member (run) access to each agent used within the workforce

This means when sharing a workforce, you must:

* Share the workforce with the user at the desired permission level
* Share each agent in the workforce at a matching (or higher) permission level

### Example scenarios

| Goal                      | Workforce Role | Agent Roles          |
| ------------------------- | -------------- | -------------------- |
| User can run workforce    | Member         | Member on all agents |
| User can edit workforce   | Admin          | Admin on all agents  |
| User can view but not run | Viewer         | Viewer on all agents |

<Info>
  **Note:** If a user has run access to a workforce but lacks permissions on one or more agents, they will see a "Cannot run agent" warning in the workforce builder and will be unable to execute the workflow.
</Info>

<Tip>
  Learn more about [sharing workforces](/docs/build/workforces/share-your-workforce) as cloneable templates.
</Tip>

***

## Transitioning to RBAC

If your organization is being migrated from legacy permissions to RBAC, this section covers what changes, what stays the same, and what actions you need to take.

### Before RBAC (legacy permissions)

The legacy permission system had three roles — Admin, Editor, and Viewer — applied at the project and organization level only:

<CardGroup>
  <Card title="Admin" icon="user-tie-hair-long">
    Full read and write access to all assets and settings.
  </Card>

  <Card title="Editor" icon="pen-to-square">
    Read and write access to all datasets, knowledge sets, and agents. Could run agents and tools.
  </Card>

  <Card title="Viewer" icon="book-open-reader">
    Read access to all datasets, knowledge sets, and agents. Could run agents.
  </Card>
</CardGroup>

There was no asset-level granularity. All users with a given role had the same access to every asset in the project by default. Shared credentials (API keys, OAuths) applied project-wide.

### During the migration

The Relevance AI team handles the technical migration to RBAC. You do not need to manually migrate roles, but you should review and adjust assignments after migration is complete.

**How legacy roles map to RBAC roles by default:**

| Legacy role | Default RBAC role (project level) |
| ----------- | --------------------------------- |
| Admin       | Admin                             |
| Editor      | Editor                            |
| Viewer      | Viewer                            |

<Warning>
  The Viewer role changes significantly under RBAC. Legacy Viewers could run agents — RBAC Viewers cannot run anything and can only view assets they have been explicitly granted access to. Users who only need to interact with agents via chat should be assigned the Chat role, not Viewer.
</Warning>

After migration, review every user mapped to Viewer and determine whether they should be:

* **Chat** — for users who only need to use agents through [Relevance Chat](/docs/get-started/chat/introduction)
* **Viewer** — for users who genuinely only need read access to asset configurations and outputs
* **Member** — for users who need to run agents

### After RBAC is enabled

Once RBAC is active for your organization:

<AccordionGroup>
  <Accordion title="Asset-level permissions are enforced" icon="lock">
    Access to individual agents, tools, and knowledge bases is controlled separately from project-level roles.
  </Accordion>

  <Accordion title="Editors and Admins keep full access" icon="user-shield">
    Project Admin and Editor roles cascade to asset Admin on every asset in the project automatically.
  </Accordion>

  <Accordion title="Everyone else needs explicit grants" icon="ban">
    Members, Viewers, and Chat users receive a 403 error on any asset they have not been granted access to.
  </Accordion>

  <Accordion title="Credentials can be scoped per tool" icon="key">
    Instead of one set of project-wide credentials, asset admins can assign specific authentication accounts per tool.
  </Accordion>

  <Accordion title="Permissions live in the Share modal" icon="share-nodes">
    Assign access on each individual agent, tool, or knowledge base from its Share modal.
  </Accordion>
</AccordionGroup>

### Action items for admins

After RBAC is enabled, complete these steps to restore appropriate access for your team:

<Steps>
  <Step title="Review user roles immediately">
    Check that the default role mappings match your intended access levels, particularly for legacy Viewer accounts.
  </Step>

  <Step title="Assign asset-level permissions">
    Use the Share modal on each agent, tool, and knowledge base that non-Admin/Editor users need to access.
  </Step>

  <Step title="Use RBAC Groups for bulk assignment">
    If many users need the same access, create a group and assign asset permissions to the group rather than to individual users.
  </Step>

  <Step title="Audit chat-only users">
    Confirm that users who only need [Relevance Chat](/docs/get-started/chat/introduction) access are assigned the Chat project role, not Viewer.
  </Step>

  <Step title="Verify critical users can access required assets">
    Test access for key team members, especially those with Member or Viewer roles, before announcing the migration is complete.
  </Step>
</Steps>

***

## Permission inheritance and cascading

Project roles fall into two groups: those that cascade Admin access to every asset automatically, and those that start with no asset access until it is explicitly granted.

<CardGroup>
  <Card title="Cascades to asset Admin" icon="circle-check">
    **Admin** and **Editor** automatically receive Admin access on every asset in the project — no per-asset setup required.
  </Card>

  <Card title="Starts with no access" icon="lock">
    **Member**, **Viewer**, and **Chat** receive nothing by default. Each asset must be shared with them explicitly, or they get a 403 error.
  </Card>
</CardGroup>

<Note>
  To grant access at scale, assign asset permissions to an RBAC Group rather than to each user individually.
</Note>

***

## Technical implementation notes

The following is relevant primarily for API users and developers integrating directly with the authorization system.

<Tabs>
  <Tab title="Member/Operator">
    The "Member" role label in the UI maps to the `operator` role in the authorization system. When querying permissions via the API, use `operator` rather than `member`.
  </Tab>

  <Tab title="OpenFGA">
    The platform uses OpenFGA for authorization. Organizations not yet migrated to RBAC are still served by a legacy authorization layer.
  </Tab>

  <Tab title="Embeddable agents">
    Agents shared via public embed links (Chat UI, Chat Widget) may use the legacy authorization layer. Test permissions explicitly when deploying embedded agents.
  </Tab>
</Tabs>

***

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can I access role-based access controls without upgrading to Enterprise?">No. This feature is available for Enterprise subscriptions only.</Accordion>
  <Accordion title="I'm on an Enterprise subscription but do not have access to this feature yet. How do I get access?">This feature is gradually being rolled out to Enterprise customers. Please reach out to your sales representative to express your interest in receiving this feature.</Accordion>
  <Accordion title="How do I limit someone to only access Chat?">To limit a user to only access [Relevance Chat](/docs/get-started/chat/introduction), assign them the **Chat** role at the project level. Users with the Chat role are automatically redirected to Chat and cannot access the main web application. They can still interact with agents and have LLM conversations, but will need appropriate asset-level permissions (Member or higher) on specific agents to run them in Chat. Chat role users who authenticate the MCP server are automatically restricted to Chat mode — they cannot create, edit, or delete assets via MCP. This restriction is enforced server-side.</Accordion>
  <Accordion title="Is there a role that allows a user to approve escalations or interact with agent tasks, but not edit the agent?">Yes, assigning a user the Asset Member role for a specific agent allows them to create tasks on the agent, approve/reject escalations, and view outputs, but not edit or delete the agent's configuration.</Accordion>
  <Accordion title="How many organization admins should I have?">At least two. Besides the Owner, Admin is the only organization role that can manage users, settings, API keys, and project roles, so a single Admin is a continuity risk if they leave or are unavailable. Keeping two or more preserves access while still maintaining clear accountability.</Accordion>
</AccordionGroup>


# Directory Sync and Groups
Source: https://relevanceai.com/docs/enterprise/rbac-groups

Sync your identity provider groups to Relevance AI for bulk user management

<Warning>
  RBAC Groups is gradually being rolled out to our Enterprise customers. If you have an Enterprise subscription with Relevance AI and do not have access to this feature yet, please reach out to your sales representative to share your interest in this feature.

  You will not be able to access this feature if you are not on an Enterprise subscription.
</Warning>

RBAC Groups enables organizations to connect their identity provider (IDP) such as Entra ID or Okta directly to Relevance AI. This allows bulk assignment of users via groups to projects and assets, as well as setting their permissions. Instead of manually assigning hundreds of users individually, organizations can assign entire groups (like "Marketing Team" or "Sales Operations") to projects and agents.

This feature dramatically simplifies user management for large teams by leveraging your existing organizational structure from your identity provider.

## Key Features

<Columns>
  <Card title="Direct IDP Integration" icon="link">
    Connect identity providers like Entra ID or Okta directly to Relevance AI through WorkOS
  </Card>

  <Card title="Bulk User Assignment" icon="users">
    Assign entire groups to projects and assets instead of managing individual users
  </Card>

  <Card title="Group-Level Permissions" icon="shield-halved">
    Set permissions at the group level for efficient access control across your organization
  </Card>

  <Card title="Automatic Synchronization" icon="arrows-rotate">
    Groups sync from your IDP approximately every 60 minutes
  </Card>

  <Card title="Read-Only Groups" icon="lock">
    IDP groups are managed at the source and cannot be edited within Relevance AI
  </Card>

  <Card title="Unified Interface" icon="grid">
    Manage group assignments alongside individual users in the same UI
  </Card>
</Columns>

<Info>
  Groups are managed through your identity provider and sync automatically to Relevance AI. Changes made in your IDP will be reflected in Relevance AI approximately every 60 minutes.
</Info>

## Setting Up RBAC Groups

To use RBAC Groups, your organization must first connect your identity provider through WorkOS. WorkOS provides secure integrations with major identity providers including Entra ID (formerly Azure AD) and Okta.

<Note>
  For detailed setup instructions, see the [Setting Up Directory Sync](#setting-up-directory-sync) section below.
</Note>

Once your IDP is connected, groups from your identity provider will automatically sync to Relevance AI and become available for assignment.

## Organization-Level Groups

<img alt="Organization Groups Overview" />

<Note>
  Organization-level groups are only accessible to organization admins and owners.
</Note>

At the organization level, you can view all groups that have been synced from your identity provider. This provides a centralized view of your organizational structure within Relevance AI.

### Accessing organization groups

To view organization-level groups:

1. Click **Settings** in the sidebar
2. Select **Organization**
3. Click the **User Groups** tab

The groups page displays the name, source (e.g., Entra ID, Okta), and member count for each group synced from your identity provider. Note that member counts only reflect users who have already logged into Relevance AI—not your total IDP group size. The count increases as more users sign in for the first time.

To view who is in a group and the permissions they have, simply click on the group. You'll see the complete membership list and group details.

<Info>
  IDP groups are read-only in Relevance AI. To modify group membership, make changes in your identity provider and they will sync automatically.
</Info>

## Project-Level Groups

<img alt="Adding Groups to Projects" />

Groups can be assigned to projects to give all group members access to that project. This is particularly useful for onboarding entire teams or departments to specific projects.

When adding a group to a project, you'll assign them a [project-level role](/docs/enterprise/rbac#project-level-controls) (Admin, Editor, Member, Chat, or Viewer) which determines what they can do within that project.

### Adding groups to projects

<div>
  <iframe title="Add groups to projects" />
</div>

To add a group to a project:

1. Head to the **Invite to project** screen (either through Settings or by clicking your profile picture in Relevance AI)
2. Click **Assign Group**
3. Select the group(s) you want to add
4. Select the role you want the group to have (Admin, Editor, Member, Chat, or Viewer)

<Warning>
  When you add a group to a project, all members of that group will be added to the project with the specified role. If any group members are not already part of the project, they will be automatically added.
</Warning>

## Asset-Level Groups

<img alt="Adding Groups to Assets" />

Groups can also be assigned directly to individual assets (agents, tools, knowledge bases, or workforces). This provides granular control over who can access and use specific resources.

When adding a group to an asset, you'll assign them an [asset-level role](/docs/enterprise/rbac#asset-level-controls) (Admin, Member, or Viewer) which determines their permissions for that specific asset.

### Adding groups to assets

<div>
  <iframe title="Add groups to assets" />
</div>

To add a group to an asset:

1. Open the asset you want to set permissions on
2. Click **Share**
3. Click **Groups**
4. Choose the group you want to add
5. Set the permission for the group (Admin, Member, or Viewer)

<Info>
  If a group is not already part of the asset's parent project, adding it to the asset will automatically add all group members to the project as well.
</Info>

## How Group Synchronization Works

RBAC Groups maintains synchronization with your identity provider to ensure access control stays current with your organizational structure.

<Columns>
  <Card title="Hourly Sync" icon="clock">
    Groups automatically sync from your IDP approximately every 60 minutes
  </Card>

  <Card title="New User Sign-Up" icon="user-plus">
    Group permissions are applied automatically during the sign-up process for new users
  </Card>

  <Card title="Background Processing" icon="gears">
    Syncs happen automatically without user intervention
  </Card>
</Columns>

### New User Sign-Up Process

When onboarding new users with Directory Sync and Groups enabled, follow this workflow:

<Steps>
  <Step title="Enable SSO for the user">
    Add the user to your identity provider and assign them access to the Relevance AI SSO application.
  </Step>

  <Step title="Verify required attributes">
    Ensure the user has the following required attributes correctly configured in your identity provider:

    * First name
    * Last name
    * Email address

    These attributes are required for successful SSO sign-up. See the [SSO Setup documentation](/docs/enterprise/sso-setup) for more details.
  </Step>

  <Step title="Add user to groups">
    Assign the user to the appropriate groups in your identity provider.
  </Step>

  <Step title="Wait for sync">
    The sync worker runs approximately every 60 minutes (could be slightly longer). Wait for the next sync cycle to complete.
  </Step>

  <Step title="User signs up">
    The user can now sign up to Relevance AI through SSO. Their group-based permissions will be automatically applied during the sign-up process.
  </Step>

  <Step title="User appears in Relevance AI">
    Only after the user completes sign-up and is officially created in the Relevance AI platform will their email address appear in the users list within groups in Relevance AI.
  </Step>
</Steps>

<Warning>
  **Important**: Users will not appear in your Relevance AI groups or user lists until they have completed their first SSO sign-up. Group membership and permissions are configured during the sign-up process, not before.
</Warning>

### Adding users to groups

When a user is added to an IDP group:

1. The change is detected during the next sync (approximately every 60 minutes)
2. The user automatically gains access to all projects and assets assigned to that group
3. The user inherits the group's role and permissions for each resource
4. No manual intervention is required in Relevance AI

<Note>
  For users not yet signed up to the platform, their group permissions will be applied during the sign-up process. Once they sign up, the user will be reflected as a member in the group.
</Note>

### Removing users from groups

When a user is removed from an IDP group:

1. The change is detected during the next sync (approximately every 60 minutes)
2. The user's group-based access to projects and assets is revoked
3. If the user has individual (direct) permissions, those are retained
4. If the user only had group-based access, they lose access to the resource

<Warning>
  Users removed from IDP groups lose their group-based permissions immediately upon sync. Ensure you understand the impact before removing users from groups in your identity provider.
</Warning>

## User Deprovisioning

User deprovisioning automatically removes user access when they are deprovisioned in your Identity Provider (IdP). This ensures that access control stays synchronized with your organization's user lifecycle management.

### When is a user deprovisioned?

Any of the following actions in your Identity Provider (IdP) will deprovision a user (they lose access via Directory Sync):

* Deprovisioning the user from the SCIM app in the IdP
* Removing the user from the group that grants access to the SCIM app (when that group controls provisioning to the app)
* Marking the user as inactive in the IdP
* Deleting the user in the IdP

<Warning>
  If your app is assigned to only one group, removing a user from that group (or removing the group itself) will deprovision the user. **Best practice:** Assign the Relevance app to a dedicated group (e.g., "Relevance App Users") to ensure that removing users from other role-based groups like "Marketing" or "Engineering" does not accidentally deprovision them from the application.
</Warning>

## Email address updates for IdP-managed users

Email address changes are not currently supported for SSO/SCIM managed users. When a user's email address changes in your identity provider, Relevance AI treats it as a new user — the original account's permissions, group memberships, owned assets, and history are not transferred to the new email address.

### What happens when an email changes

If you rename an existing IdP user's email address, the Directory Sync process provisions a new Relevance AI account for the new email address. The original account remains active until deprovisioned separately, and no data or permissions are transferred between the two accounts.

<Warning>
  Do not rename an existing IdP user to change their email address. This creates a new account rather than updating the existing one, and the original account's assets and history cannot be automatically transferred.
</Warning>

### Workaround for email address changes

To move a user to a new email address while preserving their access:

<Steps>
  <Step title="Create a new user in your IdP">
    Create a new identity provider user with the new email address. Do not rename the existing user.
  </Step>

  <Step title="Assign the new user to SSO and Directory Sync groups">
    Add the new user to the Relevance AI SSO application and assign them to the appropriate Directory Sync groups in your identity provider.
  </Step>

  <Step title="User signs in with the new email">
    The new user signs in to Relevance AI via SSO. Their group-based permissions are applied automatically during sign-up.
  </Step>

  <Step title="Remove the old user from your IdP">
    Deprovision the original user in your identity provider. This removes their access from Relevance AI automatically. See [User deprovisioning](#user-deprovisioning) for details.
  </Step>
</Steps>

<Note>
  Assets, agents, tools, or history owned by the original account are not automatically transferred to the new account. If you need to transfer ownership, do so manually before deprovisioning the old user.
</Note>

## Setting Up Directory Sync

To use RBAC Groups, you need to configure directory sync between your identity provider and Relevance AI through WorkOS. This guide walks you through the complete setup process.

### Prerequisites

Before setting up Directory Sync, you must first configure [Single Sign-On (SSO)](/docs/enterprise/sso-setup) for your organization. Directory Sync builds on top of SSO to enable automated user provisioning and group management.

### Setup Process

<Steps>
  <Step title="Enable the Groups feature">
    Contact your dedicated sales representative or reach out to our support team via your Slack channel or [other support methods](/docs/get-started/support) to enable the "Groups" feature for your organization. Wait for confirmation before proceeding to the next step.
  </Step>

  <Step title="Configure your identity provider">
    Follow the WorkOS integration guide for your specific identity provider:

    * **Microsoft Entra ID (Azure AD)**: [Entra ID SCIM Integration](https://workos.com/docs/integrations/entra-id-scim)
    * **Google Workspace**: [Google Directory Sync Integration](https://workos.com/docs/integrations/google-directory-sync)
    * **Okta**: [Okta SCIM Integration](https://workos.com/docs/integrations/okta-scim)
    * **JumpCloud**: [JumpCloud SCIM Integration](https://workos.com/docs/integrations/jumpcloud-scim)

    <Note>
      Some UI elements in the WorkOS documentation may differ from their current interface.
    </Note>

    <Warning>
      **Important for Microsoft Entra ID users:** The provisioning service cannot read or provision users in nested groups—only immediate group members are supported. This limitation also affects single sign-on. **Best practice:** Directly assign users to groups rather than using nested group structures. [Learn more](https://learn.microsoft.com/en-us/entra/identity/app-provisioning/how-provisioning-works)
    </Warning>
  </Step>

  <Step title="Confirm setup with Relevance AI">
    After completing the configuration in your identity provider, contact your dedicated sales representative or reach out to our support team via your Slack channel or [other support methods](/docs/get-started/support) to confirm setup. We will monitor and validate that your sync is working correctly.
  </Step>
</Steps>

### Current Limitations

The following features are not yet supported in the current version of RBAC Groups:

<AccordionGroup>
  <Accordion title="Assigning Individual Roles to Group Members">
    Users who have access only through groups (with no direct access) will not appear in the project user list. They will still have access based on their group membership.

    If you want to give a specific user a different role than their group, you must:

    1. Invite them directly to the project (even if they already have access through a group)
    2. Assign them the desired role

    <Note>
      The user will always receive the **highest permission level** from all their sources. For example, if they have Admin access through a group and you assign them Viewer access directly, they will retain Admin permissions because the highest permission always prevails.
    </Note>
  </Accordion>

  <Accordion title="Group Membership Visibility">
    Group membership is only visible on the organization settings page (requires admin access), not on the project invite page.
  </Accordion>

  <Accordion title="Empty Agent Folders">
    Folders without accessible agents will still be visible to users, even if they cannot access any agents within them.
  </Accordion>

  <Accordion title="Permission Downgrade">
    Users can downgrade their own group's permissions without warning or confirmation. Exercise caution when assigning admin permissions.
  </Accordion>

  <Accordion title="Email address updates for IdP-managed users">
    Changing a user's email address in your identity provider creates a new Relevance AI account rather than updating the existing one. Permissions, group memberships, and owned assets are not transferred automatically. See [Email address updates for IdP-managed users](#email-address-updates-for-idp-managed-users) for the recommended workaround.
  </Accordion>
</AccordionGroup>

<Tip>
  For questions or assistance with directory sync setup, contact your dedicated sales representative or reach out to our support team via your Slack channel or [other support methods](/docs/get-started/support).
</Tip>

***

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can I access RBAC Groups without upgrading to Enterprise?">
    No. RBAC Groups is available for Enterprise subscriptions only.
  </Accordion>

  <Accordion title="Which identity providers are supported?">
    RBAC Groups supports major identity providers including Entra ID (formerly Azure AD) and Okta through WorkOS. For a complete list of supported providers, see the [WorkOS integrations documentation](https://workos.com/docs/integrations).
  </Accordion>

  <Accordion title="Can I assign different roles to the same group at different levels?">
    Yes. A group can have different roles at the project level versus the asset level. For example, a group could be Members at the project level but Admins on a specific agent within that project.
  </Accordion>

  <Accordion title="What happens if a user is in multiple groups with different roles?">
    If a user belongs to multiple groups assigned to the same resource with different roles, they receive the highest permission level among all their group memberships and individual assignments.
  </Accordion>

  <Accordion title="Can I use RBAC Groups with non-SSO users?">
    RBAC Groups requires users to authenticate via SSO through your identity provider. Users who sign in with email/password or other non-SSO methods cannot be managed through IDP groups.
  </Accordion>

  <Accordion title="What's the difference between RBAC and RBAC Groups?">
    RBAC (Role-Based Access Control) provides the permission framework and roles at organization, project, and asset levels. RBAC Groups extends this by allowing you to assign those roles to entire groups of users at once, rather than individually.
  </Accordion>

  <Accordion title="What happens if a user's email address changes in our identity provider?">
    Email address changes are not currently supported for IdP-managed users. Changing a user's email in your identity provider creates a new Relevance AI account rather than updating the existing one — permissions, group memberships, and owned assets are not transferred. See [Email address updates for IdP-managed users](#email-address-updates-for-idp-managed-users) for the step-by-step workaround.
  </Accordion>
</AccordionGroup>


# Single Sign-On (SSO) Setup
Source: https://relevanceai.com/docs/enterprise/sso-setup

Configure secure authentication with your identity provider using SAML 2.0 or OIDC protocols

<Note>
  SSO is available for Enterprise customers only. If you're not on an Enterprise subscription, [contact our sales team](https://relevanceai.com/contact) to learn more.
</Note>

Single Sign-On (SSO) enables your organization to authenticate users through your identity provider, providing centralized access control, enhanced security, and a seamless login experience.

## Supported Identity Providers

Relevance AI supports SSO through a wide range of identity providers via WorkOS, including:

* **Okta**
* **Azure AD / Microsoft Entra ID**
* **Google Workspace**
* **Auth0**
* And many more

We support both SAML 2.0 and OIDC protocols. For a complete list of supported identity providers, see [WorkOS Integrations](https://workos.com/docs/integrations).

***

## Prerequisites

Before setting up SSO, ensure you have:

* Your organization's email domain
* IT admin contact at your organization
* Admin access to your identity provider

<Warning>
  Setting up SSO requires coordination between your Relevance AI representative and your IT administrator. Plan to have your IT admin available during setup.
</Warning>

### Required User Attributes

Before completing SSO setup, ensure that your identity provider is configured to send the following required user attributes:

| Attribute  | Accepted Field Names        |
| ---------- | --------------------------- |
| First Name | `firstName` or `first_name` |
| Last Name  | `lastName` or `last_name`   |
| Email      | `email`                     |

These attributes must be correctly mapped in your identity provider for users to successfully sign up to Relevance AI via SSO. If these attributes are missing or incorrectly mapped, user sign-up will fail or result in incomplete user profiles.

Your IT administrator should verify these attribute mappings during the SSO configuration process in WorkOS.

<Note>
  Email is used as the unique account identifier in Relevance AI. If a user's email address changes in your identity provider, Relevance AI provisions a new account rather than updating the existing one — permissions, group memberships, and owned assets are not transferred. See [Email address updates for IdP-managed users](/docs/enterprise/rbac-groups#email-address-updates-for-idp-managed-users) for details and the recommended workaround.
</Note>

***

## Setup Process

<Steps>
  <Step title="Contact Your Relevance AI Representative" icon="handshake">
    Reach out to your Relevance AI representative to initiate SSO setup. Provide:

    * Your organization's email domain
    * Contact information for your IT administrator
    * Your preferred identity provider (e.g., Okta, Azure AD, Google Workspace, Auth0, or others)
  </Step>

  <Step title="Receive Setup Link" icon="envelope">
    Your IT administrator will receive an email with a setup link and instructions for completing the SSO configuration through WorkOS.

    <Info>
      Your IT admin should refer to the WorkOS integration guide for your specific identity provider. Visit [WorkOS Integrations](https://workos.com/docs/integrations) and select your provider for detailed setup instructions. Don't worry if you select the wrong one - the configuration can be redone.
    </Info>
  </Step>

  <Step title="Complete SSO Configuration" icon="gear">
    Your IT administrator will:

    1. Click the setup link provided in the email
    2. Complete the SSO configuration through the WorkOS interface
    3. Follow the WorkOS process: [Using Admin Portal](https://workos.com/docs/admin-portal/c-using-admin-portal)

    <Warning>
      **Important:** When clicking the setup link, your IT admin will be directed to your organization's WorkOS environment (created by Relevance AI). They should **not** create a new WorkOS account. If they see a billing screen or are prompted to pay, they are in the wrong organization. Contact your Relevance AI representative if this occurs.
    </Warning>

    <Note>
      Admin permissions in your identity provider are required to complete this step.
    </Note>
  </Step>

  <Step title="Test SSO Connection" icon="vial">
    Your IT administrator should test the SSO connection to ensure it works properly.
  </Step>

  <Step title="Assign User Access" icon="users">
    Your IT administrator will assign SSO access to the requested users in your identity provider system.
  </Step>

  <Step title="Confirm Setup" icon="check">
    Contact your Relevance AI representative to confirm successful setup.
  </Step>
</Steps>

***

## Next Steps

Once SSO is configured, you can enhance your organization's access management with these additional Enterprise features:

<CardGroup>
  <Card title="Role-Based Access Controls (RBAC)" icon="shield-halved" href="/enterprise/rbac">
    Control user permissions at organization, project, and asset levels with granular role-based access controls.
  </Card>

  <Card title="Directory Sync and Groups" icon="arrows-rotate" href="/enterprise/rbac-groups">
    Sync your identity provider groups to Relevance AI for bulk user management and automated provisioning.
  </Card>
</CardGroup>

***

## Troubleshooting

<AccordionGroup>
  <Accordion title="Setup link has expired">
    If the initial setup link expires, contact your Relevance AI representative to generate a new setup link. Confirm access with your IT administrator once you receive the new link.
  </Accordion>
</AccordionGroup>

***

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="How long does SSO setup take?">
    The setup process typically takes 30-60 minutes once your IT administrator receives the setup link.
  </Accordion>

  <Accordion title="What is WorkOS?">
    WorkOS is the SSO infrastructure provider that handles the secure connection between Relevance AI and your identity provider. You don't need a WorkOS account - your IT admin will use it only during the setup process. When your IT admin clicks the setup link, they'll be directed to your organization's WorkOS environment that Relevance AI has already created. They should not see any billing screens or be prompted to create a new account.
  </Accordion>

  <Accordion title="Can I use multiple identity providers?">
    No, each Relevance AI organization can only connect to one identity provider at a time. Contact your sales representative if you need to discuss alternative options.
  </Accordion>

  <Accordion title="What happens if I need to change identity providers?">
    Contact your Relevance AI representative. They will coordinate disabling your current SSO configuration and setting up a new one with your new identity provider.
  </Accordion>
</AccordionGroup>


# Event Streaming for Observability (OpenTelemetry)
Source: https://relevanceai.com/docs/enterprise/streaming-events

Stream audit logs and execution traces to your own infrastructure for compliance, analytics, and observability

Stream audit logs and execution traces to your own infrastructure for compliance, analytics, and observability.

## Overview

Relevance AI delivers events in [**OpenTelemetry (OTEL)**](https://opentelemetry.io/) format - an open, vendor-neutral standard for telemetry data. This provides several advantages:

* **Interoperability**: OTEL is supported by most major observability platforms, making it easy to route data to your existing tools
* **Future-proof**: As an [industry standard backed by CNCF](https://www.cncf.io/projects/opentelemetry/), OTEL ensures your data pipelines won't be locked into proprietary formats
* **Rich semantics**: Built-in support for traces, logs, and metrics with standardized attribute naming conventions
* **Correlation**: Trace IDs link related events across agent invocations, LLM calls, and workforce executions

You don't need to run an OTEL collector to use this feature. Events are delivered directly to your S3 bucket where you can:

* Query them directly using Athena, BigQuery, or similar tools
* Ingest into your data lake (Snowflake, Databricks, etc.)
* Forward to any OTEL-compatible backend for visualization and alerting

<Note>Enterprise customers can enable PII redaction to automatically protect sensitive information in logs. Contact your Account Manager to learn more.</Note>

<Note>S3 is currently the only supported destination. Support for direct OTEL collector endpoints is on our roadmap.</Note>

<Note>Because events are delivered to your own bucket, you control their retention through your bucket's lifecycle policy. On the Relevance side, telemetry only passes through a temporary staging bucket that is automatically purged after 30 days. This is independent of the [Data Retention](/docs/enterprise/data-retention) policy, which governs platform data such as conversations, tool runs, and versions.</Note>

***

## Setup

### Prerequisites

* AWS account with permissions to create S3 buckets and bucket policies
* Relevance AI Enterprise plan

### 1. Create an S3 bucket

Create a bucket in the **same AWS region** as your Relevance data:

| Relevance Region | AWS Region                |
| ---------------- | ------------------------- |
| Australia        | `ap-southeast-2` (Sydney) |
| Europe           | `eu-west-2` (London)      |
| US               | `us-east-1` (N. Virginia) |

### 2. Configure bucket policy

Add this policy to allow Relevance to write events to your bucket:

```json theme={null}
{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Sid": "AllowRelevanceEventConsumer",
      "Effect": "Allow",
      "Principal": {
        "AWS": "RELEVANCE_EVENT_CONSUMER_ROLE_ARN"
      },
      "Action": ["s3:PutObject", "s3:GetBucketLocation"],
      "Resource": [
        "arn:aws:s3:::YOUR_BUCKET_NAME",
        "arn:aws:s3:::YOUR_BUCKET_NAME/*"
      ]
    }
  ]
}
```

Replace:

* `YOUR_BUCKET_NAME`: Your S3 bucket name
* `RELEVANCE_EVENT_CONSUMER_ROLE_ARN`: Contact your Relevance team for the region-specific IAM role ARN

### 3. Provide configuration to Relevance

Send your Account Manager or support team:

* **Bucket Name**: Your S3 bucket name
* **Region**: AWS region of the bucket
* **Prefix**: S3 prefix for events (e.g., `relevance-events/`)

***

## PII Redaction (Enterprise feature)

PII (Personally Identifiable Information) redaction is an org-level feature that automatically scrubs sensitive information like email addresses, phone numbers, credit card numbers, and names before your event data leaves the platform and is delivered to your S3 destination.

PII redaction applies at the point of data export, specifically when telemetry and audit logs are being written to your S3 bucket. It does not apply to live agent conversations in real-time or data stored internally on Relevance AI's side. Think of it as a "scrub before delivery" mechanism for your downstream data pipeline.

<Note>PII redaction is a contractual Enterprise feature. Contact your Account Manager to enable this capability for your organization.</Note>

### What gets redacted

There are two layers of redaction:

**1. Structured fields (always scrubbed automatically)**

The following fields are redacted from exported data regardless of whether PII scrubbing is explicitly enabled:

* User email addresses
* IP addresses
* Device info

**2. Unstructured text (scrubbed when PII redaction is enabled for your org)**

When PII redaction is enabled, the following fields are scanned and scrubbed using a Presidio-powered ML engine:

* LLM input messages (what you send to an agent/tool)
* LLM output messages (what the agent responds)
* System instructions

### Supported PII entities

The following entity types can be detected. By default, all supported entities are targeted. You can optionally limit detection to specific entity types via configuration.

<CardGroup>
  <Card title="EMAIL_ADDRESS" icon="envelope" />

  <Card title="PHONE_NUMBER" icon="phone" />

  <Card title="PERSON" icon="user" />

  <Card title="CREDIT_CARD" icon="credit-card" />

  <Card title="US_SSN" icon="id-card" />

  <Card title="IP_ADDRESS" icon="globe" />

  <Card title="IBAN_CODE" icon="building-columns" />

  <Card title="URL" icon="link" />

  <Card title="DATE_TIME" icon="calendar" />

  <Card title="LOCATION" icon="location-dot" />
</CardGroup>

### Redaction actions

When PII is detected, the following actions can be applied:

| Action    | Description                                                                                        |
| --------- | -------------------------------------------------------------------------------------------------- |
| `replace` | Substitutes the detected value with a placeholder like `<EMAIL_ADDRESS>`. **This is the default.** |
| `mask`    | Replaces the detected value with `****`                                                            |
| `redact`  | Removes the detected value entirely                                                                |
| `hash`    | Hashes the detected value (one-way)                                                                |

### Prerequisites

Before configuring PII redaction, ensure:

* OTEL log shipping is enabled for your organization
* The `jsonl_to_otel_json` transformation is applied
* Your Enterprise plan includes PII redaction (contact your Account Manager)

### Configuration

| Field             | Type      | Required | Description                                                                                 |
| ----------------- | --------- | -------- | ------------------------------------------------------------------------------------------- |
| `enabled`         | boolean   | Yes      | Enable or disable PII redaction                                                             |
| `action`          | string    | Yes      | Action to take on detected PII. Options: `"replace"`, `"redact"`, `"mask"`, or `"hash"`     |
| `entities`        | string\[] | No       | Specific PII entity types to target. If omitted, all supported entities are used            |
| `target_fields`   | string\[] | No       | Specific data fields to scan for PII                                                        |
| `score_threshold` | float     | No       | Minimum confidence level for PII detection (0.0–1.0). Higher values require more confidence |

<Warning>There is currently no self-serve UI to toggle PII redaction. It is configured at the organization level by the Relevance AI team. Contact your Account Manager to have it enabled.</Warning>

***

## Prompt injection detection (Enterprise feature)

Prompt injection detection is an org-level feature that scans incoming user input for prompt injection attempts. When an attempt is detected, the result is recorded on the agent's execution trace so you can monitor and alert on it in your own observability stack.

Detection runs as a post-processing scan that is independent of the underlying LLM model. It flags and records attempts; it does not block or rewrite the user's message.

<Note>Prompt injection detection requires event streaming to be enabled, because results surface only on the OTEL traces delivered to your S3 bucket. It is a contractual Enterprise feature with no self-serve UI — contact your Account Manager to enable it for your organization.</Note>

### Where it surfaces

Detection results are added as attributes on the [`invoke_agent`](#invoke_agent) span:

| Attribute                                  | Type    | Description                                                       |
| ------------------------------------------ | ------- | ----------------------------------------------------------------- |
| `relevance_ai.prompt_injection.detected`   | boolean | Whether a prompt injection attempt was detected in the user input |
| `relevance_ai.prompt_injection.confidence` | double  | Confidence score for the detection (0.0–1.0)                      |

These attributes are only present when prompt injection detection is enabled for your organization. Query them in your observability tool the same way you would any other span attribute.

***

## Delivery format

**Format**: Gzipped JSON following the OpenTelemetry JSON specification.

**File path pattern**:

```
{prefix}/customer-otel-{type}-formatted/org_id={org_id}/dt={YYYY-MM-DD}/year={YYYY}/month={MM}/day={DD}/hour={HH}/minute={MM}/{type}_{org_id}_{timestamp}_{uuid}.json.gz
```

**Event types**:

* `logs` - Audit logs for administrative and lifecycle events
* `traces` - Execution traces for agents, workforces, and LLM completions

***

## Logs

Audit logs provide a complete record of activity across your organization for security monitoring, compliance reporting, and operational visibility. Events capture who did what, when, and from where. Enterprise customers can enable PII redaction to automatically protect sensitive information in logs.

<Note>We're actively expanding our event coverage. If there are specific operations or attributes you'd like to see, let your Account Manager know.</Note>

### Structure

```json theme={null}
{
  "resourceLogs": [{
    "scopeLogs": [{
      "logRecords": [...]
    }]
  }]
}
```

### Base attributes

Included on all log records:

| Attribute                      | Type   | Required | Description                            |
| ------------------------------ | ------ | -------- | -------------------------------------- |
| `relevance_ai.organization_id` | string | Yes      | Organization ID                        |
| `relevance_ai.project_id`      | string | Yes      | Project ID                             |
| `relevance_ai.user_id`         | string | No       | User ID that performed the action      |
| `relevance_ai.user_email`      | string | No       | Email of the user                      |
| `relevance_ai.user_type`       | string | No       | Type of user (`user`, `api_key`, etc.) |
| `relevance_ai.ip_address`      | string | No       | IP address of the request              |
| `relevance_ai.device_info`     | string | No       | User agent / device information        |

### Log record properties

| Property           | Type   | Description                                  |
| ------------------ | ------ | -------------------------------------------- |
| `timeUnixNano`     | int64  | Timestamp in nanoseconds since Unix epoch    |
| `severityNumber`   | int    | OTEL severity level (9 = Info)               |
| `severityText`     | string | `"Info"`                                     |
| `body.stringValue` | string | The event name                               |
| `attributes`       | array  | Key-value pairs with base + event attributes |

### Supported events

### Agent events

**`agent_created`** - Emitted when a new agent is created (from scratch, cloned, or duplicated).

| Attribute                                   | Type   | Required | Description                   |
| ------------------------------------------- | ------ | -------- | ----------------------------- |
| `relevance_ai.event.agent_id`               | string | Yes      | ID of the newly created agent |
| `relevance_ai.event.cloned_from_agent_id`   | string | No       | Source agent ID if cloned     |
| `relevance_ai.event.cloned_from_region`     | string | No       | Source region if cloned       |
| `relevance_ai.event.cloned_from_project_id` | string | No       | Source project if cloned      |

**`agent_updated`** - Emitted when an agent's configuration is updated directly (outside draft/publish workflow).

| Attribute                     | Type   | Required | Description             |
| ----------------------------- | ------ | -------- | ----------------------- |
| `relevance_ai.event.agent_id` | string | Yes      | ID of the updated agent |

**`agent_deleted`** - Emitted when an agent is permanently deleted.

| Attribute                     | Type   | Required | Description             |
| ----------------------------- | ------ | -------- | ----------------------- |
| `relevance_ai.event.agent_id` | string | Yes      | ID of the deleted agent |

**`agent_draft_saved`** - Emitted when work-in-progress changes are saved to an agent.

| Attribute                       | Type   | Required | Description                   |
| ------------------------------- | ------ | -------- | ----------------------------- |
| `relevance_ai.event.agent_id`   | string | Yes      | ID of the agent being edited  |
| `relevance_ai.event.version_id` | string | Yes      | ID of the draft version saved |

**`agent_published`** - Emitted when agent changes are published to make them live.

| Attribute                       | Type   | Required | Description                     |
| ------------------------------- | ------ | -------- | ------------------------------- |
| `relevance_ai.event.agent_id`   | string | Yes      | ID of the agent being published |
| `relevance_ai.event.version_id` | string | Yes      | ID of the version now live      |

### Tool events

**`tool_created`** - Emitted when a new tool is created (from scratch or cloned).

| Attribute                                   | Type   | Required | Description                  |
| ------------------------------------------- | ------ | -------- | ---------------------------- |
| `relevance_ai.event.tool_id`                | string | Yes      | ID of the newly created tool |
| `relevance_ai.event.cloned_from_tool_id`    | string | No       | Source tool ID if cloned     |
| `relevance_ai.event.cloned_from_region`     | string | No       | Source region if cloned      |
| `relevance_ai.event.cloned_from_project_id` | string | No       | Source project if cloned     |

**`tool_deleted`** - Emitted when a tool is permanently deleted.

| Attribute                    | Type   | Required | Description            |
| ---------------------------- | ------ | -------- | ---------------------- |
| `relevance_ai.event.tool_id` | string | Yes      | ID of the deleted tool |

**`tool_draft_saved`** - Emitted when work-in-progress changes are saved to a tool.

| Attribute                       | Type   | Required | Description                   |
| ------------------------------- | ------ | -------- | ----------------------------- |
| `relevance_ai.event.tool_id`    | string | Yes      | ID of the tool being edited   |
| `relevance_ai.event.version_id` | string | No       | ID of the draft version saved |

**`tool_published`** - Emitted when tool changes are published to make them live.

| Attribute                       | Type   | Required | Description                    |
| ------------------------------- | ------ | -------- | ------------------------------ |
| `relevance_ai.event.tool_id`    | string | Yes      | ID of the tool being published |
| `relevance_ai.event.version_id` | string | Yes      | ID of the version now live     |

### Workforce events

**`workforce_created`** - Emitted when a new workforce is created (from scratch, cloned, or duplicated).

| Attribute                                     | Type   | Required | Description                       |
| --------------------------------------------- | ------ | -------- | --------------------------------- |
| `relevance_ai.event.workforce_id`             | string | Yes      | ID of the newly created workforce |
| `relevance_ai.event.cloned_from_workforce_id` | string | No       | Source workforce ID if cloned     |
| `relevance_ai.event.cloned_from_region`       | string | No       | Source region if cloned           |
| `relevance_ai.event.cloned_from_project_id`   | string | No       | Source project if cloned          |

**`workforce_deleted`** - Emitted when a workforce is permanently deleted.

| Attribute                         | Type   | Required | Description                 |
| --------------------------------- | ------ | -------- | --------------------------- |
| `relevance_ai.event.workforce_id` | string | Yes      | ID of the deleted workforce |

**`workforce_draft_saved`** - Emitted when work-in-progress changes are saved to a workforce.

| Attribute                         | Type   | Required | Description                      |
| --------------------------------- | ------ | -------- | -------------------------------- |
| `relevance_ai.event.workforce_id` | string | Yes      | ID of the workforce being edited |
| `relevance_ai.event.version_id`   | string | No       | ID of the draft version saved    |

**`workforce_published`** - Emitted when workforce changes are published to make them live.

| Attribute                         | Type   | Required | Description                         |
| --------------------------------- | ------ | -------- | ----------------------------------- |
| `relevance_ai.event.workforce_id` | string | Yes      | ID of the workforce being published |
| `relevance_ai.event.version_id`   | string | No       | ID of the version now live          |

### Permission events

**`project_user_role_updated`** - Emitted when a user's role within a project is updated. This occurs when an admin changes another user's access level in the project settings.

| Attribute                           | Type   | Required | Description                               |
| ----------------------------------- | ------ | -------- | ----------------------------------------- |
| `relevance_ai.event.target_user_id` | string | Yes      | ID of the user whose role was changed     |
| `relevance_ai.event.project_role`   | string | Yes      | The new project role assigned to the user |

**`organization_user_role_updated`** - Emitted when a user's role within an organization is updated. This occurs when an admin changes another user's organization-level access.

| Attribute                              | Type   | Required | Description                                    |
| -------------------------------------- | ------ | -------- | ---------------------------------------------- |
| `relevance_ai.event.target_user_id`    | string | Yes      | ID of the user whose role was changed          |
| `relevance_ai.event.organization_role` | string | Yes      | The new organization role assigned to the user |

### Example log record

```json theme={null}
{
  "resourceLogs": [{
    "scopeLogs": [{
      "logRecords": [{
        "timeUnixNano": 1768742472616000000,
        "severityNumber": 9,
        "severityText": "Info",
        "body": { "stringValue": "agent_deleted" },
        "attributes": [
          { "key": "relevance_ai.organization_id", "value": { "stringValue": "f6fb76a4-..." }},
          { "key": "relevance_ai.project_id", "value": { "stringValue": "3acde218-..." }},
          { "key": "relevance_ai.user_id", "value": { "stringValue": "941afb4d-..." }},
          { "key": "relevance_ai.user_email", "value": { "stringValue": "user@example.com" }},
          { "key": "relevance_ai.user_type", "value": { "stringValue": "user" }},
          { "key": "relevance_ai.ip_address", "value": { "stringValue": "119.18.1.156" }},
          { "key": "relevance_ai.device_info", "value": { "stringValue": "Mozilla/5.0..." }},
          { "key": "relevance_ai.event.agent_id", "value": { "stringValue": "3d47a9f5-..." }}
        ],
        "droppedAttributesCount": 0,
        "traceId": "",
        "spanId": ""
      }]
    }]
  }]
}
```

***

## Traces

Traces track execution flows for agents, workforces, and LLM completions. Use traces to understand performance, debug issues, and analyze agent behavior.

<Note>We're actively expanding our trace coverage. If there are specific spans or attributes you'd like to see, let your Account Manager know.</Note>

### Structure

```json theme={null}
{
  "resourceSpans": [{
    "scopeSpans": [{
      "spans": [...]
    }]
  }]
}
```

### Resource attributes

Resource attributes identify the service producing telemetry data. These attributes are attached to all spans and logs exported from Relevance AI.

| Attribute      | Type   | Value            | Description                                     |
| -------------- | ------ | ---------------- | ----------------------------------------------- |
| `service.name` | string | `"Relevance AI"` | Logical name of the service producing telemetry |

This attribute is applied to all spans - `chat`, `invoke_agent`, `multi_agent_system_trigger`, and `condition_trigger`. In your observability tool, you can filter all Relevance activity with a single query like `service.name = "Relevance AI"`, cleanly separating Relevance spans from your own services.

**Example structure:**

```json theme={null}
{
  "resourceSpans": [{
    "resource": {
      "attributes": [
        { "key": "service.name", "value": { "stringValue": "Relevance AI" } }
      ]
    },
    "scopeSpans": [{ "spans": [...] }]
  }]
}
```

### Base attributes

Included on all spans:

| Attribute                      | Type   | Required | Description                    |
| ------------------------------ | ------ | -------- | ------------------------------ |
| `relevance_ai.organization_id` | string | Yes      | Organization ID                |
| `relevance_ai.project_id`      | string | Yes      | Project ID                     |
| `relevance_ai.user_id`         | string | No       | User ID that triggered the run |
| `relevance_ai.user_email`      | string | No       | Email of the user              |

### Span properties

| Property            | Type   | Description                               |
| ------------------- | ------ | ----------------------------------------- |
| `traceId`           | string | 32-character hex trace identifier         |
| `spanId`            | string | 16-character hex span identifier          |
| `parentSpanId`      | string | Parent span ID (if child span)            |
| `name`              | string | Span name (e.g., `invoke_agent My Agent`) |
| `kind`              | int    | Span kind (3 = CLIENT)                    |
| `startTimeUnixNano` | int64  | Start time in nanoseconds                 |
| `endTimeUnixNano`   | int64  | End time in nanoseconds                   |
| `attributes`        | array  | Key-value pairs with span details         |
| `status.code`       | int    | 1 = OK, 2 = ERROR                         |

### Trace hierarchy

Spans share `traceId` and link via `parentSpanId`:

```
multi_agent_system_trigger
├── condition_trigger
└── invoke_agent
    ├── chat
    └── invoke_agent (sub-agent)
        └── chat
```

### Supported spans

### `invoke_agent`

Records a complete agent conversation/invocation.

**Name**: `invoke_agent` or `invoke_agent {agent_name}`

| Attribute                                  | Type    | Required | Description                                                                                                                                                                  |
| ------------------------------------------ | ------- | -------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `gen_ai.operation.name`                    | string  | Yes      | `"invoke_agent"`                                                                                                                                                             |
| `gen_ai.agent.id`                          | string  | Yes      | Agent UUID                                                                                                                                                                   |
| `gen_ai.agent.name`                        | string  | No       | Agent name                                                                                                                                                                   |
| `gen_ai.agent.description`                 | string  | No       | Agent description                                                                                                                                                            |
| `gen_ai.conversation.id`                   | string  | Yes      | Conversation/task UUID                                                                                                                                                       |
| `relevance_ai.agent.relevance_model`       | string  | Yes      | Relevance model ID                                                                                                                                                           |
| `relevance_ai.agent.runtime`               | string  | Yes      | `"default"` or `"phone_call"`                                                                                                                                                |
| `relevance_ai.agent.metadata`              | object  | Yes      | Custom metadata key-value pairs                                                                                                                                              |
| `relevance_ai.agent.knowledge_used`        | array   | Yes      | Knowledge set IDs used                                                                                                                                                       |
| `relevance_ai.agent.escalation.reason`     | string  | No       | Escalation reason                                                                                                                                                            |
| `relevance_ai.agent.escalation.context`    | string  | No       | Escalation context                                                                                                                                                           |
| `relevance_ai.agent.version_id`            | string  | No       | The active version ID of the agent at invocation time. Use this to compare performance across agent versions                                                                 |
| `relevance_ai.prompt_injection.detected`   | boolean | No       | Whether a prompt injection attempt was detected in the user input. Only present when [prompt injection detection](#prompt-injection-detection-enterprise-feature) is enabled |
| `relevance_ai.prompt_injection.confidence` | double  | No       | Confidence score (0.0–1.0) for the prompt injection detection. Only present when prompt injection detection is enabled                                                       |

### `chat`

Records a single LLM inference call.

**Name**: `chat {model_name}`

<Warning>This span was previously named `llm_completion`. If you have existing dashboards or queries filtering on `gen_ai.operation.name = "llm_completion"` or span names containing `llm_completion`, update them to use `"chat"`. The rename aligns with the [OpenTelemetry GenAI semantic conventions](https://opentelemetry.io/docs/specs/semconv/gen-ai/).</Warning>

| Attribute                        | Type   | Required | Description                                                                         |
| -------------------------------- | ------ | -------- | ----------------------------------------------------------------------------------- |
| `gen_ai.operation.name`          | string | Yes      | `"chat"`                                                                            |
| `gen_ai.request.model`           | string | Yes      | Requested model ID                                                                  |
| `gen_ai.request.temperature`     | int    | No       | Temperature setting                                                                 |
| `gen_ai.request.max_tokens`      | int    | No       | Max tokens setting                                                                  |
| `gen_ai.system_instructions`     | string | No       | System prompt                                                                       |
| `gen_ai.input.messages`          | string | Yes      | JSON-stringified input messages                                                     |
| `gen_ai.tool.definitions`        | string | Yes      | JSON-stringified tool definitions (use `JSON.parse()` to access as structured data) |
| `gen_ai.response.id`             | string | Yes      | Provider's response ID                                                              |
| `gen_ai.response.model`          | string | Yes      | Actual model used                                                                   |
| `gen_ai.response.finish_reasons` | array  | Yes      | Finish reasons (e.g., `["stop"]`)                                                   |
| `gen_ai.output.messages`         | string | Yes      | JSON-stringified output messages                                                    |
| `gen_ai.usage.input_tokens`      | int    | Yes      | Input tokens used                                                                   |
| `gen_ai.usage.output_tokens`     | int    | Yes      | Output tokens generated                                                             |

**Attribute**: `gen_ai.tool.definitions`

<Warning>`gen_ai.tool.definitions` was previously an array type (`arrayValue`). It is now a JSON string (`stringValue`). If you consume this field in your pipeline or queries, you need to call `JSON.parse()` on the value to work with the tool definitions as structured data. The content is identical - the same tool definitions are present, just serialized as a string for better compatibility across OTEL exporters and backends.</Warning>

### `multi_agent_system_trigger`

Records a complete workforce execution.

**Name**: `multi_agent_system_trigger`

| Attribute                                  | Type   | Required | Description                                                                                                       |
| ------------------------------------------ | ------ | -------- | ----------------------------------------------------------------------------------------------------------------- |
| `gen_ai.operation.name`                    | string | Yes      | `"multi_agent_system_trigger"`                                                                                    |
| `relevance_ai.workforce.workforce_id`      | string | Yes      | Workforce UUID                                                                                                    |
| `relevance_ai.workforce.workforce_task_id` | string | Yes      | Task UUID (for correlation)                                                                                       |
| `relevance_ai.workforce.type`              | string | Yes      | `"chat"` or `"default"`                                                                                           |
| `relevance_ai.workforce.metadata`          | object | Yes      | Custom metadata                                                                                                   |
| `relevance_ai.workforce.status`            | string | Yes      | Final status                                                                                                      |
| `relevance_ai.workforce.version_id`        | string | No       | The active version ID of the workforce at trigger time. Use this to compare performance across workforce versions |

### `condition_trigger`

Records a condition node evaluation in a workforce.

**Name**: `condition_trigger`

| Attribute                                     | Type   | Required | Description                      |
| --------------------------------------------- | ------ | -------- | -------------------------------- |
| `gen_ai.operation.name`                       | string | Yes      | `"condition_trigger"`            |
| `relevance_ai.condition.workforce_node_id`    | string | Yes      | Node ID in workforce             |
| `relevance_ai.condition.workforce_node_label` | string | No       | Node label/name                  |
| `relevance_ai.condition.input`                | string | Yes      | JSON-stringified condition input |
| `relevance_ai.condition.decisions`            | array  | Yes      | Decision objects (see below)     |
| `relevance_ai.condition.reasoning`            | string | No       | LLM reasoning                    |

**Decision object**:

```json theme={null}
{ "node": { "id": "node-id", "name": "Node Name" }, "will_run": true }
```

***

## Attribute value types

| Type    | JSON Format                                                        |
| ------- | ------------------------------------------------------------------ |
| String  | `{ "stringValue": "text" }`                                        |
| Integer | `{ "intValue": 123 }`                                              |
| Double  | `{ "doubleValue": 1.23 }`                                          |
| Boolean | `{ "boolValue": true }`                                            |
| Array   | `{ "arrayValue": { "values": [...] }}`                             |
| Object  | `{ "kvlistValue": { "values": [{ "key": "k", "value": {...} }] }}` |

***

## Resources

### Understanding OpenTelemetry

* [What is OpenTelemetry?](https://opentelemetry.io/docs/what-is-opentelemetry/) - Official introduction to OTEL concepts

### Compatible observability platforms

OTEL is supported by most major observability platforms. See the [OpenTelemetry Registry](https://opentelemetry.io/ecosystem/registry/) for a full list of compatible vendors and integrations.

For a detailed example of ingesting OTEL data, see [Honeycomb's OpenTelemetry guide](https://docs.honeycomb.io/send-data/opentelemetry/).

### Querying OTEL data directly

OTEL JSON files in S3 can be queried directly using SQL tools like [AWS Athena](https://docs.aws.amazon.com/athena/latest/ug/what-is.html).


# User Level Authentication
Source: https://relevanceai.com/docs/enterprise/user-level-authentication

Enable agents to use individual user authentication for secure, private access to integrations

<Warning>
  User Level Authentication is an Enterprise-only feature that is being rolled out. If you don't have access yet, please reach out to your sales representative.
</Warning>

User Level Authentication enables agents to operate using each user's individual authentication accounts for integrations, allowing multiple users to safely and privately use the same shared agent (especially in Chat) while only accessing their own data. Builders can configure this behavior per agent, and users only need to authenticate once per tool.

## How User Level Authentication works

User Level Authentication changes how authentication works when multiple people use the same agent. Instead of everyone using a single shared account to access integrations (like Google Drive, Slack, or HubSpot), each user connects their own individual work accounts.

<Columns>
  <Card title="Builders configure" icon="wrench">
    Choose which tool inputs require user level authentication using simple toggles
  </Card>

  <Card title="Users authenticate" icon="key">
    Connect once per tool when first using the agent - credentials are saved for future use
  </Card>

  <Card title="Credentials persist" icon="lock">
    Authentication is automatically remembered and reused for all future agent runs
  </Card>

  <Card title="Defaults apply" icon="star">
    Set preferred accounts for each integration that auto-populate in new agents
  </Card>
</Columns>

This means that when multiple team members use the same agent, each person only sees and accesses their own data within the connected integrations.

## Why use User Level Authentication

<Columns>
  <Card title="Enhanced Security" icon="shield-check">
    Each user accesses integrations with their own credentials and permissions, reducing the security risks of shared accounts.
  </Card>

  <Card title="Privacy Protection" icon="user-lock">
    Users only see their own data. When an agent queries Google Drive or searches through emails, it only accesses the data that user has permission to see.
  </Card>

  <Card title="Access Control" icon="user-shield">
    Users can only access what they're authorized to see in the external integration, based on their own account permissions rather than a shared account's permissions.
  </Card>

  <Card title="Simplified Management" icon="users-gear">
    No need to manage shared credentials or worry about what happens when team members leave. Each user manages their own connections.
  </Card>

  <Card title="Audit Trail" icon="clipboard-list">
    Actions taken by agents are attributed to individual user accounts in external integrations, making it easier to track who did what.
  </Card>
</Columns>

## Requirements for User Level Authentication

User Level Authentication works with tools that use **OAuth authentication only**.

**Supported:**

* Tool steps with OAuth account inputs (e.g., Google Sheets, Slack, HubSpot, Notion, Trello)
* Any integration that uses OAuth to connect user accounts

**Not supported:**

* Tools that use API key authentication
* Python code steps (even when using OAuth via the Integration helper class)
* Custom API calls with API key or bearer token authentication

If your tool uses API key authentication or Python steps, you'll need to use a shared account instead of User Level Authentication.

## Where User Level Authentication works

User Level Authentication is supported in the following contexts:

<Columns>
  <Card title="Chat" icon="message-lines">
    Users are prompted to connect their own account when an agent requests an authenticated action
  </Card>

  <Card title="Run tab" icon="play">
    Same experience when running an agent directly in the agent builder
  </Card>

  <Card title="Workforces" icon="users">
    When a workforce delegates to an agent with user level authentication configured, the user is prompted to authenticate
  </Card>
</Columns>

### Limitations

<Warning>
  **Embedded chat is NOT currently supported.** Users who embed agents on external sites via the embed widget cannot use user level authentication. For embedded agents, you'll need to use a shared account for those integrations instead.
</Warning>

## Setting up user level authentication

As a builder, you can configure User Level Authentication at the agent level.

<div>
  <iframe title="Setting up User Level Authentication" />
</div>

To enable User Level Authentication:

1. Head into your Agent and click Tools.
2. Head to the Tool you want to set User Level Authentication on, and find the Input for the OAuth account. From here, click the agent input setting, which is currently set to 'Set manually'.
3. Then, set this to 'Per-user authentication'.

When this is enabled, any agent using this tool will require each user to authenticate with their own account for this integration.

<Note>
  If you have [asset-level authentication controls](/docs/enterprise/rbac#permissions) enabled through RBAC, users can also choose from project-level shared accounts instead of authorizing their individual accounts.
</Note>

<Info>
  When dynamic authentication is enabled on a shared agent, all team members — not just admins — can add their own OAuth accounts. Members can only manage their own private accounts and cannot view or modify project-level shared accounts, which remain admin-managed.
</Info>

## User experience in Chat

When users interact with an agent in Chat that has User Level Authentication enabled, the authentication flow is seamless and intuitive.

<div>
  <iframe title="Using User Level Authentication in chat" />
</div>

### First-time authentication

All team members — not just admins — will see the authentication prompt when using an agent with dynamic authentication enabled for the first time.

<Steps>
  <Step title="Open the authentication prompt">
    When you run an agent that requires User Level Authentication for the first time, a pop-up appears with the guidance "Connect your account to use this tool."
  </Step>

  <Step title="Open the account dropdown">
    Click the **Select connected account** dropdown to choose an account. If your project has shared accounts available, they appear here alongside your personal options.
  </Step>

  <Step title="Add your own account">
    Click **Add account** to start the OAuth login flow for that integration.
  </Step>

  <Step title="Log in">
    Follow the on-screen steps to log in. Your credentials are saved automatically for future runs.
  </Step>
</Steps>

Members can only add and manage their own private accounts. Project-level shared accounts are managed by project admins and cannot be modified by members.

### Subsequent uses

After the first authentication:

* The agent automatically uses the user's saved credentials
* No interruption to the agent's execution
* Users can change their default credentials in project settings if needed

### Multiple tools

Authentication is required per tool, not per integration. If an agent uses two Google Drive tools, for example, you'll be prompted to authenticate each tool separately on the first run. After initial setup, all credentials are remembered and reused automatically.

## Managing default authentication

Users can manage their default authentication accounts for each integration at the project level, making it easy to control which accounts are used by agents.

### Setting integration defaults

To set or change your default account for an integration:

1. Click on 'Integrations & API Keys' in the sidebar
2. Click on the integration you want to configure

<Tip>
  Integration defaults pre-populate when you create new assets or use assets for the first time, saving you time and ensuring consistency across your agent interactions.
</Tip>

### Privacy and account visibility

All team members — not just admins — can add their own OAuth accounts when dynamic authentication is enabled on an agent. Members can connect their own private credentials without needing admin intervention, and can only see and manage their own private accounts. They cannot view or modify project-level shared accounts.

To protect privacy and security:

* **Private accounts are hidden** - Your personal accounts won't be visible to other users
* **Auths are private by default** - When you connect a new account, it's automatically set as private
* **Shared accounts** - If your organization has shared project-level accounts, you'll see those as options alongside your personal accounts

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What happens if I need to change my authentication credentials?">
    If you need to change your authentication credentials, you can do so easily. During Chat, you'll receive modal prompts that allow you to select existing credentials, add new ones, or mark a new credential as default. You can also manage your default credentials by going to Integrations & API Keys in the sidebar and selecting the integration you want to reconfigure.
  </Accordion>

  <Accordion title="Can I use both shared accounts and individual accounts in the same project?">
    Yes, if you have asset-level authentication controls enabled. When User Level Authentication is configured, you will see both your individual accounts and any shared project-level accounts available for selection. This gives you flexibility to choose the appropriate account based on the context.
  </Accordion>

  <Accordion title="What integrations support User Level Authentication?">
    User Level Authentication works with all integrations that support OAuth authentication in Relevance AI. This includes popular integrations like Google Drive, Gmail, Slack, HubSpot, Salesforce, and many others.
  </Accordion>

  <Accordion title="Is this feature only available on Enterprise plans?">
    Yes, User Level Authentication is an Enterprise-only feature. If you would like access to this feature, please reach out to your sales representative.
  </Accordion>

  <Accordion title="Where can I use User Level Authentication?">
    User Level Authentication works in Chat, the Run tab (when running agents directly in the agent builder), and in Workforces (when a workforce delegates to an agent with user level authentication configured). However, it is NOT currently supported in embedded chat widgets on external websites.
  </Accordion>

  <Accordion title="Why can't I enable User Level Authentication on my tool?">
    User Level Authentication only works with OAuth-based integrations. If you don't see the option to set 'Per-user authentication' on your agent input setting, it's likely because:

    * Your tool uses API key authentication instead of OAuth
    * Your tool uses a Python code step

    To use User Level Authentication, you need to use a tool that supports OAuth authentication (like Google Sheets, Slack, HubSpot, etc.). For tools that use API keys or Python steps, you'll need to use a shared account instead.
  </Accordion>

  <Accordion title="My integration uses API keys or Python code steps. Can I use User Level Authentication?">
    No. User Level Authentication only supports OAuth-based integrations. If your integration uses API key authentication, Python code steps, or custom API calls with bearer tokens, you'll need to use a shared account instead. Only integrations that use OAuth to connect user accounts (like Google Sheets, Slack, HubSpot, Notion, etc.) are compatible with User Level Authentication.
  </Accordion>

  <Accordion title="Can non-admin team members add their own OAuth accounts?">
    Yes. When dynamic authentication is enabled on a shared agent, all team members can add their own OAuth accounts — this is not limited to admins. Members see an "Add account" button with the guidance "Connect your account to use this tool." Members can only manage their own private accounts; they cannot view or modify project-level shared accounts, which remain admin-managed.
  </Accordion>
</AccordionGroup>


# Deep Researcher
Source: https://relevanceai.com/docs/get-started/chat/chat-agents/deep-researcher

Conduct comprehensive research on any topic using the Deep Researcher agent in Chat

Deep Researcher is a built-in Agent in Chat that conducts comprehensive research on a topic. Mention it with `@Deep Researcher` in any conversation to get started.

## How it works

1. Provide a research topic or question.
2. Deep Researcher searches the web and synthesizes findings into a detailed report.
3. Review the results and ask follow-up questions or request deeper investigation in the same conversation.

You can combine Deep Researcher with other agents in the same chat. For example, have Deep Researcher gather information on a topic, then pass the results to [Slide Builder](/docs/get-started/chat/slide-builder) to create a presentation or [Website Builder](/docs/get-started/chat/chat-agents/website-builder) to build a landing page.

## Credit cost

Deep Researcher typically costs 100+ credits per run. Credit usage varies based on the breadth and depth of the research.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="How do I use Deep Researcher?">
    Mention `@Deep Researcher` in any Chat conversation and describe the topic or question you want researched.
  </Accordion>

  <Accordion title="Can I use Deep Researcher with other agents?">
    Yes. You can @ mention Deep Researcher alongside your own agents or other built-in agents in the same conversation.
  </Accordion>

  <Accordion title="How detailed is the research output?">
    Deep Researcher produces a comprehensive report with sourced findings. You can ask follow-up questions to explore specific areas in more depth.
  </Accordion>
</AccordionGroup>


# Image Generator
Source: https://relevanceai.com/docs/get-started/chat/chat-agents/image-generator

Generate images from text prompts using the Image Generator agent in Chat

Image Generator is a built-in Agent in Chat that creates images from text prompts. Mention it with `@Image Generator` in any conversation to get started.

## How it works

1. Describe the image you want — include details about subject, style, composition, and mood.
2. Image Generator creates an image based on your prompt.
3. Review the result and ask for changes or variations in the same conversation.

You can combine Image Generator with other agents in the same chat. For example, use an agent to draft marketing copy, then ask Image Generator to create matching visuals.

## Credit cost

Image Generator typically costs 10+ credits per run.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="How do I use Image Generator?">
    Mention `@Image Generator` in any Chat conversation and describe the image you want to create.
  </Accordion>

  <Accordion title="Can I use Image Generator with other agents?">
    Yes. You can @ mention Image Generator alongside your own agents or other built-in agents in the same conversation.
  </Accordion>

  <Accordion title="Can I generate multiple images in one conversation?">
    Yes. You can keep prompting Image Generator in the same conversation to create additional images or variations.
  </Accordion>
</AccordionGroup>


# Website Builder
Source: https://relevanceai.com/docs/get-started/chat/chat-agents/website-builder

Build websites from a prompt or description using the Website Builder agent in Chat

Website Builder is a built-in Agent in Chat that generates complete websites from a text prompt or description. Mention it with `@Website Builder` in any conversation to get started.

## How it works

1. Describe the website you want — include details about layout, content, style, and purpose.
2. Website Builder generates a complete website based on your prompt.
3. Review the result and ask for changes in the same conversation.

You can combine Website Builder with other agents in the same chat. For example, use [Deep Researcher](/docs/get-started/chat/chat-agents/deep-researcher) to gather content on a topic, then ask Website Builder to turn it into a site.

## Credit cost

Website Builder typically costs 100+ credits per run. Credit usage varies based on the complexity and length of the generated site.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="How do I use Website Builder?">
    Mention `@Website Builder` in any Chat conversation and describe the website you want to create.
  </Accordion>

  <Accordion title="Can I use Website Builder with other agents?">
    Yes. You can @ mention Website Builder alongside your own agents or other built-in agents in the same conversation.
  </Accordion>

  <Accordion title="Can I iterate on the generated website?">
    Yes. After Website Builder generates a site, you can ask it to make changes in the same conversation — update copy, adjust layout, change colors, and so on.
  </Accordion>
</AccordionGroup>


# Chat
Source: https://relevanceai.com/docs/get-started/chat/introduction

Use AI Agents and Workforces in a chat interface on desktop, mobile, and browser

<img alt="Relevance Chat" />

Chat is where you interact with your AI Agents and Workforces. @ mention any Agent, combine multiple Agents in one conversation, or use pre-built Agents for tasks like generating images, deep research, and building slides — all from [chat.relevanceai.com](https://chat.relevanceai.com), on desktop and mobile.

Once you build an Agent or clone one from the Marketplace, you can use it in Chat — and [control which Agents are visible](/docs/get-started/chat/using-agents#agent-visibility) to your team.

## What you can do with Chat

<CardGroup>
  <Card title="Using Agents" icon="robot" href="/get-started/chat/using-agents">
    Mention any Agent in your project to bring its capabilities into a conversation
  </Card>

  <Card title="Saved prompts" icon="bookmark" href="/get-started/chat/saved-prompts">
    Reusable prompt templates shared across your team
  </Card>

  <Card title="Built-in Agents" icon="wand-magic-sparkles" href="/get-started/chat/using-agents#built-in-agents">
    Pre-built Agents for research, images, and slides
  </Card>
</CardGroup>

## Getting started

Navigate to [chat.relevanceai.com](https://chat.relevanceai.com) to start chatting. Chat works on desktop and mobile browsers, and is also available as a desktop app. Native [Android and iOS apps](/docs/get-started/chat/mobile-app) are also available.

### Selecting an LLM

You can choose which LLM to use for your conversation, or select **Auto** to let Chat pick the best model for your request.

<div>
  <iframe title="How to select an LLM in Chat" />
</div>

Once you've selected your model, you can [run Agents and Workforces](/docs/get-started/chat/using-agents) in your Chat conversations.

## Supported file types

Upload documents, images, and data files to give your conversations context, or generate new files from Agent responses. The following extensions are fully supported — other file types can still be uploaded but may not be processed reliably.

<Tip>
  Chat can preview 70+ file types directly in the conversation. Pasting from clipboard is limited to images only.
</Tip>

<CardGroup>
  <Card title="Documents" icon="file-lines">
    pdf, doc, docx, ppt, pptx, xls, xlsx, csv, txt, md, xml, json, abw
  </Card>

  <Card title="Images" icon="image">
    png, jpg, jpeg, webp, gif, svg, avif, bmp
  </Card>

  <Card title="Media" icon="film">
    **Audio:** mp3, wav, ogg, m4a, aac, aiff, flac

    **Video:** mp4, mpeg, mov, webm, avi, flv, mpg, wmv
  </Card>

  <Card title="Archives" icon="file-zipper">
    zip, bin, arc
  </Card>
</CardGroup>

### File capabilities by provider

Each provider supports different file types for reading (uploading files as context) and creating/exporting (generating new files from Chat responses).

<Warning>
  File export is not supported for organizations in the AU region.
</Warning>

<AccordionGroup>
  <Accordion title="OpenAI">
    **Size limits:** 32 MiB per file, 50 MiB combined per message

    **Document files:**

    * `.docx` — read and create/export
    * `.docx` with images — create/export only
    * `.xlsx` — read (converted to text) and create/export
    * `.csv` — read and create/export
    * `.pdf` — read and create/export
    * `.pdf` with images — read and create/export
    * `.txt` — read and create/export
    * `.md` — read only (create/export available on request)

    **Image files:**

    * `.png` — read and create/export
    * `.jpg` / `.jpeg` — read and create/export (exported as PNG)

    **Not supported:** `.pptx`
  </Accordion>

  <Accordion title="Anthropic">
    **Size limits:** 32 MiB per file, 32 MiB combined per message

    **Document files:**

    * `.docx` — read and create/export
    * `.docx` with images — create/export only
    * `.xlsx` — read (converted to text) and create/export
    * `.csv` — read and create/export
    * `.pdf` — read and create/export
    * `.pdf` with images — read and create/export
    * `.txt` — read and create/export
    * `.md` — read (limited) only (create/export available on request)

    **Image files:**

    * `.png` — read and create/export
    * `.jpg` / `.jpeg` — read and create/export (exported as PNG)

    **Not supported:** `.pptx`
  </Accordion>

  <Accordion title="Gemini">
    **Size limits:** 32 MiB per file, 50 MiB combined per message

    **Document files:**

    * `.docx` — read and create/export
    * `.docx` with images — create/export only
    * `.xlsx` — read (converted to text) and create/export
    * `.csv` — read and create/export
    * `.pdf` — read and create/export
    * `.pdf` with images — read and create/export
    * `.txt` — read and create/export
    * `.md` — read only (create/export available on request)

    **Image files:**

    * `.png` — read and create/export
    * `.jpg` / `.jpeg` — read and create/export (exported as PNG)

    **Video and audio (Gemini Pro only):**

    * Video: `.mp4`, `.mpeg`, `.mov`, `.avi`, `.flv`, `.mpg`, `.webm`, `.wmv`
    * Audio: `.wav`, `.mp3`, `.aiff`, `.aac`, `.ogg`, `.flac`

    **Not supported:** `.pptx`

    Video and audio files are only supported with Gemini. If you attach a video using an Anthropic or OpenAI model, the send button is blocked.
  </Accordion>
</AccordionGroup>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="How does Chat usage affect my credits?">
    Chat credit usage will show up in the **Plan & Billing** credit monitoring section as **'Chat Orchestrator'**.

    You'll be charged credits based on:

    * The LLM model you're using in Chat
    * How much you use Chat (number of messages and interactions)

    Built-in agents like [Slide Builder](/docs/get-started/chat/slide-builder) will also appear on this page when you use them in Chat.

    <Info>
      To monitor your Chat credit usage, navigate to **Settings > Plan & Billing** to see a detailed breakdown of your credit expenses. Learn more about [Plans and credits](/docs/admin/subscriptions/plans).
    </Info>
  </Accordion>

  <Accordion title="Can I use different LLMs in Chat?">
    Yes, you can select the LLM you want to use in Chat by following these steps:

    1. Click on the cog icon in the top right of your screen
    2. Choose the 'Chat Model' on the pop-up that appears

    You can also add to the System Prompt for your Chat from this screen.
  </Accordion>

  <Accordion title="Can I use Relevance Chat on mobile?">
    Yes. Relevance Chat is available as a native [Android and iOS app](/docs/get-started/chat/mobile-app), and you can also use it in any mobile browser at [chat.relevanceai.com](https://chat.relevanceai.com). The main Relevance AI platform ([app.relevanceai.com](https://app.relevanceai.com)) is not available on mobile — the mobile apps are Chat-only.
  </Accordion>

  <Accordion title="Can Chat remember my previous conversations?">
    Yes! Chat Memory allows Relevance Chat to remember and reference your recent conversations. You can ask questions like "What did I ask about earlier?" or "Who did I compare X against recently?" and get contextual answers based on your chat history.

    **Important limitations:** Chat Memory only remembers history within the same project, and you can only access your own chat history, not other users' conversations in the same project. This is different from Agent memory, which is conversation-specific memory within individual Agent interactions.
  </Accordion>

  <Accordion title="How do I switch between projects in Chat?">
    You can switch between projects using the project selector in the bottom left of the Chat interface. You can also use **Cmd + K** (Mac) or **Ctrl + K** (Windows) for quick access. When you switch projects, you'll have access to all agents available in that project.

    For more details, see [Learn how to switch projects in Chat](/docs/admin/project-management/switch-projects#switching-projects-in-chat).
  </Accordion>

  <Accordion title="What are Edge regions and how do I select them?">
    Edge regions (US Edge, AU Edge, and EU Edge) are available in the Chat browser extension and desktop app. If your project is in an Edge region, it will appear in the project switcher labeled with **(Edge)** — for example, **US (Edge)**.

    To select an Edge region project, open the project switcher and look for the **(Edge)** label next to the region name. For detailed instructions, see [how to switch projects in Chat](/docs/admin/project-management/switch-projects#selecting-an-edge-region-in-chat).

    Edge regions aren't available in the [Android and iOS apps](/docs/get-started/chat/mobile-app).
  </Accordion>

  <Accordion title="What keyboard shortcuts are available in Chat?">
    Chat supports the following keyboard shortcuts:

    | Action                           | Mac    | Windows |
    | -------------------------------- | ------ | ------- |
    | Toggle left panel (chat history) | Cmd+\\ | Ctrl+\\ |
    | Toggle right panel (Super GTM)   | Cmd+B  | Ctrl+B  |
    | Switch projects                  | Cmd+K  | Ctrl+K  |
  </Accordion>

  <Accordion title="Which Agents will show up in Chat - all the Agents in my organization, or just the ones in a specific project?">You'll be able to select the project you want to use Chat on in the bottom left. You'll be able to use all of the Agents you have access to in this project. If you want to access Agents in another project, you'll need to switch to that project in the bottom left.</Accordion>
  <Accordion title="Can I call multiple Workforces in the same Chat?">Not at this time, but we hope to add this feature soon!</Accordion>
  <Accordion title="Can I call a Workforce and an Agent in the same Chat?">Not at this time, but we hope to add this feature soon!</Accordion>
</AccordionGroup>


# Mobile app
Source: https://relevanceai.com/docs/get-started/chat/mobile-app

Use Relevance Chat on Android and iOS

<img alt="Relevance Chat on iOS" />

Take your Agents and Workforces on the go. Dictate messages with voice-to-text, attach photos straight from the camera, share content into Chat from any app on your device, and @ mention any Agent in the conversation.

Most features work the same as on [chat.relevanceai.com](https://chat.relevanceai.com), so workflows you've already set up keep working on mobile. This page covers what's different, including [mobile-only features](#mobile-only-features) and [features that aren't available yet](#not-available-in-the-mobile-app).

## Where to download the Chat mobile app

<CardGroup>
  <Card title="Android" icon="brand-android" href="https://play.google.com/store/apps/details?id=com.relevanceai.android&hl=en">
    Get Relevance Chat on Google Play
  </Card>

  <Card title="iOS" icon="brand-apple" href="https://apps.apple.com/au/app/relevance/id6757512568">
    Get Relevance Chat on the App Store
  </Card>
</CardGroup>

## How to sign in to the mobile app

<div>
  <iframe title="Signing in to the Relevance Chat mobile app" />
</div>

<Steps>
  <Step title="Open the Relevance Chat app">
    Launch the app on your Android or iOS device.
  </Step>

  <Step title="Tap Enterprise login">
    The sign-in page opens inside the app.
  </Step>

  <Step title="Choose a sign-in method">
    Sign in with email and password, SSO, Google, or Apple.
  </Step>
</Steps>

## What's available on mobile

<Tabs>
  <Tab title="Mobile-only features">
    <CardGroup>
      <Card title="Voice-to-text input" icon="microphone">
        Dictate directly into the chat input instead of typing
      </Card>

      <Card title="Camera attachments" icon="camera">
        Attach photos straight from the camera, alongside files and gallery images
      </Card>

      <Card title="Share into Chat" icon="share">
        Use the Android or iOS share sheet from any app to start a new chat with the shared text, image, or one or more files
      </Card>

      <Card title="Native share sheet" icon="arrow-up-from-bracket">
        Share a chat message or slide deck out through the OS share sheet
      </Card>

      <Card title="Light and dark theme" icon="circle-half-stroke">
        Follow the system theme, or override it in Settings
      </Card>

      <Card title="Thinking selector" icon="brain">
        Choose Off, Standard, or Extended thinking for the model under Settings → Preferences
      </Card>
    </CardGroup>
  </Tab>

  <Tab title="What works the same as Chat on web">
    The mobile apps share the core Chat experience with the web app, so workflows you've already set up keep working on the go.

    <CardGroup>
      <Card title="@-mention Agents" icon="at" href="/get-started/chat/using-agents">
        Bring any Agent into a conversation with @ mentions
      </Card>

      <Card title="Saved prompts" icon="bookmark" href="/get-started/chat/saved-prompts">
        Reuse prompt templates shared across your team
      </Card>

      <Card title="Slide builder" icon="presentation" href="/get-started/chat/slide-builder">
        Generate slide decks from a chat conversation
      </Card>

      <Card title="Model selection" icon="sparkles">
        Pick the model used for new conversations
      </Card>

      <Card title="File attachments" icon="paperclip">
        Attach files to messages from your device
      </Card>

      <Card title="Image generation" icon="image">
        Generate images directly inside a chat
      </Card>

      <Card title="Agent, Tool, and Workforce execution" icon="robot">
        Run Agents, Tools, and Workforces from chat
      </Card>

      <Card title="Message feedback" icon="thumbs-up">
        Rate responses to improve future replies
      </Card>

      <Card title="Project switching" icon="arrow-right-arrow-left">
        Move between projects without signing out
      </Card>

      <Card title="Markdown rendering" icon="markdown">
        Messages render formatted markdown the same as on web
      </Card>
    </CardGroup>
  </Tab>
</Tabs>

## How to navigate the mobile app

<div>
  <iframe title="Navigating the Relevance Chat mobile app" />
</div>

Two parts of the screen do most of the work — the **action bar** next to the chat input (for what you can do *inside* a conversation), and the **side menu** (for moving *between* conversations and reaching your account).

<CardGroup>
  <Card title="Inside a chat" icon="plus">
    Tap the **+** next to the message input to open the action menu. From there, browse the drop-down to pick a pre-built Agent, the slide builder, or the image generator. Tap the microphone to dictate instead of typing.
  </Card>

  <Card title="Between chats" icon="bars">
    Tap the menu icon to open the side panel. Start a new conversation with **New chat**, tap a previous chat to reopen it, or jump to your account from the same menu.
  </Card>
</CardGroup>

## How to sign out of the Relevance Chat app

<div>
  <iframe title="Signing out of the Relevance Chat mobile app" />
</div>

<Steps>
  <Step title="Open the side menu">
    Tap the menu icon in the top corner of the chat screen.
  </Step>

  <Step title="Go to Account">
    Open the **Account** section to see the user you're signed in as.
  </Step>

  <Step title="Tap Sign out">
    Confirm to sign out of the app.
  </Step>
</Steps>

## FAQ

<AccordionGroup>
  <Accordion title="Where can I download the mobile app?">
    Get Relevance Chat on [Google Play](https://play.google.com/store/apps/details?id=com.relevanceai.android\&hl=en) for Android or the [App Store](https://apps.apple.com/au/app/relevance/id6757512568) for iOS.
  </Accordion>

  <Accordion title="Which sign-in methods are supported?">
    Tap **Enterprise login** on the sign-in page, then sign in with email and password, SSO, Google, or Apple.
  </Accordion>

  <Accordion title="How do I switch projects on mobile?">
    Open settings and go to **Workspace** to switch projects. You'll also see your organization and region there (read-only).
  </Accordion>

  <Accordion title="How do I change the model used in chats?">
    Open settings and go to **Chat** to pick the model used for new conversations.
  </Accordion>

  <Accordion title="How do I adjust thinking level or theme?">
    Open settings and go to **Preferences** to set the thinking level (Off, Standard, Extended) and choose a light, dark, or system theme.
  </Accordion>

  <Accordion title="Where do I find the settings menu?">
    Open the settings menu to manage your account and customize Chat. It groups options into **Account** (signed-in user, sign out), **Workspace** (project, organization, region), **Chat** (model), **Preferences** (thinking level, theme), and **About** (app version and links).
  </Accordion>

  <Accordion title="How do I see the app version?">
    Open settings and go to **About** to see the installed app version and reference links.
  </Accordion>

  <Accordion title="Which features are web-only?">
    A few features aren't in the mobile apps yet — Edge regions, browser extension page-aware actions, keyboard shortcuts, and clipboard paste for files and images. Use the Chat browser extension or desktop app if you need them.
  </Accordion>

  <Accordion title="Are Edge regions available on mobile?">
    No. US Edge, AU Edge, and EU Edge are only available in the [Chat browser extension and desktop app](/docs/admin/project-management/switch-projects#selecting-an-edge-region-in-chat).
  </Accordion>

  <Accordion title="Do browser extension features work on mobile?">
    No. Page-aware actions that rely on the browser extension itself aren't available in the mobile apps.
  </Accordion>

  <Accordion title="Do keyboard shortcuts work on mobile?">
    No. The shortcuts listed in the Chat overview don't apply on mobile.
  </Accordion>

  <Accordion title="Can I paste files or images from the clipboard?">
    No. Use the camera or attachment picker instead.
  </Accordion>
</AccordionGroup>


# Saved prompts
Source: https://relevanceai.com/docs/get-started/chat/saved-prompts

Create, use, and manage reusable prompt templates in Chat

<Note>
  Saved prompts are currently in limited availability and being rolled out to select customers.
</Note>

Saved prompts are reusable prompt templates that help standardize common queries across your team. They allow you to create, manage, and share frequently-used prompts with everyone in your project, ensuring consistency and saving time on repetitive tasks.

## Saving a prompt

Save any prompt from your Chat conversations to make it available to all users in your project through the @ menu.

<div>
  <iframe title="How to save a prompt" />
</div>

<Steps>
  <Step title="Start a new conversation">
    Enter the prompt you want to save.
  </Step>

  <Step title="Bookmark the prompt">
    After sending your prompt, click the bookmark icon that appears under it.
  </Step>

  <Step title="Name and save">
    Give the prompt a descriptive name and click **Save**.
  </Step>
</Steps>

<Info>
  For Enterprise customers, the ability to create and manage saved prompts can be controlled through role-based access controls (RBAC). Learn more about [RBAC and user permissions](/docs/enterprise/rbac).
</Info>

## Using saved prompts

All users in the project can access and use saved prompts in their Chat conversations. Saved prompts appear alongside Agents in the @ menu, making it easy to reuse standardized prompt templates.

<div>
  <iframe title="How to use a saved prompt" />
</div>

1. Type @ in the chat input field to open the menu
2. Click 'Prompts' in the pop-up that appears
3. Select the saved prompt you want to use from the list

## Deleting a saved prompt

To delete a saved prompt:

1. Type @ in the chat and click 'Prompts' in the pop-up
2. Click the 'X' button next to the prompt you want to remove
3. Click 'Delete' on the confirmation pop-up to confirm deletion

<Tip>
  Combine saved prompts with @ mentioned Agents for workflows that use both reusable templates and specialized AI capabilities.
</Tip>


# Slide Builder
Source: https://relevanceai.com/docs/get-started/chat/slide-builder

Use Slide Builder in Chat to create AI-generated slide decks

<div>
  <iframe />
</div>

Slide Builder is a built-in Agent in Chat that allows you to generate slides based on your prompt. Slide Builder first creates a presentation outline showing all planned slides, allowing you to review and refine the structure before building. You can give Slide Builder a simple prompt, use built-in Agents like Deep Researcher to find info on a topic and then generate slides, or use your Agents and Workforces to produce an output that Slide Builder can then turn into slides.

## Key features

<CardGroup>
  <Card title="AI-generated" icon="wand-magic-sparkles">
    Tell Slide Builder what you want, and it will create a slide deck based on your prompts or instructions.
  </Card>

  <Card title="Outline preview" icon="list-check">
    Review a complete presentation outline with slide titles and descriptions before any slides are built.
  </Card>

  <Card title="Dynamic updates" icon="pen-to-square">
    Ask Slide Builder to modify or update your slides.
  </Card>

  <Card title="Content customization" icon="file-lines">
    Share your pre-existing text, slides, notes and Slide Builder will use your information in the slides.
  </Card>

  <Card title="Agent integration" icon="robot">
    Work with your Relevance AI agents or Workforces to research or generate content.
  </Card>

  <Card title="Branding and design customization" icon="palette">
    Your slides can look and feel exactly like your brand.
  </Card>

  <Card title="Export and reuse" icon="file-export">
    Export your slides in multiple formats including PDF, Images, PPTX, and JSON to re-use them in a new chat or share with others.
  </Card>
</CardGroup>

## Getting started

<div>
  <iframe title="Getting started with Slide Builder" />
</div>

1. **Navigate to Slides**: In [Relevance Chat](https://chat.relevanceai.com/chat/start), select the Slides feature.
2. **Plan your content and slide design**: Look at [Generating content and designing slides](/docs/get-started/chat/slide-builder#generating-content-and-designing-slides) for tips.
3. **Review and refine**: Review the generated slides and prompt Slide Builder to make changes.
4. **Present and share**: Use the slides directly in Slide Builder, share with others, or export them using the Export button.

### Generating content and designing slides

<Steps>
  <Step title="Generate content using prompts">
    Tell Slide Builder what content you want on your slide deck. Alternatively, use `@` to integrate your Relevance AI agent or Workforce.
  </Step>

  <Step title="Generate content with a PDF or other files">
    Upload a PDF or file with the information you want and Slide Builder will display it on your slides.
  </Step>

  <Step title="Branding and design customization using BrandKit">
    Upload screenshots of the branding or design you want displayed on the slides and Slide Builder will match your style.
  </Step>

  <Step title="Remix slides">
    Use the Remix feature to instantly refresh your slide designs while keeping all your content intact. Ask Slide Builder to "remix" your slides to explore different visual styles and layouts.
  </Step>

  <Step title="Incremental building">
    Request to see the outline first, then build slides one at a time or all at once. Example: "Create 10 slides about frogs, but only build slide 1" lets you preview before completing the full deck.
  </Step>
</Steps>

## BrandKits and Templates

Slide Builder includes two features to help you create consistent, professional presentations: **BrandKits** and **Templates**.

### BrandKits

A BrandKit ensures your slides are on-brand every time. It defines the visual identity of your presentations, including colors, typography, logos, and even the tone and voice of your content.

Slide Builder comes with **two default BrandKits** to get you started, but you can also create your own custom BrandKit.

#### What's included in a BrandKit

<CardGroup>
  <Card title="Moodboard images" icon="images">
    Visual references that set the overall aesthetic
  </Card>

  <Card title="Color palette" icon="swatchbook">
    The colors that represent your brand
  </Card>

  <Card title="Font styles" icon="font">
    Typography for headers and body text
  </Card>

  <Card title="Logos" icon="image">
    Your brand logos to include in slides
  </Card>

  <Card title="Brand tone" icon="face-smile">
    The personality of your brand (e.g., professional, friendly, bold)
  </Card>

  <Card title="Brand voice" icon="comment">
    How your brand communicates (e.g., formal, conversational)
  </Card>

  <Card title="AI design guidelines" icon="sliders">
    Specific instructions for preferred layouts or slide styles
  </Card>
</CardGroup>

#### Creating a BrandKit

You can create a BrandKit in two ways:

1. **From brand images**: Upload existing brand assets and let AI extract your colors, tone, and font styles automatically.
2. **From scratch**: Manually define each element of your brand, giving you complete control over every detail.

#### How to create a BrandKit

<div>
  <iframe title="How to create a BrandKit" />
</div>

1. Open Slide Builder and click **New BrandKit**, then select **From scratch** to manually build your BrandKit.
2. Upload moodboard images that represent your brand's visual style using the upload icon.
3. Add brand colors using the plus icon under **Colors**, select your desired colors, and click **Done** to confirm.
4. Select font styles for your slide headers and body content.
5. Add your logo by clicking the logo area or dragging and dropping your company image file.
6. Enter brand tone and voice in the provided text fields to define how your brand communicates.
7. Click the **Slide instructions** tab and enter your slide design guidelines.
8. Click **Done** to complete your BrandKit configuration.
9. Select your BrandKit from the list, enter your presentation topic and content, then click send to generate your slides.

### Templates

A Template captures the structure and formula of slides you've already created, allowing you to replicate successful presentations or use them as inspiration for new ones. This makes them perfect for standardizing recurring presentations like weekly reports, pitch decks, or training materials.

#### Creating a Template

Templates are created from existing slides. Once you've built a slide deck you're happy with, you can save it as a Template to reuse later.

#### How to create a Template

<div>
  <iframe title="How to create a Template" />
</div>

1. Click the **Convert to template** button on a previously created slide deck.
2. Enter a template name in the input field, then click **Convert to template** to save your slides as a reusable template.
3. In a new chat, click **Slide Builder** and select your template from the **Templates** section.
4. Enter your presentation topic along with any content to include, then click **Send** to generate your slides.

<Tip>
  You can use BrandKits and Templates together. Set your BrandKit first to define your visual identity, then apply a Template to structure your slides. Slide Builder will combine both for a consistent, on-brand presentation.
</Tip>

## Exporting slides

You can export your slides in multiple formats to share, present, or reuse them in other contexts. Slide Builder supports four export formats:

* **PDF**: Export your entire slide deck as a single PDF file, perfect for sharing or printing.
* **Images**: Download each slide as an individual image file, useful for embedding slides in other documents or presentations.
* **PPTX**: Export your slides as a PowerPoint presentation file, compatible with Microsoft PowerPoint and other presentation software.
* **JSON**: Export the slide data structure to upload into a new chat session, allowing you to recreate slides or use them as a reference for future presentations.

### How to export slides

<div>
  <iframe title="How to export slides" />
</div>

Follow these steps to export your slides:

1. **Open your slides in Relevance Chat**: Navigate to your slide deck in [Relevance Chat](https://chat.relevanceai.com/chat/start).
2. **Click Export**: In the top right of the slides pane, click the **Export** button.
3. **Select the desired format**: Choose from PDF, Images, PPTX, or JSON export options.

### Export limitations and notes

* **Animations and moving elements**: When presenting slides in chat, they may include animations or moving elements. However, when exported, these dynamic elements will be lost. Exported slides will show the static final state of each slide.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="How do I edit the slides?">
    Just tell Slide Builder what changes you want to make. It's AI-generated!
  </Accordion>

  <Accordion title="Can I use my own content in the slides?">
    Yes! You can upload your own files and documents, or integrate your Relevance AI agents to help populate the slides with your own content.
  </Accordion>

  <Accordion title="Can I review the outline before slides are built?">
    Yes! Slide Builder creates a presentation outline first, showing all planned slide titles and descriptions. You can request changes to the outline before building the actual slides.
  </Accordion>

  <Accordion title="Can I build slides one at a time?">
    Yes! You can request to build slides incrementally, such as "Create a 10-slide presentation but only build the first 3 slides." This helps you preview the style before completing the full deck.
  </Accordion>

  <Accordion title="Which export format should I use?">
    Choose the format based on your needs: Use **PDF** for sharing or printing complete presentations, **Images** for embedding individual slides in other documents, or **JSON** to save and recreate your slides in a new chat session.
  </Accordion>

  <Accordion title="Do I need a Relevance AI account?">
    To use the Slide Builder feature, you will need to sign up and create an account.
  </Accordion>
</AccordionGroup>

## Follow along on YouTube

We also have a number of YouTube tutorials you can use to follow along with and learn how to use Slide Builder.

<Tabs>
  <Tab title="Slide Builder Introduction">
    <Frame>
      <iframe title="We Built an AI That Makes Slide Decks for You (Inside Relevance Chat)" />
    </Frame>
  </Tab>

  <Tab title="BrandKits and Templates">
    <Frame>
      <iframe title="BrandKits and Templates in Slide Builder" />
    </Frame>
  </Tab>
</Tabs>


# File system
Source: https://relevanceai.com/docs/get-started/chat/super-gtm/file-system

Persistent storage for documents and reference material that the agent can access across all conversations

The file system is the agent's persistent storage. It holds shared documents, reference material, deliverables, and working files that the agent can read from or write to as instructed. It's completely separate from your integration connections (HubSpot, Gmail, Slack, etc.) — the agent works with those directly.

Think of the file system as a shared drive for your project. Anything you want the agent (or your team) to have access to across conversations lives here.

## Where to find files

You can view and manage your files in the **Files** tab. From here you can browse all project files, upload new documents, and see what the agent has saved.

<Frame>
  <img alt="Files tab in the Super GTM interface" />
</Frame>

## Key areas

<Tabs>
  <Tab title="Project files">
    Persistent storage shared across your project. This is where shared knowledge lives — company context, reference documents, templates, and anything the team needs access to across conversations.

    To add something to project files, just ask the agent:

    > "Save this report to the project files"

    > "Create a document in project files called Q3 Campaign Brief"

    You can also ask the agent to read, search, or update anything already stored there:

    > "What's in the project files?"

    > "Find the competitive analysis in project files"

    > "Update the onboarding checklist with the new step"

    Project files are managed by project admins.
  </Tab>

  <Tab title="Session files">
    Workspace scoped to a conversation. The agent uses this for working files — exploration results, execution plans, and task lists. Session files persist across sessions in a folder matching the conversation ID, but they're separate from shared project files.

    You generally don't need to interact with session files directly. The agent manages them as part of its workflow.
  </Tab>
</Tabs>

## How skills use the file system

Skills can optionally pull context from the file system. Each skill can have its own `/references` or `/assets` folder for skill-specific context, or reference shared project files when the same document is relevant across multiple skills.

When a skill runs, the agent fetches the relevant files at that moment — you can update a reference file once and all skills that use it pick up the new version automatically.

This is useful for storing things like:

* Templates that skills use as a starting point
* Reference documents (brand guidelines, process docs, competitive intel)
* Deliverables the agent produces that should persist across conversations
* Ways of working that the whole team should follow


# Integrations
Source: https://relevanceai.com/docs/get-started/chat/super-gtm/integrations

Connect your GTM tools to Super GTM via OAuth, API key, or built-in integrations

Super GTM connects to your go-to-market tech stack so the agent can read data and take actions across your tools. Integrations are managed from the Super GTM settings panel.

## Connecting integrations

<div>
  <iframe title="Connecting integrations" />
</div>

<Steps>
  <Step title="Open the settings panel">
    Click the gear icon next to the Super GTM toggle.
  </Step>

  <Step title="Find the integration">
    Navigate to the **Integrations** section and locate the tool you want to connect.
  </Step>

  <Step title="Authorize">
    Click **Connect account** (OAuth) or **Add API key** (API key integrations) and follow the prompts.
  </Step>
</Steps>

### Connection methods

<Tabs>
  <Tab title="OAuth">
    You click **Connect**, log in to the third-party service, and Relevance securely stores your tokens. No manual key management needed.

    1. Click **Connect** on the integration
    2. A modal opens and redirects you to the third-party login (e.g. Google, HubSpot)
    3. Authorize the requested scopes
    4. The integration shows as connected and is immediately available

    Each user must connect their own OAuth integrations individually. For per-user OAuth setup at the organizational level, see [user-level authentication](/docs/enterprise/user-level-authentication).
  </Tab>

  <Tab title="API key">
    You paste in a key from the third-party platform's settings page.

    Enter API keys via the integrations settings page — accessible from the Super GTM settings panel (click the external link icon on the integration) or from your project's Integrations page directly.

    A project admin can configure API key integrations at the project level (e.g. a shared Gong API key for the whole team), so individual users don't need to connect personally — they inherit the project's credentials. Admins can manage these in the builders app under [Integrations & API Keys](/docs/get-started/core-concepts/api-integration).
  </Tab>
</Tabs>

<Note>
  Depending on how your organization has configured each external application, you may need to request admin access to enable OAuth or to obtain an API key. Check with your IT team or the admin of the external application.
</Note>

### Managing connections

All active integrations are visible in the Super GTM settings panel. Each integration has one of three states:

<CardGroup>
  <Card title="Connected and enabled" icon="circle-check">
    Integration is active and the agent can use it
  </Card>

  <Card title="Connected but disabled" icon="circle-pause">
    Credentials exist but you've toggled it off in your settings
  </Card>

  <Card title="Not connected" icon="circle-xmark">
    No credentials stored — a "Connect" button is shown
  </Card>
</CardGroup>

### Write approval

Every integration action is classified as either a read or a write (create, update, delete). You control how the agent handles writes.

<Tabs>
  <Tab title="Ask before writing (default)">
    The agent pauses and asks for your approval before writing to any integration — e.g. "I'm about to create a contact in HubSpot — approve?"
  </Tab>

  <Tab title="Never ask">
    The agent executes writes automatically without asking for confirmation.
  </Tab>

  <Tab title="Per-integration overrides">
    Override write approval on a per-integration basis from the settings modal. For example: auto-approve Slack posts but always require approval for HubSpot CRM updates.
  </Tab>
</Tabs>

<Info>
  Certain built-in Agents (like Explorer) are hardcoded to read-only — they can query and search integrations but cannot write anything, regardless of your approval settings.
</Info>

### Setup details

<AccordionGroup>
  <Accordion title="HubSpot">
    **Connection method:** OAuth

    Gives Super GTM read and write access to your HubSpot CRM — search contacts, companies, and deals; create and update records; manage sequences and pipelines; pull activity history.

    HubSpot requests a broad set of optional scopes (contacts, companies, deals, files, tickets, automation sequences, sales email, pipeline management). This is intentional to enable the full range of Super GTM workflows. You can review and approve scopes during the OAuth flow.

    **How to connect:**

    1. In the Super GTM settings panel, find **HubSpot** under Integrations
    2. Click **Connect account**
    3. Sign in to HubSpot and authorize access when prompted
    4. Select the HubSpot portal to connect if you have multiple

    **Things to know:**

    * Each user must authorize their own HubSpot connection
    * If your HubSpot portal restricts app installations, a HubSpot admin needs to enable App Marketplace access for your user — see [Troubleshooting](#troubleshooting) below
    * The agent respects your HubSpot user permissions — it can only access records you have permission to see
  </Accordion>

  <Accordion title="Salesforce">
    **Connection method:** OAuth

    **Additional field:** **Sandbox account** checkbox — enable this if you're connecting a Salesforce sandbox (test/staging) environment. Routes login to test.salesforce.com instead of login.salesforce.com.

    Connects Super GTM to your Salesforce org — full API access with read and write CRM data, SOQL queries, and actions across Salesforce objects.

    **How to connect:**

    1. In the Super GTM settings panel, find **Salesforce** under Integrations
    2. If you're connecting a sandbox (test/staging) environment, check **Sandbox account** before proceeding — this routes login to test.salesforce.com instead of login.salesforce.com
    3. Click **Connect account**
    4. Sign in to Salesforce and authorize the connection
    5. If your org requires admin approval for connected apps, request access from your Salesforce admin

    **Things to know:**

    * Salesforce organizations often require admin approval for new OAuth apps — see [Troubleshooting](#troubleshooting) below if the connection fails
    * The agent operates within your Salesforce user permissions and sharing rules
    * Custom objects and fields are accessible if they're visible to your user profile
  </Accordion>

  <Accordion title="Apollo">
    **Connection method:** API key

    Gives Super GTM access to Apollo's sales intelligence database — search and enrich contact and company data, find contact information, pull prospect details.

    **How to connect:**

    1. Log in to Apollo and navigate to Settings > Integrations > API
    2. Generate an API key
    3. In the Super GTM settings panel, find **Apollo** under Integrations
    4. Click **Add API key** and paste your Apollo API key

    **Things to know:**

    * Apollo API keys are scoped to your Apollo account — each user needs their own key, or an admin can add a shared project-level key
    * API usage counts against your Apollo plan's API limits
  </Accordion>

  <Accordion title="Gong">
    **Connection method:** API key

    Connects Super GTM to Gong call recordings and transcripts — pull call data, surface key moments, incorporate conversation intelligence into deal briefings.

    **How to connect:**

    1. In Gong, navigate to Settings > Ecosystem > API
    2. Generate an API key (requires Gong admin access)
    3. In the Super GTM settings panel, find **Gong** under Integrations
    4. Click **Add API key** and enter your Gong API credentials

    **Things to know:**

    * Gong API access typically requires Technical Admin permissions in Gong — check with your Gong admin
    * The agent can only access calls that are already processed and indexed in Gong
  </Accordion>

  <Accordion title="Salesloft">
    **Connection method:** API key

    Connects Super GTM to your Salesloft account for sales engagement workflows.

    **How to connect:**

    1. Log in to Salesloft and navigate to your API settings
    2. Generate an API key
    3. In the Super GTM settings panel, find **Salesloft** under Integrations
    4. Click **Add API key** and paste your Salesloft API key
  </Accordion>

  <Accordion title="Slack">
    **Connection method:** OAuth

    Connects Super GTM to your Slack workspace — read and post messages, search conversations, list users, manage channels, add reactions.

    **How to connect:**

    1. In the Super GTM settings panel, find **Slack** under Integrations
    2. Click **Connect account**
    3. Select your Slack workspace and authorize the requested permissions
    4. If your workspace restricts app installations, request approval from your Slack admin

    **Things to know:**

    * Slack workspace admins can restrict which apps users can install — if the authorization fails, you may need admin approval
    * The agent can only read messages in channels it has been added to, and public channels
    * Posting to channels requires the agent to have access to those channels
  </Accordion>

  <Accordion title="Gmail">
    **Connection method:** OAuth (via GSuite)

    Connects Super GTM to your Gmail account — compose, read, send, and draft emails, manage labels, modify settings.

    **How to connect:**

    1. In the Super GTM settings panel, find **GSuite** under Integrations
    2. Click **Connect account** and sign in with your Google account
    3. Grant the requested Gmail permissions
    4. If your Google Workspace organization restricts third-party app access, contact your Google Workspace admin

    **Things to know:**

    * Gmail access is part of the GSuite integration, which also covers Google Calendar, Google Docs, Google Drive, and Google Sheets
    * Google Workspace organizations often have policies restricting which apps can access Gmail — your admin may need to approve the connection
    * Sent emails appear in your Gmail Sent folder as normal
  </Accordion>

  <Accordion title="Microsoft Outlook">
    **Connection method:** OAuth

    Connects Super GTM to your Microsoft Outlook account — read, send, and draft emails, manage labels and mailbox settings.

    **How to connect:**

    1. In the Super GTM settings panel, find **Microsoft Outlook** under Integrations
    2. Click **Connect account**
    3. Sign in with your Microsoft account and grant the requested permissions
    4. If your Microsoft 365 tenant restricts app consent, your IT admin may need to grant admin consent

    **Things to know:**

    * Microsoft 365 organizations often require admin consent for third-party OAuth apps — contact your IT admin if the authorization fails
    * The connection covers your personal Outlook mailbox; shared mailboxes may require additional configuration
  </Accordion>

  <Accordion title="Microsoft Teams">
    **Connection method:** OAuth

    Gives Super GTM access to your Microsoft Teams environment — read and send chat and channel messages, read meeting transcripts, manage online meetings.

    **How to connect:**

    1. In the Super GTM settings panel, find **Microsoft Teams** under Integrations
    2. Click **Connect account**
    3. Sign in with your Microsoft account and authorize the connection
    4. Admin consent is typically required for Teams app integrations — work with your IT admin

    **Things to know:**

    * Teams API access almost always requires Microsoft 365 admin consent — plan for this when setting up
    * Meeting transcript access requires that transcription is enabled in your Teams tenant
  </Accordion>

  <Accordion title="Google Calendar">
    **Connection method:** OAuth (via GSuite)

    Connects Super GTM to your Google Calendar — check availability, create events, get scheduling context.

    **How to connect:**
    Google Calendar access is included with the GSuite integration. Connect via the **GSuite** option in the Integrations section.

    **Things to know:**

    * Calendar access is part of the GSuite integration alongside Gmail, Google Docs, Google Drive, and Google Sheets
    * The agent can only create events on calendars you have write access to
    * Meeting invitations sent by the agent will come from your Google account
  </Accordion>

  <Accordion title="Microsoft Calendar">
    **Connection method:** OAuth

    Connects Super GTM to Microsoft Calendar (via Microsoft 365) — create and manage calendar events, including shared calendars.

    **How to connect:**

    1. In the Super GTM settings panel, find **Microsoft Calendar** under Integrations
    2. Click **Connect account**
    3. Sign in with your Microsoft account and authorize the connection

    **Things to know:**

    * Microsoft Calendar access is separate from Microsoft Outlook — you can connect either or both
    * Shared calendar access depends on your Microsoft 365 permissions
  </Accordion>

  <Accordion title="Google Docs">
    **Connection method:** OAuth (via GSuite)

    Connects Super GTM to Google Docs — read and edit documents, pull content from existing docs, create new documents.

    **How to connect:**
    Google Docs access is included with the GSuite integration. Connect via the **GSuite** option in the Integrations section.

    **Things to know:**

    * Google Docs access is part of the GSuite integration alongside Gmail, Google Calendar, Google Drive, and Google Sheets
    * The agent can only access documents shared with or owned by your Google account
  </Accordion>

  <Accordion title="Google Drive">
    **Connection method:** OAuth (via GSuite)

    Connects Super GTM to Google Drive — read and write files, search across your Drive.

    **How to connect:**
    Google Drive access is included with the GSuite integration. Connect via the **GSuite** option in the Integrations section.

    **Things to know:**

    * Google Drive access is part of the GSuite integration
    * The agent can only access files shared with or owned by your Google account
  </Accordion>

  <Accordion title="Google Sheets">
    **Connection method:** OAuth (via GSuite)

    Connects Super GTM to Google Sheets — read and write spreadsheet data.

    **How to connect:**
    Google Sheets access is included with the GSuite integration. Connect via the **GSuite** option in the Integrations section.

    **Things to know:**

    * Google Sheets access is part of the GSuite integration
    * The agent can only access spreadsheets shared with or owned by your Google account
  </Accordion>

  <Accordion title="Notion">
    **Connection method:** OAuth

    Connects Super GTM to your Notion workspace — read and create pages, query and create databases, create comments, manage blocks, search.

    **How to connect:**

    1. In the Super GTM settings panel, find **Notion** under Integrations
    2. Click **Connect account**
    3. Sign in to Notion and select which pages and databases to share with the integration
    4. Notion requires you to explicitly grant access to each page or database — the agent can only see what you've shared

    **Things to know:**

    * Notion's permission model is selective: you must share specific pages or databases during the OAuth flow. The agent cannot see anything you haven't explicitly shared.
    * If you want the agent to access new Notion pages in the future, you'll need to update the connection to include them
  </Accordion>

  <Accordion title="Linear">
    **Connection method:** OAuth or API key

    Linear is the only integration that supports both OAuth and API key. You can choose whichever you prefer — both methods work interchangeably. If either is configured, the integration shows as connected.

    Connects Super GTM to Linear — read, create, update, and archive issues, manage comments, read projects and labels.

    **How to connect:**

    *OAuth (per user):*

    1. In the Super GTM settings panel, find **Linear** under Integrations
    2. Click **Connect account** and authorize via your Linear account

    *API key (project-level):*

    1. In Linear, navigate to Settings > Security & access > Personal API keys
    2. Create a new API key
    3. In the Super GTM settings panel, find **Linear** under Integrations
    4. Click **Add API key** and paste your Linear API key
  </Accordion>

  <Accordion title="Ashby">
    **Connection method:** API key

    Connects Super GTM to Ashby for recruiting and ATS workflows.

    **How to connect:**

    1. Log in to Ashby and navigate to your API settings
    2. Generate an API key
    3. In the Super GTM settings panel, find **Ashby** under Integrations
    4. Click **Add API key** and paste your Ashby API key
  </Accordion>

  <Accordion title="Avoma">
    **Connection method:** API key

    Connects Super GTM to Avoma for meeting intelligence — access call recordings, transcripts, and meeting notes.

    **How to connect:**

    1. Log in to Avoma and navigate to your API settings
    2. Generate an API key
    3. In the Super GTM settings panel, find **Avoma** under Integrations
    4. Click **Add API key** and paste your Avoma API key
  </Accordion>

  <Accordion title="Firmable">
    **Connection method:** API key

    Connects Super GTM to Firmable for company data enrichment.

    **How to connect:**

    1. Log in to Firmable and navigate to your API settings
    2. Generate an API key
    3. In the Super GTM settings panel, find **Firmable** under Integrations
    4. Click **Add API key** and paste your Firmable API key
  </Accordion>

  <Accordion title="ZoomInfo">
    **Connection method:** API key

    Connects Super GTM to ZoomInfo for contact and company intelligence. ZoomInfo is the only integration that requires three separate fields instead of a single API key.

    **Required fields:**

    * **ZoomInfo Email** — your ZoomInfo account email address
    * **ZoomInfo Client ID** — your ZoomInfo API client ID
    * **ZoomInfo Private Key** — your ZoomInfo private key (this is a multi-line key, pasted into a larger text area)

    **How to connect:**

    1. Log in to ZoomInfo and navigate to your account settings
    2. Locate your **Email**, **Client ID**, and **Private Key**
    3. In the Super GTM settings panel, find **ZoomInfo** under Integrations
    4. Enter all three values
  </Accordion>

  <Accordion title="Cursor">
    **Connection method:** API key

    Connects Super GTM to Cursor — list, launch, and stop agents, view conversation history, send follow-up prompts.

    **How to connect:**

    1. In Cursor, navigate to your account settings and find the API section
    2. Generate an API key
    3. In the Super GTM settings panel, find **Cursor** under Integrations
    4. Click **Add API key** and paste your Cursor API key
  </Accordion>
</AccordionGroup>

### Troubleshooting

<AccordionGroup>
  <Accordion title="Salesforce OAuth error when connecting">
    When connecting Salesforce to Relevance AI, you may encounter one of these OAuth errors:

    <Frame>
      <img alt="Salesforce OAuth Error page showing OAUTH_APPROVAL_ERROR_GENERIC" />
    </Frame>

    This error occurs when the Relevance AI connected app in Salesforce needs to be configured to allow admin-approved users.

    <Frame>
      <img alt="Relevance AI OAuth Connection Failed page showing OAUTH_APP_ACCESS_DENIED error" />
    </Frame>

    A Salesforce administrator must complete these steps:

    <Steps>
      <Step title="Navigate to Setup">
        In Salesforce, click the gear icon in the top right and select **Setup**.

        <Frame>
          <img alt="Salesforce Setup menu accessed from the gear icon" />
        </Frame>
      </Step>

      <Step title="Open Connected Apps OAuth Usage">
        Search for and open **Connected Apps OAuth Usage**.

        <Frame>
          <img alt="Connected Apps OAuth Usage page in Salesforce Setup" />
        </Frame>
      </Step>

      <Step title="Install the Relevance AI app if needed">
        If the **Relevance AI** app is not already installed, click **Install**. This redirects to the app settings page — skip to step 5.

        <Frame>
          <img alt="Block and Install buttons for connected apps" />
        </Frame>
      </Step>

      <Step title="Go to Manage Connected Apps">
        Navigate to **Manage Connected Apps** and click **Relevance AI**.

        <Frame>
          <img alt="Manage Connected Apps page listing Relevance AI" />
        </Frame>

        This opens the Connected App Detail page for Relevance AI.

        <Frame>
          <img alt="Relevance AI Connected App Detail page" />
        </Frame>
      </Step>

      <Step title="Click Edit Policies">
        Click the **Edit Policies** button on the Connected App Detail page.

        <Frame>
          <img alt="Edit Policies button on the Connected App Detail page" />
        </Frame>
      </Step>

      <Step title="Configure permitted users">
        Under **OAuth Policies**, set the **Permitted Users** dropdown to one of:

        **All users may self-authorize** — any Salesforce user can connect to Relevance AI.

        <Frame>
          <img alt="Permitted Users dropdown showing All users may self-authorize and Admin approved users are pre-authorized" />
        </Frame>

        **Admin approved users are pre-authorized** — only users with specific profiles or permission sets can connect. If you choose this option, scroll down to the **Profiles** and **Permission Sets** sections to assign access.

        <Frame>
          <img alt="Profiles and Permission Sets sections on the Connected App page" />
        </Frame>

        Use **Manage Profiles** or **Manage Permission Sets** to select which users can connect.

        <Frame>
          <img alt="Application Profile Assignment page for selecting authorized profiles" />
        </Frame>
      </Step>
    </Steps>

    <Note>
      If you don't see the Relevance AI app in Connected Apps OAuth Usage, initiate the connection from the Super GTM settings panel first — this registers the app in your Salesforce org.
    </Note>
  </Accordion>

  <Accordion title="HubSpot app installation blocked by permissions">
    If a user can't install the Relevance App from the HubSpot App Marketplace, they likely don't have the **App Marketplace access** permission enabled on their account. A HubSpot Super Admin (or any admin with permission to manage users) needs to grant this.

    <Steps>
      <Step title="Open HubSpot settings">
        Log in to HubSpot and click the **Settings** gear icon in the top navigation bar.
      </Step>

      <Step title="Navigate to Users & Teams">
        In the left sidebar, click **Users & Teams**. This shows a list of all users in your HubSpot account.
      </Step>

      <Step title="Select the user">
        Find the user who needs access and click **Edit permissions**. This opens the permissions editor.
      </Step>

      <Step title="Enable App Marketplace access">
        Scroll down to **Account / Settings Access** (or search for "App Marketplace") and toggle **App Marketplace access** to **ON**. You may also want to enable **App Marketplace uninstall access** if the user should be able to remove apps later.
      </Step>

      <Step title="Save changes">
        Click **Save** to apply the new permissions. The user now has App Marketplace access and can install the Relevance App.
      </Step>
    </Steps>

    To grant access to multiple users at once, select the checkboxes next to each user in the Users & Teams list (they must share the same seat type), click **Edit permissions** at the top, toggle on **App Marketplace access**, and save.

    On HubSpot Enterprise, you can also add **App Marketplace access** to an existing Permission Set under **Settings** > **Users & Teams** > **Permission Sets** — all users assigned to that set will automatically receive the permission.

    <Note>
      If the permission changes don't take effect immediately, have the user open HubSpot in a private/incognito browser window. If it works there, they just need to clear their browser cache and cookies, then reload HubSpot.
    </Note>
  </Accordion>
</AccordionGroup>


# Super GTM
Source: https://relevanceai.com/docs/get-started/chat/super-gtm/introduction

An AI super agent for your entire go-to-market stack, with persistent skills, memory, and deep integrations

<Warning>
  Super GTM is currently in closed beta (started March 2, 2026) and only available for Enterprise customers. If you're an Enterprise customer and want to join the beta, contact your account team.
</Warning>

Super GTM is a Chat mode that gives you an AI super agent capable of working across your entire go-to-market (GTM) technology stack — including CRM, calendar, email, calls, support tickets, and more. It doesn't just read from your tools; it takes actions across them, with your approval.

What sets Super GTM apart from standard Chat is its persistent layer of context: skills that run repeatable workflows, a file system that stores documents and reference material across sessions, and memory that learns your preferences and history over time. These three features work together to make the agent capable, consistent, and personal.

## What you can do with Super GTM

<CardGroup>
  <Card title="Deep integrations" icon="plug" href="/get-started/chat/super-gtm/integrations">
    Connect your CRM, calendar, email, call recordings, support tickets, and other GTM tools
  </Card>

  <Card title="Skills" icon="wand-magic-sparkles" href="/get-started/chat/super-gtm/skills">
    Define repeatable workflows the agent can activate automatically whenever you need them
  </Card>

  <Card title="File system" icon="folder-open" href="/get-started/chat/super-gtm/file-system">
    Store documents and reference material that persist across conversations
  </Card>

  <Card title="Memory" icon="brain" href="/get-started/chat/super-gtm/memory">
    The agent learns your preferences, feedback, and context — and carries it into every session
  </Card>

  <Card title="Scheduled tasks" icon="clock" href="/get-started/chat/super-gtm/scheduled-tasks">
    Run skills automatically on a recurring schedule
  </Card>

  <Card title="Subagents" icon="robot" href="/get-started/chat/super-gtm/subagents">
    Connect any Agent you've built in Relevance AI so Super GTM can call it as a subagent
  </Card>
</CardGroup>

## Getting started

### Prerequisites

* You must be on an Enterprise plan with Relevance AI.
* Requires enrollment in the Super GTM closed beta. Contact your account team to request access.

### Enrollment

Super GTM is enabled by Relevance staff after discussion with your Account Executive (AE).

<CardGroup>
  <Card title="By user" icon="user">
    Provide a list of email addresses to your AE
  </Card>

  <Card title="By domain" icon="globe">
    Enable for all users with a specific email domain
  </Card>
</CardGroup>

Once enrolled, you'll see the Super GTM option in your Chat interface.

### Accessing Super GTM

<Tabs>
  <Tab title="Web">
    <div>
      <iframe title="Accessing Super GTM" />
    </div>

    <Steps>
      <Step title="Navigate to Chat">
        Go to [chat.relevanceai.com](https://chat.relevanceai.com)
      </Step>

      <Step title="Open the model picker">
        Click the model picker in the chat interface
      </Step>

      <Step title="Enable Super GTM">
        Toggle on the Super GTM mode
      </Step>

      <Step title="Start working">
        Start working across your GTM stack
      </Step>
    </Steps>
  </Tab>

  <Tab title="iOS (TestFlight)">
    <Steps>
      <Step title="Install TestFlight">
        [Download the TestFlight app](https://apps.apple.com/us/app/testflight/id899247664) on your iPhone
      </Step>

      <Step title="Join the beta">
        [Open the invite link](https://testflight.apple.com/join/xx1C1Z6y) to join the Relevance AI iOS beta
      </Step>

      <Step title="Install the app">
        Install Relevance AI from within TestFlight. Turn on automatic updates so the app stays up-to-date.
      </Step>

      <Step title="Toggle Super GTM">
        Open the model picker and enable the Super GTM mode toggle
      </Step>
    </Steps>
  </Tab>
</Tabs>

## Quick start

<Note>
  By default, the agent asks for your confirmation before any write action (send, create, update, or delete). You can adjust this per integration in your [integration settings](/docs/get-started/chat/super-gtm/integrations#write-approval).
</Note>

<Steps>
  <Step title="Connect your integrations">
    Click the gear icon next to the Super GTM toggle and connect the tools you use — CRM, calendar, email, and more. See [integrations](/docs/get-started/chat/super-gtm/integrations) for setup instructions.
  </Step>

  <Step title="Ask the agent something">
    Start with a real task. Try something like:

    > "What deals are closing this month?"

    > "Prep me for my next meeting"

    > "Draft a follow-up email to the attendees from my last call"

    The agent pulls from your connected tools and responds with actionable output.
  </Step>

  <Step title="Create your first skill">
    When you find a workflow you want to repeat, ask the agent to turn it into a [skill](/docs/get-started/chat/super-gtm/skills). Skills save the full process so the agent can run it again without you re-explaining it.
  </Step>
</Steps>

Share feedback with your account team or [contact support](/docs/get-started/support) at any time.


# Memory
Source: https://relevanceai.com/docs/get-started/chat/super-gtm/memory

How the agent remembers your preferences, corrections, and context across conversations

Memory is how the agent learns about you across conversations. It stores your preferences, corrections, project context, and references so the agent gets better over time — without you having to repeat yourself.

The agent checks your memory at the start of every conversation, so it always has your context before it starts working.

## Where to find memory

<Frame>
  <img alt="Context and Memories section in the Overview tab" />
</Frame>

You can view and manage your memories in the **Overview** tab, under **Context & Memories**. From here you can see all stored memories, review what the agent has learned about you, and remove anything that's outdated or incorrect.

<Note>
  Memory is scoped per user. Your memory is not shared with other users in your project.
</Note>

## How it works

The agent maintains a memory entry for each thing it learns about you. At the start of each conversation, the agent reads a memory index and opens the individual entries relevant to your request. Those are loaded into the agent's context alongside your message.

The agent doesn't load everything every time — it loads what's relevant. The more clearly your memories are described, the better the selection works.

## Memory types

<AccordionGroup>
  <Accordion title="User memory">
    Information about you — your role, goals, expertise, and preferences. The agent uses this to tailor its work to who you are. For example, you might store that you're a VP of Sales managing a team of 8 AEs focused on enterprise accounts, that you prefer data-driven insights and concise communication, or that you have deep HubSpot experience but are new to Slack integrations.
  </Accordion>

  <Accordion title="Feedback memory">
    Corrections and guidance you've given the agent. These are the most impactful memories — they prevent the agent from repeating mistakes. For example, you might correct the agent to keep email drafts professional but warm with no corporate jargon, to avoid including pricing in first outreach emails, or to always present data as bullet points rather than paragraphs.
  </Accordion>

  <Accordion title="Project memory">
    Context about ongoing work, goals, and timelines that the agent wouldn't otherwise know. For example, you might store that the enterprise team has a \$2.4M pipeline target for Q3 closing September 30, that the Q2 campaign is focused on enterprise expansion, or that a product launch is scheduled for June 15 with deliverables due by June 10.
  </Accordion>

  <Accordion title="Reference memory">
    Pointers to where information lives, so the agent knows where to look. For example, you might tell the agent that deal stage definitions are in the HubSpot wiki under "Sales Process > Deal Stages", that the competitive analysis is saved in project files under /research/competitors.md, or that customer feedback is tracked in a Notion database called "Product Feedback".
  </Accordion>
</AccordionGroup>

## Building up memory over time

You don't need to set up memory all at once. The agent builds it naturally as you work together.

<Tabs>
  <Tab title="Tell it to remember">
    Ask the agent to save something to memory at any time:

    > "Remember that I prefer bullet points over paragraphs"

    > "Remember that the Q3 target is \$2.4M"

    The agent will confirm what it stored and which memory type it used.
  </Tab>

  <Tab title="Correct it">
    When you correct the agent — "No, don't include pricing in the first email" — it saves a feedback memory so it doesn't make the same mistake next time.
  </Tab>

  <Tab title="Tell it to forget">
    If a memory is outdated or incorrect:

    > "Forget what you know about the Q2 campaign"

    The agent will find and remove the relevant memory.
  </Tab>
</Tabs>

## How skills and memory work together

[Skills](/docs/get-started/chat/super-gtm/skills) and memory complement each other:

* **Skills define the process.** A skill tells the agent *how* to do something — the steps, the data sources, the output format.
* **Memory personalizes the execution.** Memory tells the agent *your way* of doing it — your tone, your preferences, your priorities.

A well-written skill references memory where it matters. For example, a post-meeting follow-up skill tells the agent to check your memory for communication preferences before drafting emails. The skill is the same for everyone on the team, but the output adapts to each person's style.

This means you can share a project skill with your whole team, and each person's agent executes it differently based on their individual memory.


# Scheduled tasks
Source: https://relevanceai.com/docs/get-started/chat/super-gtm/scheduled-tasks

Set up recurring prompts that Super GTM acts on automatically at a time you define

Scheduled tasks let you configure recurring prompts that Super GTM runs on your behalf at a set time. Instead of manually typing the same request every morning, you write the instruction once, pick a schedule, and the agent handles it from there.

When a schedule fires, the agent receives your saved message exactly as if you had typed it into a conversation. It pulls from your connected [integrations](/docs/get-started/chat/super-gtm/integrations), uses your [skills](/docs/get-started/chat/super-gtm/skills), references your [files](/docs/get-started/chat/super-gtm/file-system), applies [memory](/docs/get-started/chat/super-gtm/memory), and delivers the output — just like a normal Super GTM interaction.

<Note>
  Scheduled tasks is in limited rollout. Your organization needs access enabled before the feature appears. Contact your account team or [reach out to support](/docs/get-started/support) to check availability.
</Note>

## Where to find schedules

The **Schedules** section lives in the Super GTM panel on the **Overview** tab, below the Calendar and Recently Viewed sections.

## Setting up a scheduled task

<Steps>
  <Step title="Open the Schedules section">
    In the Super GTM **Overview** tab, scroll to the **Schedules** section and click to add a new schedule.
  </Step>

  <Step title="Write your message">
    Enter a natural language instruction — this is the prompt the agent receives when the schedule fires. Write it the same way you would in a conversation.

    Examples:

    * "Summarize all new Gong calls from the last 7 days"
    * "Check my pipeline for deals that haven't had activity in 2 weeks and flag them"
    * "Find 5 new prospects matching our ICP and add them to my outreach list"
  </Step>

  <Step title="Set the frequency and time">
    Choose how often the task should run (daily, weekly, monthly, or annually), the time of day, and your timezone.
  </Step>

  <Step title="Save and enable">
    Save the schedule. It runs automatically at the next scheduled time.
  </Step>
</Steps>

<Tip>
  Write your scheduled message the same way you'd ask the agent in a conversation. The more specific your instruction, the more useful the output.
</Tip>

## Frequency options

<CardGroup>
  <Card title="Daily" icon="sun">
    Runs every day at a set time — e.g. morning pipeline digest at 9am
  </Card>

  <Card title="Weekly" icon="calendar-week">
    Runs on specific days each week — e.g. Monday kick-off report, Friday wrap-up
  </Card>

  <Card title="Monthly" icon="calendar">
    Runs once a month on a set day — e.g. account health review on the 1st
  </Card>

  <Card title="Annually" icon="calendar-check">
    Runs once a year on a set date — e.g. annual renewal check-in
  </Card>
</CardGroup>

<Info>
  All schedules run in your selected timezone. Full IANA timezone support (e.g. Australia/Sydney, America/New\_York).
</Info>

## Managing schedules

You can have up to **10 active schedules**. All management happens from the Schedules section of the Overview tab.

<Tabs>
  <Tab title="Pause and resume">
    Temporarily stop a schedule without deleting it. Paused schedules show a "(paused)" label in the list. Resume at any time from the same menu.
  </Tab>

  <Tab title="Edit">
    Update the message, frequency, time, or timezone for an existing schedule. Changes take effect from the next scheduled run.
  </Tab>

  <Tab title="Delete">
    Remove a schedule permanently when it's no longer needed. This cannot be undone.
  </Tab>
</Tabs>

Each schedule displays the last time it was triggered, so you can confirm it's running as expected.

## Use cases

<CardGroup>
  <Card title="Daily pipeline summaries" icon="chart-line">
    "Every morning, summarize my open deals and highlight any at-risk accounts"
  </Card>

  <Card title="Weekly prospect research" icon="magnifying-glass">
    "Every Monday, find 5 new prospects matching our ICP criteria and add them to my outreach list"
  </Card>

  <Card title="Recurring reporting" icon="calendar-week">
    "Every Friday at 5pm, prepare a summary of this week's customer conversations from Gong"
  </Card>

  <Card title="Follow-up nudges" icon="bell">
    "Every Tuesday, remind me of any contacts I haven't followed up with in 2+ weeks"
  </Card>
</CardGroup>


# Skills
Source: https://relevanceai.com/docs/get-started/chat/super-gtm/skills

Repeatable workflows that your agent activates automatically based on what you ask for

Skills are repeatable workflows that the agent activates based on what you ask for. Each skill defines the instructions, integrations, output format, and reference material for a specific task.

When your request matches a skill, the agent loads it and follows the instructions step by step. You don't need to explain how to do the work — the agent already has everything it needs.

## Where to find skills

View and manage your skills in the **Skills** tab in the right-side panel.

<Frame>
  <img alt="Skills tab in the Super GTM right-side menu" />
</Frame>

Browse all available skills (project, user, and built-in), see which are active, and open any skill to view or edit its contents.

## How the agent knows what's available

The agent sees all available skills at the start of every conversation. When your request matches a skill, it activates the skill and loads the full instructions.

<Tip>
  You don't need to reference skills by name. Just describe what you need, and the agent will match it to the right skill if one exists.
</Tip>

To reference a skill directly, use the `/` menu. Type `/` in the chat input to see your available skills, then filter by typing part of the skill name. Use the arrow keys to navigate and press Enter or Tab to insert the skill into your message.

## How to create a skill

<div>
  <iframe title="Creating a skill" />
</div>

Ask the agent:

> "Create a skill that does \[description of the workflow]"

The agent activates the built-in **Skill Creator** and walks you through defining the workflow, integrations, output format, and whether it should be a project skill (shared) or user skill (personal). The agent handles formatting and saving. You can also ask the agent to update an existing skill at any time.

### Example: creating a post-meeting follow-up skill

<Accordion title="Building the skill from a conversation">
  After a sales call, you ask the agent:

  > "Summarize my call with Relevance AI. Pull the action items, update the deal in HubSpot, and draft a follow-up email to the attendees."

  The agent finds the meeting in Google Calendar, reads the transcript, extracts action items, updates HubSpot, and drafts the email. You refine:

  > "Always include the deal stage and next scheduled meeting in the summary. Format the action items as a table with columns for action, owner, and deadline."

  The agent adjusts the output. Now you tell it:

  > "Make a skill out of this conversation."

  The agent activates the Skill Creator, reviews the conversation, and builds a skill that captures the full workflow — steps, integrations, output format, and your formatting preferences. It saves the skill to your catalog. From then on, the agent can run that same workflow whenever you need it.
</Accordion>

<Accordion title="Using the skill afterward">
  From then on, whenever you say something like:

  * "Follow up on my call with Relevance AI"
  * "I just got off a call with Sarah — process it"
  * "Summarize my last meeting and update the CRM"

  The agent matches your request to the skill and runs the same workflow.

  **What you get back:**

  * Meeting summary (date, attendees, duration, deal stage, next scheduled meeting)
  * Action items table (action, owner, deadline)
  * Confirmation of what was updated in HubSpot
  * A draft follow-up email ready to send or edit
  * Any open questions that need attention

  The skill captures the process, not specific data. The agent adapts it to whatever meeting and CRM records are relevant.
</Accordion>

## Types of skills

<Tabs>
  <Tab title="Project skills">
    Shared workflows available to everyone in your project. Use these for recurring tasks your team handles the same way — post-meeting follow-ups, pipeline reviews, or weekly reporting.

    Only admins can create and edit project skills. All project members can use them.
  </Tab>

  <Tab title="User skills">
    Personal workflows private to you. Use these for tasks tailored to your role or preferences — your specific outreach style, how you like reports formatted, or a workflow only you need.

    Create and edit your own user skills freely. Other team members can discover and read them, but can't modify them.
  </Tab>

  <Tab title="Built-in skills">
    Skills that ship with the platform. These include:

    * **Skill Creator** — helps you build new skills (see [How to create a skill](#how-to-create-a-skill) above)
    * **Meeting Prep & Summary** — prepares you for upcoming meetings using your calendar, transcripts, and notes. Requires a meeting integration like Google Calendar. Use it as a reference when creating your own skills.

    Built-in skills are read-only and available to everyone.
  </Tab>
</Tabs>

## Running skills on a schedule

<Frame>
  <img alt="Schedule configuration for running skills automatically" />
</Frame>

Skills don't have to be triggered by a conversation. Schedule the agent to run a skill automatically on a recurring basis — daily, weekly, monthly, or annually. Set up schedules in the **Overview** tab in the right-side panel.

<Accordion title="Example: scheduling a Follow up all calls skill">
  You have a skill called **Follow up all calls** that checks recent meetings, lists outstanding action items, and flags clients you haven't followed up with. You schedule it to run every weekday morning at 9am.

  Each morning, the agent runs the skill, checks your calendar and CRM, and sends you a summary — missed follow-ups, overdue action items, and calls without a recap.
</Accordion>

This is useful for any skill that benefits from regular execution:

<CardGroup>
  <Card title="Daily pipeline reviews" icon="chart-line">
    Summarize deal movements, stale opportunities, and next actions each morning.
  </Card>

  <Card title="Weekly reporting" icon="calendar-week">
    Generate end-of-week reports covering activity metrics, closed deals, and upcoming priorities.
  </Card>

  <Card title="CRM hygiene checks" icon="broom">
    Flag missing fields, outdated contacts, and deals without recent activity.
  </Card>

  <Card title="Monday meeting prep" icon="mug-hot">
    Brief you on upcoming meetings, attendee context, and open action items for the week.
  </Card>
</CardGroup>

To set up a schedule, go to the **Schedules** section in the **Overview** tab. Choose the skill, set the cadence, and the agent runs it automatically. See [scheduled tasks](/docs/get-started/chat/super-gtm/scheduled-tasks) for full details.

## Tips for writing and managing skills

The agent already knows how to create, refine, and consolidate skills through its built-in Skill Creator. These tips help you get more out of your skills.

<AccordionGroup>
  <Accordion title="Scope each skill to one workflow">
    Each skill should handle one specific workflow — "post-meeting follow-up" or "weekly pipeline review," not "handle all my sales tasks." If you find yourself writing a skill that covers multiple unrelated workflows, split it into separate skills. You can always consolidate later once you've confirmed each piece works well on its own.
  </Accordion>

  <Accordion title="Describe when it should activate">
    Include the kinds of things you'd say when you need this workflow. The agent uses these to match your requests to the right skill. Narrowly focused skills with clear descriptions produce more accurate matches.
  </Accordion>

  <Accordion title="Be specific about where to look">
    Specify which integrations to use — "check HubSpot for the deal" is better than "check the CRM."
  </Accordion>

  <Accordion title="Define the output">
    Describe what the result should look like so the agent delivers consistent output every time.
  </Accordion>

  <Accordion title="Reference memory">
    If the skill should respect your preferences (email tone, formatting style), tell it to check your [memory](/docs/get-started/chat/super-gtm/memory).
  </Accordion>

  <Accordion title="Don't over-specify">
    Define a clear process, but let the agent adapt to the data it finds. Specific steps are good; hard-coded values are not.
  </Accordion>

  <Accordion title="Start with a user skill, then promote to project">
    Use project skills for workflows the whole team runs the same way — post-meeting follow-ups, pipeline reviews, or weekly reporting. Use user skills for workflows tailored to your personal style or role. Start as a user skill and iterate through a few real scenarios until the results are consistent, then ask an admin to recreate it as a project skill. Each person's agent applies their own [memory](/docs/get-started/chat/super-gtm/memory) when running a shared project skill, so the same skill adapts to individual preferences without needing separate versions.
  </Accordion>

  <Accordion title="Keep your catalog lean">
    Fewer well-defined skills outperform a large collection of vague ones. The agent sees all available skills at the start of every conversation, and when too many overlap or have ambiguous descriptions, matching accuracy drops. Periodically review your skills in the **Skills** tab — consolidate overlapping skills (ask the agent: "consolidate my meeting follow-up skills into one") and remove any that haven't been triggered in several weeks. If your team uses project skills, treat the shared catalog as a curated set of your highest-value workflows.
  </Accordion>
</AccordionGroup>


# Subagents
Source: https://relevanceai.com/docs/get-started/chat/super-gtm/subagents

Connect any Agent you've built in Relevance AI as a subagent so Super GTM can call it during a conversation

Subagents let you connect any Agent you've built in the Relevance AI builder to Super GTM. Once connected, Super GTM sees the subagent in every conversation and can invoke it when a user request matches what the Agent does.

This is the bridge between the Agents your team has already built and the Super GTM chat experience — the work of designing, prompting, and configuring an Agent in the builder is reused directly inside Super GTM.

<Info>
  Subagents replace the deprecated agents-as-skills approach. Previously, Super GTM would become a connected Agent by reading its instructions and running its tools inline. Subagents instead run each connected Agent in its own conversation with its own context.
</Info>

## How it works

Super GTM acts as an orchestrator. Connected subagents appear in its catalog at the start of every conversation, and Super GTM decides when to delegate to one based on the user's request and the subagent's description.

Each subagent runs as a full agent in its own conversation — separate context, its own reasoning, its own tools. The only context shared with Super GTM is the message Super GTM sends in. When the subagent responds, its output is returned to Super GTM, which then continues with the user. If Super GTM needs more from the same subagent later in the session, it sends a follow-up message into that same subagent conversation rather than starting fresh.

<Tabs>
  <Tab title="What Super GTM sees">
    For each connected subagent, Super GTM sees:

    * The Agent's **name**
    * The Agent's **description**

    These two fields drive routing — Super GTM matches the user's request against the description to decide whether to invoke the subagent. A clear, specific description that explains what the Agent does and when to use it is the single biggest factor in getting reliable routing.
  </Tab>

  <Tab title="What the subagent sees">
    When Super GTM first invokes a subagent, the subagent receives a single message with the instructions for the task. It does not see the rest of the Super GTM conversation, the user's memory, or any other context Super GTM has loaded.

    The subagent runs from there using its own configuration — its system prompt, tools, integrations, and reasoning — exactly as it would when run standalone in the builder. If Super GTM sends a follow-up message later in the same session, it lands in the same subagent conversation and the subagent keeps its prior context.
  </Tab>

  <Tab title="How output flows back">
    When the subagent responds, its output is returned to Super GTM. Super GTM treats that output as the result of the delegation and uses it to continue working on the user's original request — summarizing it, combining it with other tool outputs, or passing it into another step.

    If a tool inside the subagent requires approval, the approval request is passed up to the Super GTM UI for the user to approve or reject. The subagent pauses until the user responds, then continues.
  </Tab>
</Tabs>

## Connecting an Agent as a subagent

<Steps>
  <Step title="Open the Agent in the builder">
    Go to [app.relevanceai.com](https://app.relevanceai.com) and open the Agent you want to connect.
  </Step>

  <Step title="Make sure the Agent has a name and description">
    The Agent must have both before it can be used as a subagent. The description is what Super GTM reads when deciding whether to route to this Agent — explain clearly what the Agent does and when to use it.
  </Step>

  <Step title="Open advanced settings">
    In the agent edit view, navigate to the **Advanced** section.
  </Step>

  <Step title="Toggle on Can be used by Super GTM">
    Enable the **Can be used by Super GTM** setting. The Agent is now available as a subagent in every Super GTM conversation in the project.
  </Step>
</Steps>

<Warning>
  The Agent's description is the most important field for subagent performance. Super GTM reads it on every conversation to decide when to route a request to this Agent. If the description is vague, missing, or written for end users rather than for the orchestrator, routing will be unreliable.
</Warning>

## Writing a description that routes well

The description should answer two questions clearly:

1. **What does the Agent do?** Be specific about the task — "drafts personalized cold outreach emails based on a prospect's LinkedIn profile" beats "writes emails."
2. **When should Super GTM use it?** Describe the kind of request that should trigger it — "use when the user wants to write a first-touch email to a new prospect."

<Accordion title="Example: a well-written subagent description">
  **Agent name:** Cold Email Writer

  **Description:** Drafts personalized cold outreach emails for new prospects. Takes a LinkedIn URL or company name as input, researches the prospect, and produces a short first-touch email matching our team's voice and outreach playbook. Use when the user asks for an outreach email, an intro email, or wants to start a sequence with a new prospect. Do not use for follow-ups on existing conversations — those should go to the Follow-up Email Writer.

  This works because it explains exactly what the agent produces, what it needs as input, when to route to it, and when *not* to. Super GTM uses all of that to make a routing decision.
</Accordion>

<Accordion title="Example: a description that won't route well">
  **Agent name:** Email Agent

  **Description:** Helps with emails.

  This fails because Super GTM has no way to distinguish it from any other email-related capability — the built-in skills, an integration, or another subagent. Most email-related requests will not route here, and the ones that do will be inconsistent.
</Accordion>

## Tips for managing subagents

<AccordionGroup>
  <Accordion title="Treat the description as a routing prompt, not user-facing copy">
    The description is read by Super GTM, not the end user. Write it for an LLM making a routing decision. Include trigger phrases users might say, the kinds of inputs the agent expects, and any cases where this agent should *not* be used.
  </Accordion>

  <Accordion title="Connect agents that do something specific">
    Subagents work best when each one owns a well-defined task. A "research a prospect" subagent and a "draft a follow-up email" subagent will route more reliably than a single "do anything sales-related" subagent.
  </Accordion>

  <Accordion title="Keep your catalog lean">
    Super GTM sees every connected subagent at the start of every conversation. As the catalog grows, the orchestrator has more tools and agents to choose between, and routing accuracy can drop. We recommend keeping the total number of connected subagents under \~20 per project. Super GTM also has 10+ built-in tools and agents already, so the effective catalog is larger than just your subagents.
  </Accordion>

  <Accordion title="Iterate on routing by watching real conversations">
    If Super GTM routes to the wrong agent — or fails to route to your subagent when it should — the fix is almost always in the description. Update it to clarify when the agent should be used, then test the same request again.
  </Accordion>

  <Accordion title="Compare with skills before connecting">
    A [skill](/docs/get-started/chat/super-gtm/skills) is the right choice when the workflow is a sequence of steps Super GTM itself should run. A subagent is the right choice when the work is best done by a separate agent with its own configuration — its own system prompt, tools, or integrations. If a skill can do the job, prefer the skill.
  </Accordion>
</AccordionGroup>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can Super GTM run multiple instances of the same subagent at once?">
    Not currently. Each connected agent has a single conversation per Super GTM session, and follow-up messages from Super GTM go to that same conversation rather than spinning up a new instance.
  </Accordion>

  <Accordion title="Can Super GTM run subagents in parallel?">
    Not currently. When Super GTM invokes a subagent, it waits for that subagent to respond before continuing — invocations happen one at a time, sequentially.
  </Accordion>

  <Accordion title="Can Super GTM send a follow-up message to an existing subagent conversation?">
    Yes. If Super GTM needs more from a subagent it has already invoked, it sends a follow-up message into the same subagent conversation rather than starting a new one. The subagent keeps its prior context for that session.
  </Accordion>

  <Accordion title="What if the subagent uses a tool that requires approval?">
    Approval requests are passed up from the subagent to the Super GTM UI, where the user approves or rejects them just like they would for any other approval-gated action. Once the user responds, the subagent continues from where it paused.
  </Accordion>

  <Accordion title="Is there a limit to how many subagents I can connect?">
    There is no hard technical limit, but we recommend connecting no more than \~20 subagents per project. Beyond that, Super GTM has more capabilities to choose between and routing accuracy starts to drop. Keep in mind that Super GTM already ships with 10+ built-in tools and agents, so each subagent you connect adds to that baseline.
  </Accordion>

  <Accordion title="Can other Super GTM users in my project see and use my connected subagents?">
    Yes. The "Can be used by Super GTM" setting is project-wide — once an Agent is enabled, every Super GTM user in the project sees it as a subagent. Use this in combination with the Agent's own permissions in the builder to control who can run it.
  </Accordion>

  <Accordion title="What happens if a subagent's description is missing or empty?">
    Super GTM cannot enable the toggle without a description. If you remove the description after the toggle is on, the Agent will still appear in the catalog but Super GTM will have no information to route on, so requests are unlikely to reach it.
  </Accordion>
</AccordionGroup>

## Skills, subagents, and memory

Subagents sit alongside [skills](/docs/get-started/chat/super-gtm/skills) and [memory](/docs/get-started/chat/super-gtm/memory) as the three ways you tailor Super GTM to your team's work:

* **Skills** define repeatable workflows Super GTM runs itself, step by step.
* **Subagents** delegate work to a separate agent that has its own configuration and runs in its own context.
* **Memory** personalizes how Super GTM and its subagents present output back to each user.

A common pattern is a project skill that tells Super GTM to call a specific subagent at a specific step — for example, a "post-meeting follow-up" skill that delegates email drafting to your Cold Email Writer subagent. Memory then shapes the tone of the final output for each user.


# Using Agents and Workforces in Relevance Chat
Source: https://relevanceai.com/docs/get-started/chat/using-agents

Utilize your Agents and Workforces in Chat conversations

Mention any Agent you have access to in your project using @, and Chat will integrate their capabilities into your conversation. Hover over any Agent in the @ menu to see a pop-up with its name, description, and other details — making it easy to pick the right one without leaving the chat.

<CardGroup>
  <Card title="Individual Agents" icon="robot">
    Use a single Agent for specific tasks
  </Card>

  <Card title="Multiple Agents" icon="users">
    Combine different Agents in the same conversation for complex workflows
  </Card>

  <Card title="Workforces" icon="people-group">
    Use teams of Agents working together on complex tasks
  </Card>

  <Card title="Marketplace Agents" icon="store">
    Use pre-built Agents cloned from the [Marketplace](/docs/get-started/marketplace/introduction)
  </Card>

  <Card title="Built-in Agents" icon="wand-magic-sparkles" href="#built-in-agents">
    Pre-built Agents for websites, research, images, and slides
  </Card>
</CardGroup>

## Agent visibility

By default, every Agent in your project appears in Chat's @-mention menu and "Browse agents" list. You can hide an Agent from Chat discovery using the **Visible in Chat** toggle.

<img alt="Visible in Chat setting" />

To change an Agent's visibility:

1. Open the Agent in the builder
2. Go to **Advanced Settings > General**
3. Toggle **Visible in Chat** on or off

<Note>
  Hiding an Agent from Chat does not disable it. Hidden Agents remain fully accessible in the builder's app, via API, and direct links.
</Note>

## Built-in Agents

Chat includes several pre-built Agents you can use without any setup. Mention them with @ in any conversation.

| Agent                                                            | Description                                                              | Credit cost          |
| ---------------------------------------------------------------- | ------------------------------------------------------------------------ | -------------------- |
| [Slide Builder](/docs/get-started/chat/slide-builder)                 | Generate branded slide decks from prompts, files, or other agent outputs | Varies by deck size  |
| [Website Builder](/docs/get-started/chat/chat-agents/website-builder) | Build websites from a prompt or description                              | 100+ credits per run |
| [Image Generator](/docs/get-started/chat/chat-agents/image-generator) | Generate images from text prompts                                        | 10+ credits per run  |
| [Deep Researcher](/docs/get-started/chat/chat-agents/deep-researcher) | Conduct comprehensive research on a topic                                | 100+ credits per run |

## Providing feedback on responses

<div>
  <iframe title="Providing feedback on responses" />
</div>

Rate any response using the thumbs up and thumbs down icons that appear alongside the copy button at the bottom of each message. You can optionally add written feedback in the modal dialog that appears after rating.

<CardGroup>
  <Card title="Thumbs down with written feedback" icon="thumbs-down">
    Use when a response is incorrect, incomplete, or unhelpful. Describe what went wrong to
    help improve the agent's performance.
  </Card>

  <Card title="Thumbs up" icon="thumbs-up">
    Reinforce good behaviors when a response is particularly helpful, accurate, or
    well-formatted.
  </Card>

  <Card title="Written feedback" icon="comment">
    Add details about what worked or didn't work to make your feedback more actionable.
  </Card>
</CardGroup>


# Agents
Source: https://relevanceai.com/docs/get-started/core-concepts/agents

Introduction to AI agents and building your AI Workforce!

## What are Agents?

Agents are powered by LLMs that plan and complete tasks on autopilot. They are given tools, and decide how to use tools to achieve goals prompted by you.

Unlike traditional automation that follows rigid, pre-programmed rules, AI agents can:

* **Think and reason** through complex problems
* **Adapt their approach** based on context and feedback
* **Learn from experience** by storing successful patterns
* **Make decisions** about which tools to use and when
* **Communicate naturally** with humans and other agents

## What are the benefits of an AI Agent?

<CardGroup>
  <Card title="Scales to meet demand" icon="chart-line">
    Agents can flex with your needs, handling seasonal or regional spikes in activity—especially useful in industries like education, hospitality, and finance.
  </Card>

  <Card title="Increases engagement" icon="rocket">
    Automate repetitive tasks so your team can focus on high-value work. Sales teams, for example, use agents to qualify leads, freeing reps to close deals.
  </Card>

  <Card title="Learns from experience" icon="brain">
    If an agent doesn't know something, it can escalate to a human, store the answer, and use it next time. Ours posts unknowns to Slack for fast input and learning.
  </Card>

  <Card title="Works solo or with you" icon="user-gear">
    Agents can run fully autonomously or in co-pilot mode. One of ours builds Webflow pages daily—no human touch required.
  </Card>

  <Card title="Responds flexibly" icon="arrows-split-up-and-left">
    Unlike rule-based tools, agents adapt their responses to each task. A support agent might troubleshoot one day and resolve billing the next.
  </Card>

  <Card title="Available 24/7" icon="clock">
    Agents work around the clock without breaks, ensuring consistent service and response times regardless of time zones or business hours.
  </Card>
</CardGroup>

## How AI Agents Work

Agents in Relevance AI can operate in different ways:

* **Autopilot** - They complete tasks independently.
* **Human in the loop** - Asking humans for input or approval when needed.

You can trigger agents to start work with:

* **Pre-built integrations** - Seamlessly connect with existing tools.
* **API access** - Use them in your applications.
* **User interface** - Run agents directly from our platform.

## Where are Agents most useful?

AI agents excel in roles requiring consistent execution of complex but well-defined processes, such as:

* **Customer Support**: Handling routine inquiries, troubleshooting common issues, and escalating complex problems to human agents
* **Sales Qualification**: Engaging with prospects, gathering initial information, and identifying promising leads
* **Content Creation**: Generating drafts of routine documents, reports, and communications
* **Data Analysis**: Processing and summarizing large volumes of information to extract actionable insights
* **Scheduling and Coordination**: Managing calendars, setting up meetings, and sending reminders

## Organizing your Agents

As you build and use more Agents, keeping them organized becomes essential for efficient workflows. Relevance AI provides several ways to manage and access your Agents:

### Pinning Agents for quick access

Pin your most frequently used Agents to keep them easily accessible in the Relevance AI platform. Pinned Agents appear:

* **At the top of your agents list** in the main app
* **First in the "Browse agents" section** when creating a new chat

This makes it easy to quickly access the Agents you rely on most, without scrolling through your entire collection.

<Tip>
  To pin an Agent, click the pin icon next to the Agent's name in your agents list. You can pin multiple Agents and unpin them at any time by clicking the pin icon again.
</Tip>

### Organizing with folders

You can organize your Agents into folders within your project, making it easier to categorize and find Agents based on their purpose, department, or use case. This is particularly useful for teams managing large numbers of Agents across different functions.

### Organizing with tags

Tags give you a flexible way to categorize agents that cuts across folder boundaries — unlike folders, a single agent can have multiple tags at once, making it easy to group agents by team, status, client, or any other dimension that matters to your workflow. To create a tag, hover over an agent in the agent list and click the **Add Tag** button that appears, then select a color, give the tag a name, and click **Save**. Once a tag exists, you can assign it to other agents the same way. You need Editor role to create or manage tags.

### Filtering your agent list

The agent list includes Filter options to help you narrow down your view:

* **Pinned agents** — shows only agents you've pinned.
* **My agents** — shows only agents where you are the original creator.
* **Edited by me** — shows agents you have contributed any version to, including edits to agents created by others.
* **Cloneable agents** — shows agents that have been made cloneable.
* **Embeddable agents** — shows agents that have been made embeddable.

You can also filter by agent type (**Default**, **Phone**, or **Knowledge**). With no Filter option selected, the list shows every agent you have access to in the project.

<Note>
  If you previously relied on "My agents" to find agents you had edited, those agents now appear under "Edited by me" instead. "My agents" now shows only agents you originally created, so your team can more easily distinguish between creation and contribution.
</Note>

### Customizing your agent list view

You can resize the **Agent name** and **Description** columns in the agent list table by dragging the divider between column headers. This lets you adjust how much space each column takes up based on your preferences. Other columns remain fixed-width.

## How to get the most from your Agents?

When inventing or creating an agent, you'd want to utilize domain expertise or model an agent after a Domain expert:

1. **Start with clear objectives**: Define specific tasks and goals for your agents
2. **Provide quality knowledge bases**: Equip agents with comprehensive, up-to-date information
3. **Establish escalation protocols**: Create clear paths for agents to involve humans when needed
4. **Monitor and refine**: Regularly review agent performance and make adjustments
5. **Integrate with existing systems**: Connect agents to your current tools and workflows
6. **Organize for efficiency**: Use pinning and folders to keep your most important Agents easily accessible

By thoughtfully implementing AI agents, organizations can enhance productivity, improve customer experiences, and free human employees to focus on work that requires creativity, empathy, and strategic thinking.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Do I need coding knowledge to Invent an Agent?">
    No, Invent is specifically designed to make agent creation accessible to users without technical expertise. You only need to describe what you want your agent to do in plain English.
  </Accordion>

  <Accordion title="What are tools in the context of AI agents?">
    Tools are capabilities that agents can use to perform specific actions and complete tasks. They extend what an agent can do beyond just conversation. For example, an agent might use tools to:

    * Search the internet for information
    * Access and update customer records in a CRM
    * Send emails or messages
    * Schedule appointments
    * Generate content or analyze data

    Relevance AI provides both pre-built tool templates and the ability to create custom tools for your agents.
  </Accordion>

  <Accordion title="How does Knowledge for Agents work?">
    Knowledge in Relevance AI refers to the information and context provided to agents to help them perform their tasks effectively. This is often implemented through Retrieval Augmented Generation (RAG), which allows agents to access and use specific information sources when responding to queries or completing tasks.

    By giving agents access to knowledge bases, you can ensure they provide accurate, up-to-date information specific to your business, products, or services.
  </Accordion>

  <Accordion title="What is a Multi Agent System (MAS)?">
    Yes, Relevance AI supports creating multi-agent systems where specialized AI agents work together to accomplish complex tasks. This is part of the AI Workforce feature, which provides a visual canvas where you can design, connect, and monitor teams of specialized agents.
  </Accordion>

  <Accordion title="How do agents handle situations they can't resolve?">
    Agents can be configured to escalate issues they can't handle to human team members. For example, if an agent cannot answer a customer's question, it can loop in a sales rep or manager via your preferred communication tool (email, chat, etc.). The human can provide an answer, which the agent can then use to respond to the customer and store for future reference.
  </Accordion>

  <Accordion title="What integrations are available for AI agents?">
    Relevance AI offers numerous integrations that allow agents to connect with external tools and services, including:

    * Email platforms (Gmail, Outlook)
    * CRM systems (Salesforce, HubSpot)
    * Messaging platforms (Slack, WhatsApp, Telegram)
    * Productivity tools (Google Calendar, Google Sheets)
    * Knowledge management systems (Notion, Confluence)
    * And many more

    These integrations enable agents to access data and perform actions across your entire tech stack.
  </Accordion>

  <Accordion title="Can agents operate autonomously without human oversight?">
    Yes, agents on the Relevance AI platform can complete work end-to-end without human involvement. For example, you could create an agent that checks for new information daily and produces reports or updates websites automatically.

    However, you can also configure agents to work in a co-pilot mode, where they assist humans rather than working completely independently. The best approach depends on your specific use case and comfort level with automation.
  </Accordion>

  <Accordion title="What are the system limits or quotas for AI agents?">
    Relevance AI has system quotas that define the maximum resource allocations for different aspects of the platform. These may include limits on the number of agents, conversations, knowledge bases, or API calls depending on your plan. Specific details about these quotas can be found in the [system quotas documentation](/docs/admin/system-limits).
  </Accordion>

  <Accordion title="How do I monitor and improve my agents' performance?">
    Relevance AI provides tools to monitor your agents' performance, including:

    1. **Task View**: A centralized interface for monitoring, managing, and interacting with your AI workforce
    2. **Conversation History**: Review past interactions to identify areas for improvement
    3. **Analytics**: Track metrics related to agent usage and effectiveness

    Regular monitoring helps you identify patterns, optimize agent behavior, and ensure they're meeting your business objectives.
  </Accordion>

  <Accordion title="What is concurrency and how does it affect my agents?">
    Concurrency refers to the number of tasks that can be executed simultaneously by your agents. For more details, visit our [System Quotas page](/docs/admin/system-limits).
  </Accordion>

  <Accordion title="How can I organize my Agents?">
    You can organize your Agents in several ways:

    * **Pin important Agents**: Keep your most frequently used Agents at the top of your agents list in the main app for quick access
    * **Use folders**: Organize Agents into folders based on purpose, department, or use case
    * **Project organization**: Separate Agents across different projects based on your team structure

    These organizational features help you manage large collections of Agents efficiently.
  </Accordion>
</AccordionGroup>

***

## Ready to get started with AI agents?

<CardGroup>
  <Card title="Create an Agent" icon="plus" href="/build/agents/create-an-agent">
    Build your first agent step by step
  </Card>

  <Card title="Use Agents in Chat" icon="messages" href="/get-started/chat/introduction">
    Start using agents instantly in Relevance Chat
  </Card>

  <Card title="Explore the Marketplace" icon="store" href="/get-started/marketplace/introduction">
    Clone and customize pre-built agents
  </Card>
</CardGroup>


# API integration
Source: https://relevanceai.com/docs/get-started/core-concepts/api-integration

Add custom API keys to connect services Relevance doesn't have built-in, or use the Relevance API key to access the platform programmatically.

## What is API integration?

Relevance AI has [built-in integrations](/docs/integrations/introduction) for dozens of popular services — but if you need to call a custom endpoint or connect a service that isn't on the list, you can add your own API keys. And if you want to access Relevance programmatically — via the SDK — you can generate a Relevance API key.

## What are the benefits of API integration?

<CardGroup>
  <Card title="Connect anything" icon="plug">
    Add your own API keys to connect any service or endpoint that Relevance doesn't have a built-in integration for.
  </Card>

  <Card title="Programmatic access" icon="code">
    Use the Relevance API key to interact with the platform through the SDK.
  </Card>

  <Card title="Secure by default" icon="shield-check">
    All keys are encrypted and scoped to your project. Access them safely in tools using template variables.
  </Card>

  <Card title="Full flexibility" icon="sliders">
    Call any REST endpoint, pass custom headers and parameters, and use the response in your tool steps.
  </Card>
</CardGroup>

## How to add a custom API key

If the integration you need isn't available as a built-in, you can add a custom API key:

1. Go to the **Integrations & API Keys** page from the left-hand menu
2. Click **Custom API Keys**
3. Give your key a name, paste the value, and save

<div>
  <iframe title="Add a custom API key" />
</div>

Once saved, you can access it in any tool step using the template variable:

```
{{secrets.chains_your_api_key}}
```

This lets you call any external endpoint from an API step or code step while keeping your credentials secure.

## Relevance API key

The Relevance API key is different from custom API keys — it authenticates *you* against the Relevance platform itself. Use it when you want to:

* **Use the Python or JavaScript SDK** to trigger agents, run tools, or manage resources programmatically
* **Connect agents to external platforms** by embedding them outside of Relevance

<div>
  <iframe title="Create a Relevance API key" />
</div>

To generate one, go to **Integrations & API Keys** and click **Relevance API Keys**.

<Note>
  Users with the **Editor** role or above can generate Relevance API keys. On enterprise plans, users with the **Member** role can also generate their own API key — useful for connecting MCP clients like Claude Desktop or Cursor. Members can only manage their own keys and cannot manage API keys for other users.
</Note>

<Warning>
  API keys are a form of credential — treat them like passwords. Never share them publicly or commit them to source control.
</Warning>

### Region and URL endpoints

<Snippet />

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What's the difference between a custom API key and a Relevance API key?">
    Custom API keys connect Relevance to an external service you want to call. The Relevance API key connects external code to Relevance itself — it's what you use with the SDK.
  </Accordion>

  <Accordion title="How do I use a custom API key in a tool?">
    Reference it with the template variable `{{secrets.chains_your_key_name}}` in any tool step that supports it, such as an API call or Python code step.
  </Accordion>

  <Accordion title="Are my API keys secure?">
    Yes. All keys are encrypted and scoped to your project. They're never exposed in logs or agent responses.
  </Accordion>

  <Accordion title="What about built-in integrations like HubSpot or Slack?">
    Those are covered on the [Integrations](/docs/integrations/introduction) page. This page is for connecting services that don't have a built-in integration, or for accessing Relevance programmatically.
  </Accordion>
</AccordionGroup>

***

## Next steps

<CardGroup>
  <Card title="Explore integrations" icon="puzzle-piece" href="/integrations/introduction">
    See the full list of built-in integrations and how to use them
  </Card>

  <Card title="Build a tool" icon="hammer" href="/build/tools/create-a-tool">
    Create a tool that uses your API keys to call external services
  </Card>
</CardGroup>


# Knowledge
Source: https://relevanceai.com/docs/get-started/core-concepts/knowledge

Give your agents the context they need with Relevance AI's RAG-powered knowledge system.

## What is Knowledge?

Knowledge is Relevance AI's retrieval augmented generation (RAG) system. It lets you give your agents and tools access to specific information — your docs, your data, your expertise — so they can respond with accuracy instead of relying only on their training data.

Think of it as your agent's reference library. When an agent needs to answer a question or complete a task, it searches your knowledge bases for the most relevant information and uses it to ground its response.

## What are the benefits of Knowledge?

<CardGroup>
  <Card title="More accurate responses" icon="bullseye">
    Agents reference your actual data instead of guessing. Fewer hallucinations, more reliable output.
  </Card>

  <Card title="Domain expertise on demand" icon="graduation-cap">
    Turn any agent into a specialist by giving it access to the right knowledge — product docs, playbooks, policies, whatever it needs.
  </Card>

  <Card title="Always up to date" icon="rotate">
    Update your knowledge bases as things change. Your agents automatically use the latest information.
  </Card>

  <Card title="Works across your stack" icon="plug">
    Pull in data from files, websites, Google Drive, SharePoint, Notion, and more.
  </Card>
</CardGroup>

## How Knowledge Works

Knowledge uses semantic search to find the most relevant information for each query. When an agent receives a task or question:

1. It searches your connected knowledge bases for relevant content
2. The most relevant results are passed to the agent as context
3. The agent uses that context to generate an accurate, grounded response

This is powered by retrieval augmented generation (RAG) — your agents get the right information at the right time without needing it baked into their training data.

### Connecting Knowledge to an agent

When you add a knowledge base to an agent, you choose how the agent uses it:

* **Add all to prompt** — the entire knowledge set is included in the prompt every time the agent runs. Best for small or simple datasets.
* **Allow agent to search** — the agent searches the knowledge base using RAG and pulls only what's relevant. Best for large or complex datasets.

## Where is Knowledge most useful?

Knowledge is most effective when your agents need access to specific, proprietary, or frequently changing information:

* **Customer support**: Product documentation, troubleshooting guides, and FAQs so agents resolve issues accurately
* **Sales enablement**: Battlecards, pricing sheets, and case studies to keep sales agents informed
* **Internal operations**: Company policies, SOPs, and handbooks that employees need on demand
* **Onboarding**: Training materials and process documentation for new team members
* **Research and analysis**: Reports, datasets, and reference materials for agents that synthesize information

## How to get the most from Knowledge?

1. **Keep data clean and structured**: Well-organized content with clear headings and sections improves retrieval accuracy.
2. **Be specific with what you upload**: Give agents only the knowledge they need — too much irrelevant data dilutes search results.
3. **Keep it up to date**: Regularly refresh your knowledge bases so agents always reference current information.
4. **Use enrichment**: Run [tools](/docs/build/knowledge/enrich-with-tool) over your knowledge to add summaries, tags, or derived fields that help agents find the right content faster.
5. **Use snippets for quick access**: Set up [snippets](/docs/build/knowledge/use-snippets-for-quick-access) for frequently referenced information your agents use often.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What file types can I upload to Knowledge?">
    You can upload CSV, PDF, Excel, JSON, and audio files. You can also pull content from websites or sync from integrations like Google Drive, SharePoint, and Notion.
  </Accordion>

  <Accordion title="How does semantic search differ from keyword search?">
    Keyword search looks for exact word matches. Semantic search understands meaning — so a search for "pricing information" can find content titled "how much does it cost" even though the words don't match.
  </Accordion>

  <Accordion title="Can multiple agents share the same knowledge base?">
    Yes. You can connect a single knowledge base to as many agents as you like. This is useful when multiple agents need access to the same reference material.
  </Accordion>

  <Accordion title="How do I keep my knowledge base up to date?">
    You can manually update entries, re-upload files, or use integrations like Google Drive and SharePoint that sync automatically when your source documents change.
  </Accordion>

  <Accordion title="What is knowledge enrichment?">
    Enrichment lets you run tools over your knowledge table to add or transform data — for example, generating summaries, extracting key fields, or tagging entries. This makes retrieval more effective and gives agents richer context.
  </Accordion>
</AccordionGroup>

***

## Ready to get started with Knowledge?

<CardGroup>
  <Card title="Create a Knowledge Base" icon="book" href="/build/knowledge/create-knowledge">
    Upload your data and build your first knowledge base
  </Card>

  <Card title="Connect Integrations" icon="plug" href="/build/knowledge/integrated-knowledge-sources/google-drive">
    Sync from Google Drive, SharePoint, Notion, and more
  </Card>
</CardGroup>


# MCP & Plugins
Source: https://relevanceai.com/docs/get-started/core-concepts/mcp-plugins

Build and manage your Relevance AI agents programmatically using Claude Code or any MCP-compatible AI client.

Connect your AI coding environment to Relevance AI to build agents, tools, and workforces directly from your terminal — using natural language. Instead of clicking through a UI, you build, test, and iterate through clients like Claude Code, Cursor, or any MCP-compatible AI client.

## Getting started

Pick the setup that matches your AI client.

<CardGroup>
  <Card title="OpenAI Codex" icon="robot" href="/integrations/mcp/codex">
    Connect Codex to Relevance AI with the MCP server and agent skills.
  </Card>

  <Card title="Claude Code plugin" icon="terminal" href="/integrations/mcp/claude-code">
    The fastest way to get started. Install the Relevance AI plugin for Claude Code and build agents from your terminal.
  </Card>

  <Card title="MCP server" icon="plug" href="/integrations/mcp/mcp-server">
    Connect from any MCP-compatible client — Claude Desktop, Cursor, VS Code, ChatGPT, and more.
  </Card>

  <Card title="Agent skills" icon="graduation-cap" href="/integrations/mcp/agent-skills">
    Clone the agent skills repository to give your AI coding assistant built-in knowledge of Relevance AI.
  </Card>
</CardGroup>

***

<Note>
  This page focuses on what builders can create and manage via MCP. MCP connections also work for team members who only need to run existing agents — not build or edit them. Users with Viewer or Chat project roles are automatically placed into a restricted access mode when they authenticate, so they can view and run agents without any write access. See [Access control](/docs/integrations/mcp/mcp-server#access-control) for details.
</Note>

## What you can do

Once connected, your AI client gets full access to your Relevance AI project. This goes far beyond running existing tools — you can build and manage everything in your project from clients like Claude Code.

<CardGroup>
  <Card title="Create Agents" icon="robot">
    Design and configure new Agents, set their instructions, assign Tools, and configure Triggers.
  </Card>

  <Card title="Build Tools" icon="wrench">
    Create new Tools with custom steps, inputs, and outputs.
  </Card>

  <Card title="Set up Workforces" icon="users">
    Build multi-agent workflows with Triggers, conditions, and agent-to-agent handoffs.
  </Card>

  <Card title="Trigger Agents" icon="play">
    Start conversations with your Agents and get responses.
  </Card>

  <Card title="Execute Tools" icon="bolt">
    Run any of your Relevance AI Tools directly from your AI client.
  </Card>

  <Card title="Troubleshoot Agents" icon="stethoscope">
    Diagnose issues with your Agents by reviewing conversation logs and tool outputs.
  </Card>

  <Card title="Refine Agents" icon="sliders">
    Iterate on Agent instructions, Tool configurations, and behavior based on real results.
  </Card>

  <Card title="Evaluate runs" icon="chart-line">
    Review previous agent runs, identify failures, and improve performance over time.
  </Card>

  <Card title="Update configurations" icon="gear">
    Modify agent instructions, tool settings, and workflow logic.
  </Card>
</CardGroup>

***

## Use cases

<Tabs>
  <Tab title="Build agents">
    Create and configure agents end-to-end from your AI client. Describe what you want in natural language and let your AI client handle the setup.

    **Example prompts:**

    <AccordionGroup>
      <Accordion title="Customer support agent">
        *"Create a new agent called 'Customer Support Bot' that answers questions using our FAQ knowledge base. Give it a friendly tone and make sure it escalates to a human when it can't answer."*
      </Accordion>

      <Accordion title="Lead qualification agent">
        *"Build me an agent that qualifies inbound leads from HubSpot. It should check the company size and industry, then send a personalized follow-up email via Gmail."*
      </Accordion>

      <Accordion title="Slack triage agent">
        *"Set up an agent that monitors our Slack support channel, categorizes messages by urgency, and assigns them to the right team member."*
      </Accordion>

      <Accordion title="Scheduled reporting agent">
        *"Create an agent with a scheduled trigger that runs every morning, pulls yesterday's data from Google Sheets, and posts a summary to Slack."*
      </Accordion>
    </AccordionGroup>
  </Tab>

  <Tab title="Evaluate & improve">
    Review how your agents have been performing by looking at previous conversation runs, identifying where they succeeded or failed, and making targeted improvements.

    **Example prompts:**

    <AccordionGroup>
      <Accordion title="Identify failures">
        *"Pull the last 20 conversations for my Support Agent. Identify any where the agent gave an incorrect answer or failed to resolve the issue."*
      </Accordion>

      <Accordion title="Measure qualification rates">
        *"Look at my Sales Agent's recent runs. How often is it successfully qualifying leads vs. letting unqualified ones through?"*
      </Accordion>

      <Accordion title="Find common struggles">
        *"Review the last week of conversations for my Onboarding Agent. Are there any common questions it struggles with? Suggest improvements to its instructions."*
      </Accordion>

      <Accordion title="Compare before & after">
        *"Compare the performance of my Sales Agent before and after I updated its prompt last Tuesday. Is it doing better at objection handling?"*
      </Accordion>
    </AccordionGroup>
  </Tab>

  <Tab title="Build tools">
    Create custom tools that your agents can use, combining API calls, code steps, LLM processing, and integrations — all from your AI client.

    **Example prompts:**

    <AccordionGroup>
      <Accordion title="Company research tool">
        *"Create a tool that takes a company URL, scrapes the homepage, and returns a one-paragraph summary of what the company does."*
      </Accordion>

      <Accordion title="Knowledge search tool">
        *"Build a tool that searches our knowledge base for the top 3 most relevant articles given a customer question, and formats them as a numbered list with links."*
      </Accordion>

      <Accordion title="Data enrichment tool">
        *"Make a tool that takes a CSV of records, enriches each one with LinkedIn data, and outputs a Google Sheet with the results."*
      </Accordion>

      <Accordion title="Email generator">
        *"Create a tool that generates a personalized email based on a contact's LinkedIn profile and our product's value props."*
      </Accordion>
    </AccordionGroup>
  </Tab>

  <Tab title="Troubleshoot">
    When something isn't working right, dig into agent behavior, tool failures, and configuration problems.

    **Example prompts:**

    <AccordionGroup>
      <Accordion title="Debug triggers">
        *"My Support Agent stopped responding to Slack messages yesterday. Check its trigger configuration and recent conversation logs to figure out what happened."*
      </Accordion>

      <Accordion title="Fix tool errors">
        *"The data enrichment tool is returning empty results. Look at the tool steps and check if the API call is configured correctly."*
      </Accordion>

      <Accordion title="Fix hallucination">
        *"My agent keeps hallucinating answers instead of using the knowledge base. Review its instructions and knowledge configuration and suggest fixes."*
      </Accordion>

      <Accordion title="Audit configurations">
        *"List all my agents and their triggers. I think one of them has a broken webhook — find it and show me the configuration."*
      </Accordion>
    </AccordionGroup>
  </Tab>
</Tabs>

***

## Best practices

<Steps>
  <Step title="Plan before you build">
    Before asking your AI client to create or modify anything, start by having it plan the work first. In Claude Code, you can type `/plan` to enter plan mode — this lets you and Claude align on the approach before any changes are made.

    <Tip>Instead of jumping straight to *"Build me a support agent"*, start with *"Let's plan a support agent that handles inbound Slack messages. What tools will it need? What should the escalation flow look like?"* — then review the plan and tell Claude to execute it.</Tip>
  </Step>

  <Step title="Review before you approve">
    When your AI client proposes changes — like updating an agent's instructions or modifying a tool — read through what it's about to do before confirming. This is especially important for agents that are already live and handling real conversations.
  </Step>

  <Step title="Use conversation history for context">
    When troubleshooting or refining an agent, ask your AI client to pull recent conversation logs first. This gives it real context to work with rather than guessing.

    <Tip>Prompts like *"Look at the last 10 conversations and tell me what's going wrong"* are far more effective than *"My agent isn't working well, fix it"*.</Tip>
  </Step>

  <Step title="Test with real scenarios">
    After building or updating an agent, trigger a test conversation to see how it actually behaves. Don't just review the configuration — run it. Ask your AI client to *"Send a test message to my Support Agent asking about refund policies"* and review the response.
  </Step>

  <Step title="Work across multiple projects deliberately">
    If you have separate projects for development and production, connect to both via separate MCP entries. Build and test in your dev project, then once you're happy, recreate or promote the agent in production. This keeps your live agents safe while you experiment.
  </Step>
</Steps>

***

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What is MCP?">
    The Model Context Protocol (MCP) is an open standard that allows AI clients to connect to external tools and data sources. It provides a standardized way for AI assistants to access your Relevance AI workspace, so you can build and manage agents from clients like Claude Code, Cursor, or VS Code instead of the Relevance AI web interface.
  </Accordion>

  <Accordion title="Do I need to use Claude Code?">
    No. Claude Code with the Relevance AI plugin provides the richest experience, but you can use any MCP-compatible client — Claude Desktop, ChatGPT, Cursor, VS Code, Windsurf, and more. See the [MCP server](/docs/integrations/mcp/mcp-server) page for all supported clients.
  </Accordion>

  <Accordion title="Is it free to use?">
    The MCP server and Claude Code plugin are free. You will be billed for any Relevance AI usage (agent runs, tool executions, etc.) according to your plan.
  </Accordion>

  <Accordion title="Can I use multiple AI clients at the same time?">
    Yes. You can connect to the Relevance AI MCP server from as many clients as you like simultaneously. Each client authenticates independently.
  </Accordion>

  <Accordion title="Does authentication expire?">
    Authentication tokens may expire after a period of inactivity. If you are prompted to re-authenticate, simply follow the login flow again.
  </Accordion>

  <Accordion title="Can I restrict which tools are available?">
    Run-Only mode is the primary way to restrict MCP access. Assign a user the Viewer or Chat project role to force them into a restricted mode automatically — Viewers get run-only access (no write or destructive operations), and Chat users are limited to an even narrower set of tools. Member, Editor, and Admin users can opt into run-only mode from the toggle on the OAuth consent screen. For finer-grained control — such as exposing only a specific subset of agents — organize your tools across separate projects and authenticate each connection to the appropriate project. See [Access control on the MCP server page](/docs/integrations/mcp/mcp-server#access-control) for the full role matrix.
  </Accordion>
</AccordionGroup>


# Tools
Source: https://relevanceai.com/docs/get-started/core-concepts/tools

The actions your agents use to get things done — from API calls to LLM prompts to custom code.

## What are Tools?

Tools are the actions your agents can take. They're step-by-step automations you build in a no-code builder — things like calling an API, running an LLM prompt, sending an email, searching a database, or executing custom code.

You can give tools to your agents so they know how to complete tasks, share them as standalone forms, or run them in bulk across a knowledge table.

## What are the benefits of Tools?

<CardGroup>
  <Card title="Extend what agents can do" icon="puzzle-piece">
    Without tools, agents can only talk. With tools, they can take real actions — update your CRM, send emails, search the web, and more.
  </Card>

  <Card title="No-code builder" icon="wand-magic-sparkles">
    Build tools visually by chaining steps together. No engineering required.
  </Card>

  <Card title="Reusable across agents" icon="clone">
    Build a tool once and give it to any agent. Share it with your team or publish it to the Marketplace.
  </Card>

  <Card title="Full flexibility" icon="code">
    Use pre-built integrations, call any API, or write custom Python. Whatever your workflow needs.
  </Card>
</CardGroup>

## How Tools Work

Every tool follows the same pattern: **inputs → steps → outputs**.

* **Inputs** — the data the tool receives. Text, numbers, files, or data passed from an agent or another tool.
* **Steps** — the actions the tool performs in sequence. Each step can use the output of previous steps. Steps include LLM prompts, API calls, integrations, code execution, and more.
* **Outputs** — what the tool returns when it's done. Text, links, structured data, or files that get passed back to the agent or displayed to the user.

## Where are Tools most useful?

Tools shine anywhere your agents need to take action beyond conversation:

* **CRM updates**: Automatically create, update, or enrich records in HubSpot, Salesforce, and other systems
* **Email and messaging**: Send emails, Slack messages, or notifications as part of a workflow
* **Web search and scraping**: Pull live information from the web to inform agent decisions
* **Data processing**: Transform, filter, or analyze data with code steps or LLM prompts
* **Third-party APIs**: Connect to any external service — payment processors, marketing platforms, internal systems, and more

## How to get the most from your Tools?

1. **Keep tools focused**: Each tool should do one thing well. Break complex workflows into smaller, composable tools.
2. **Write clear descriptions**: Your agent uses the tool's name and description to decide when to use it — make them specific.
3. **Define inputs carefully**: Use descriptive input names and set sensible defaults so agents pass the right data.
4. **Test step by step**: Run individual steps as you build to catch issues early instead of debugging the whole chain.
5. **Reuse and share**: Build tools once, then give them to multiple agents or share with your team via the Marketplace.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Do I need coding knowledge to build a tool?">
    No. The Tool Builder is a no-code interface where you chain steps together visually. If you need custom logic, you can add a code step with Python, but it's entirely optional.
  </Accordion>

  <Accordion title="What's the difference between a tool and an agent?">
    An agent thinks, plans, and decides what to do. A tool is a specific action the agent can take — like calling an API or sending an email. Agents use tools to get work done.
  </Accordion>

  <Accordion title="Can I use the same tool across multiple agents?">
    Yes. Tools are reusable. Build a tool once and assign it to as many agents as you like.
  </Accordion>

  <Accordion title="How do I debug a tool that isn't working?">
    Use the **Logs** tab in the Tool Builder to review past executions. You can see exactly what each step received and returned, making it easy to pinpoint where things went wrong.
  </Accordion>

  <Accordion title="Can I call external APIs from a tool?">
    Yes. You can add an API step to call any REST endpoint, pass headers and parameters, and use the response in subsequent steps.
  </Accordion>
</AccordionGroup>

***

## Ready to get started with Tools?

<CardGroup>
  <Card title="Build a Tool" icon="hammer" href="/build/tools/create-a-tool">
    Follow our guide to create your first tool step by step
  </Card>

  <Card title="Tool Style Guide" icon="palette" href="/build/tools/customize-tool/tool-style-guide">
    Learn best practices for building effective tools
  </Card>
</CardGroup>


# Workforces
Source: https://relevanceai.com/docs/get-started/core-concepts/workforces

Workforce transforms how your agents collaborate by providing a visual canvas where you can design, connect, and monitor teams of specialised agents.

## What is a Workforce?

A Workforce is a team of AI agents that work together to complete multi-step tasks. Instead of building one agent that tries to do everything, you connect specialists — each handling a different part of the process — on a visual canvas.

Think of it like a team of employees: a researcher gathers information, a writer drafts the output, and a reviewer checks the quality. Each agent focuses on what it's best at, and the Workforce handles the handoffs between them.

## What are the benefits of a Workforce?

<CardGroup>
  <Card title="Break down complex tasks" icon="diagram-project">
    Split multi-step processes across specialized agents instead of overloading a single one.
  </Card>

  <Card title="Visual workflow builder" icon="objects-column">
    Design agent teams with drag-and-drop — no code, easy to understand and modify.
  </Card>

  <Card title="Flexible routing" icon="arrows-split-up-and-left">
    Use AI-driven handoffs, fixed sequences, or conditional logic to control how work flows between agents.
  </Card>

  <Card title="End-to-end monitoring" icon="chart-line">
    Track every step of a workflow to see what's working and where things get stuck.
  </Card>
</CardGroup>

## How Workforces Work

### Connection types

When building your agent workflows, you choose how agents hand off to each other:

* **AI connection** — the agent decides when to pass control to another agent based on context. Great when you want the agent to use judgement about whether another specialist is needed.
* **Next step** — a mandatory transition. When one agent finishes, the next one starts automatically. Use this for predictable, sequential processes.

### Tool integration

You can connect tools directly within your workflow:

* Connect tools to agents to extend what they can do
* Link tools to other tools to create processing chains
* Combine agent intelligence with specialized tool functionality

### Conditional logic

Make your workflows smarter with routing rules:

* Create "if this, then that" rules to determine workflow paths
* Set up decision points that route tasks based on specific criteria
* Build adaptive workflows that respond differently depending on inputs or outcomes

## Where are Workforces most useful?

Workforces are ideal when a task requires multiple steps, skills, or handoffs that a single agent can't handle well on its own:

* **Content pipelines**: A researcher gathers information, a writer drafts the content, and a reviewer checks quality before publishing
* **Lead processing**: One agent qualifies inbound leads, another enriches the data, and a third routes them to the right sales rep
* **Customer support escalation**: A frontline agent handles common questions and hands off complex issues to a specialist agent
* **Report generation**: Agents collect data from multiple sources, analyze it, and compile a final report
* **Onboarding workflows**: Coordinate welcome emails, account setup, training material delivery, and check-ins across multiple agents

## How to get the most from your Workforce?

1. **Start with a clear workflow**: Map out your process before building — know which agents handle which steps and how they hand off.
2. **Keep agents focused**: Each agent should have a single responsibility. A researcher shouldn't also be writing and reviewing.
3. **Choose the right connection type**: Use AI connections when the agent needs judgement about handoffs. Use next step for predictable, sequential flows.
4. **Name connections clearly**: Descriptive names document the purpose of each handoff and make your workflow easier to understand.
5. **Test with varied inputs**: Run different scenarios to make sure routing works as expected and edge cases are handled.
6. **Monitor and optimize**: Review workflow performance regularly to identify bottlenecks or communication gaps between agents.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="How many agents can I add to a Workforce?">
    There's no hard limit on the number of agents in a Workforce. However, simpler workflows with clearly defined roles tend to perform better than overly complex ones.
  </Accordion>

  <Accordion title="What's the difference between an AI connection and a next step?">
    An AI connection lets the agent decide when to hand off based on context — it uses judgement. A next step is a mandatory transition that always fires when the previous agent finishes. Use AI connections for flexible routing and next steps for predictable sequences.
  </Accordion>

  <Accordion title="Can I use tools inside a Workforce?">
    Yes. You can connect tools directly to agents within your Workforce, or link tools to other tools to create processing chains.
  </Accordion>

  <Accordion title="Can I reuse agents across multiple Workforces?">
    Yes. The same agent can participate in multiple Workforces. This is useful when a specialist agent — like a data enrichment agent — is needed in several different workflows.
  </Accordion>

  <Accordion title="How do I debug a Workforce that isn't working as expected?">
    Review the workflow execution to see which agent handled each step, what it received, and what it passed on. This helps you identify where the process broke down — whether it's a routing issue, a tool failure, or a prompt that needs refining.
  </Accordion>
</AccordionGroup>

***

## Ready to get started with Workforces?

<CardGroup>
  <Card title="Create a Workforce" icon="diagram-project" href="/build/workforces/create-a-workforce">
    Create a multi-agent workflow on the visual canvas
  </Card>

  <Card title="Watch a Walkthrough" icon="circle-play" href="https://www.youtube.com/watch?v=0vATbdyGJMs">
    See how to build an AI content creation team in Workforce
  </Card>

  <Card title="Use in Chat" icon="messages" href="/get-started/chat/introduction">
    Run your workforce from Relevance Chat
  </Card>
</CardGroup>


# Introduction
Source: https://relevanceai.com/docs/get-started/introduction

Relevance AI is the home of your AI Workforce.

Relevance AI is a low/no-code platform where you can build AI agents and multi-agent teams that autonomously complete tasks, much like human employees.

Whether you're looking to automate customer support, scale your sales operations, or streamline internal workflows, Relevance AI provides the tools and infrastructure to build, deploy, and manage your AI workforce.

<img alt="The Relevance AI platform" />

## Why Relevance AI?

Traditional automation tools require rigid workflows and constant maintenance. AI agents are different—they think, adapt, and learn from experience. With Relevance AI, you can:

* **Deploy agents in minutes** using our no-code builder or pre-built templates
* **Scale on demand** without hiring or training new team members
* **Integrate seamlessly** with your existing tools and workflows
* **Maintain control** with human oversight and approval workflows
* **Customize everything** from agent behavior to escalation protocols

## Quick Links

<CardGroup>
  <Card title="Create Your First Agent" icon="wand-magic-sparkles" href="/get-started/quick-start-guide">
    Follow our step-by-step guide to build and deploy your first AI agent
  </Card>

  <Card title="Try Relevance Chat" icon="messages" href="/get-started/chat/introduction">
    Start using AI agents instantly — no setup required
  </Card>

  <Card title="Watch Video Tutorials" icon="youtube" href="https://www.youtube.com/@relevanceai">
    Learn from our video library covering agents, tools, and best practices
  </Card>

  <Card title="Get Help & Support" icon="life-ring" href="/get-started/support">
    Access our support team, community, and troubleshooting guides
  </Card>
</CardGroup>

## Core Features

<CardGroup>
  <Card title="Agents" icon="robot" href="/get-started/core-concepts/agents">
    AI entities that autonomously complete tasks like human employees
  </Card>

  <Card title="Chat" icon="messages" href="/get-started/chat/introduction">
    Built-in chat interface for human-like conversations with your agents
  </Card>

  <Card title="Workforces" icon="users-gear" href="/get-started/core-concepts/workforces">
    Multi-agent teams where specialized AI agents collaborate on complex tasks
  </Card>

  <Card title="Knowledge" icon="brain" href="/build/knowledge/create-knowledge">
    RAG solution giving agents access to specific information beyond pre-trained knowledge
  </Card>

  <Card title="Tools" icon="screwdriver-wrench" href="/get-started/core-concepts/tools">
    No-code workflow builder for integrations and automations
  </Card>

  <Card title="Marketplace" icon="store" href="/get-started/marketplace/introduction">
    Pre-built agents, tools, and workforces you can clone and customize
  </Card>
</CardGroup>

## Build your first agent

Follow along to see how to create an agent from scratch:

<div>
  <iframe title="Build an AI agent from scratch in Relevance AI" />
</div>

## How It Works

Relevance AI gives you everything you need to build, connect, and scale an AI workforce:

1. **Build agents** — create AI agents that autonomously complete tasks. Use Invent to generate one from a description, clone a pre-built agent from the Marketplace, or build from scratch.
2. **Give them tools** — equip agents with actions like sending emails, updating your CRM, searching the web, or calling any API — all built in a no-code tool builder.
3. **Add knowledge** — connect your data so agents respond with accuracy. Upload files, sync from Google Drive, SharePoint, Notion, or pull from websites.
4. **Connect them into workforces** — link multiple agents together on a visual canvas to handle complex, multi-step workflows. Each agent focuses on what it's best at, and the workforce handles the handoffs.
5. **Set guardrails and deploy** — configure escalations, approval workflows, and triggers, then put your workforce to work.

## Popular Use Cases

<AccordionGroup>
  <Accordion title="Customer Support" icon="headset">
    Deploy AI agents that handle routine inquiries, troubleshoot common issues, and escalate complex problems to human agents. Reduce response times and free your support team to focus on high-value interactions.
  </Accordion>

  <Accordion title="Sales & Lead Qualification" icon="chart-line">
    Automate prospect research, lead qualification, and initial outreach. AI agents can engage with prospects, gather information, and identify promising leads for your sales team to close.
  </Accordion>

  <Accordion title="Content Creation" icon="pen-to-square">
    Generate drafts of routine documents, reports, social media posts, and communications. Maintain your brand voice while scaling content production.
  </Accordion>

  <Accordion title="Data Analysis & Reporting" icon="chart-mixed">
    Process and summarize large volumes of information to extract actionable insights. Automate report generation and data visualization.
  </Accordion>

  <Accordion title="Scheduling & Coordination" icon="calendar-days">
    Manage calendars, set up meetings, send reminders, and coordinate across teams. Eliminate scheduling conflicts and reduce administrative overhead.
  </Accordion>
</AccordionGroup>

## Learning Resources

<CardGroup>
  <Card title="Community" icon="users" href="https://community.relevanceai.com">
    Join our Community to connect with other users and get help from the Relevance AI team
  </Card>

  <Card title="Blog & Updates" icon="newspaper" href="https://relevanceai.com/blog">
    Stay updated with the latest features, best practices, and customer stories
  </Card>
</CardGroup>

***

Ready to build your AI Workforce? [Start building now!](https://app.relevanceai.com/auth)


# Marketplace
Source: https://relevanceai.com/docs/get-started/marketplace/introduction

Discover the world's top AI Agents

<img alt="Hero image of the Marketplace" />

The Relevance AI Marketplace is a one-stop shop for finding Agents and Tools that match your needs. It offers a curated selection of ready-to-use AI agents designed to automate tasks, streamline workflows, and enhance productivity across a wide range of use cases. Whether you're looking to solve a specific problem or explore what's possible with AI, the Marketplace makes it easy to get started.

## Who can add Agents / Tools to the Marketplace?

Agents and Tools in the Marketplace have been created by Relevance AI employees (Rellies) and Relevance Builders from our community.

Agents and Tools can be submitted by verified Relevance Builders from our community. The Relevance Builder program is currently exclusive to ensure high-quality Marketplace listings. Learn more in our [Relevance Builders guide](/docs/get-started/marketplace/relevance-builders/become-a-relevance-builder).

<Note>
  **Note for Enterprise users:** If you're part of an Enterprise organization with Role-Based Access Controls (RBAC), your ability to become a Relevance Builder and submit to the Marketplace depends on your assigned permissions. Users with Viewer or Chat-only roles at the project level, or those with limited asset-level permissions, may not be able to create or submit Marketplace listings. Contact your organization administrator if you need elevated permissions.
</Note>

## Do I have to pay for Agents / Tools?

Some Agents and Tools in the Marketplace require payment. Pricing is determined by the creator of each listing. If you purchase a paid Agent or Tool and it doesn't perform as expected, you can request a refund within seven days of your purchase. Refunds are not available after this seven-day window.

We recommend first reaching out to our support team to get help before pursuing a refund, to make sure we can help you use your purchased listing first.

## I need help using an Agent / Tool I've cloned from the Marketplace

Please [reach out to our support team](/docs/get-started/support) for assistance. If you purchased an Agent / Tool, we can help contact them on your behalf if you need their help.

## How to access your purchases on the Marketplace

<div>
  <iframe title="How to access your purchases on the Marketplace" />
</div>

To view and re-clone listings you've already purchased:

1. Click 'Marketplace' in the sidebar
2. Click 'My Purchases' in the top right
3. You can then view your purchases on this page and select listings you've already purchased if you need to clone them into your project again

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Are purchases tied to a specific project?">
    Yes. When you purchase a listing, it's tied to the specific project where you made the purchase. If you purchase a listing on one project, you won't see it in My Purchases on another project.
  </Accordion>

  <Accordion title="I purchased a listing on the Marketplace, can I clone it to other projects?">
    Not always. Relevance Builders can choose to make their listings clonable or not. If they have made their listing clonable, you can clone it to any of your projects. If they have not made their listing clonable, you can only use it in the project you purchased it in. Most Builders restrict cross-project cloning for paid listings to ensure proper licensing per project.
  </Accordion>

  <Accordion title="Can I modify Agents or Tools I've purchased from the Marketplace?">
    Yes, you have full access to modify any Agent or Tool you've purchased. You can customize prompts, add or remove tools, adjust settings, and make any other changes to fit your specific needs.
  </Accordion>

  <Accordion title="How do I know if a listing is free or paid before purchasing?">
    All pricing information is clearly displayed on each Marketplace listing. Free listings are marked as "Free" while paid listings show the price. You'll see the cost before confirming your purchase.
  </Accordion>

  <Accordion title="Why don't I see free listings in My Purchases?">
    The My Purchases page only displays paid listings you've purchased. Free listings can be cloned again at any time directly from the Marketplace without needing to track them in your purchase history.
  </Accordion>
</AccordionGroup>

***

## What's next?

<CardGroup>
  <Card title="Use in Chat" icon="messages" href="/get-started/chat/introduction">
    Once you've cloned an agent, use it instantly in Relevance Chat
  </Card>

  <Card title="Become a Builder" icon="hammer" href="/get-started/marketplace/relevance-builders/become-a-relevance-builder">
    Submit your own agents and tools to the Marketplace
  </Card>
</CardGroup>


# Become a Relevance Builder
Source: https://relevanceai.com/docs/get-started/marketplace/relevance-builders/become-a-relevance-builder

Learn how to become a Relevance Builder and start sharing your Agents and Tools in the Marketplace

<Note>
  The Relevance Builder program is now exclusive to verified builders to maintain high-quality Marketplace listings. New builder applications are not currently being accepted.
</Note>

## What is a Relevance Builder?

Relevance Builders are approved builders who are able to submit listings to the Relevance Marketplace. Once you set up your Relevance Builder profile, you will also be given access to our Community discussion space for Relevance Builders, and be tagged as a Relevance Builder in the Community when posting Agents, answering questions, etc.

You can start submitting Agents and Tools to the Marketplace right away. Simply navigate to the Relevance Builders section in your project to set up your profile and begin submitting listings.

<Info>
  Want to learn more about the Marketplace? Check out our [Marketplace overview](/docs/get-started/marketplace/introduction) to discover how it works and what's available.
</Info>

## What's in this guide?

In this guide, you'll:

<CardGroup>
  <Card title="Set up your Builder profile" icon="user-gear" href="/get-started/marketplace/relevance-builders/setup-builder-profile">
    Learn how to create your Relevance Builder profile and connect your Stripe account for payments
  </Card>

  <Card title="Submit Agents to the Marketplace" icon="upload" href="/get-started/marketplace/relevance-builders/submit-agents">
    Understand our approval process and requirements for submitting Agents and Tools
  </Card>

  <Card title="Get paid for your Agents" icon="dollar-sign" href="/get-started/marketplace/relevance-builders/getting-paid">
    Learn about pricing, payments, and how to maximize your earnings
  </Card>

  <Card title="Support your Agents" icon="life-ring" href="/get-started/marketplace/relevance-builders/supporting-agents">
    Find out about support expectations and engaging with the Relevance AI Community
  </Card>
</CardGroup>

## Benefits of becoming a Relevance Builder

<AccordionGroup>
  <Accordion title="Monetize your AI expertise" icon="money-bill-wave">
    Create and sell Agents and Tools to the Relevance AI community. Set your own prices and earn revenue from your creations.
  </Accordion>

  <Accordion title="Build your reputation" icon="star">
    Showcase your skills and build credibility within the Relevance AI ecosystem. Get recognized as a Relevance Builder in the Community.
  </Accordion>

  <Accordion title="Help others succeed" icon="hands-helping">
    Share your solutions with others facing similar challenges. Your Agents can help businesses automate workflows and increase productivity.
  </Accordion>

  <Accordion title="Join a community of builders" icon="users">
    Connect with other Relevance Builders, share knowledge, and collaborate on innovative solutions.
  </Accordion>
</AccordionGroup>

## Ready to get started?

Head to the next page to learn how to [set up your Relevance Builder profile](/docs/get-started/marketplace/relevance-builders/setup-builder-profile) and start submitting to the Marketplace!


# Getting paid for your Agents
Source: https://relevanceai.com/docs/get-started/marketplace/relevance-builders/getting-paid

Learn about pricing, payments, and maximizing your earnings from Marketplace listings

If you have set up your Stripe account, you will be able to list Agents for a price of your choosing.

## How big a cut does Relevance AI take?

Relevance currently does not take any cut from the payments you receive for Marketplace listings. The small Stripe transaction fee will however come out of your fee.

<Note>
  This means you keep 100% of your listing price (minus Stripe's standard processing fees, which are typically around 2.9% + \$0.30 per transaction).
</Note>

## How much should I charge?

The price you set for an agent is up to you, though we have an absolute limit of no more than \$1000 USD. We recommend setting a lower price to ensure uptake and demand, to grow your clone count and build trust before increasing prices.

### Pricing strategies

<AccordionGroup>
  <Accordion title="Start with free or low-cost listings" icon="gift">
    When you first get started, consider listing your Agents for free or at a lower price to increase your clone count. After a while, you can then start to increase your prices as you build trust within the Relevance AI community.
  </Accordion>

  <Accordion title="Consider the value provided" icon="scale-balanced">
    Price your Agent based on the value it provides:

    * **Simple automation** (\$5 - \$25): Basic task automation or simple workflows
    * **Moderate complexity** (\$25 - \$100): Multi-step processes with integrations
    * **Advanced solutions** (\$100 - \$500): Complex workflows with multiple tools and sophisticated logic
    * **Enterprise-grade** (\$500 - \$1000): Highly specialized, industry-specific solutions
  </Accordion>

  <Accordion title="Research similar listings" icon="magnifying-glass">
    Browse the Marketplace to see how similar Agents are priced. This will help you position your listing competitively.
  </Accordion>

  <Accordion title="Offer tiered pricing" icon="layer-group">
    Consider creating multiple versions of your Agent at different price points:

    * Basic version (free or low-cost)
    * Pro version with additional features
    * Enterprise version with premium capabilities
  </Accordion>
</AccordionGroup>

## If someone buys my Agent, do they still have to pay for a Relevance AI subscription?

In most cases — yes. While our free tier gives customers access to 100 credits per day, if they need more than this to run your Agent and any others they have, they'll need to upgrade.

Purchasing listings from the Marketplace would then be an additional cost.

## Can I earn money from referring customers to the platform?

Sure! You can join our affiliate program at [https://relevance-ai.getrewardful.com/signup](https://relevance-ai.getrewardful.com/signup)

You can use this link to refer customers to Relevance AI, who can then purchase your Agents from the Marketplace.

## How can I get more people to buy my Agents and Tools?

### Set your Agents to a lower price / free to start

When you first get started, it's a good idea to list your Agents for a lower price to increase your clone count. After a while, you can then start to increase your prices as you build trust within the Relevance AI community.

### Promote your listings!

The best way to drive more customers to your listings is to promote them yourself. Here are some ideas on how to promote your listings:

<CardGroup>
  <Card title="Share in the Community" icon="users" href="https://community.relevanceai.com">
    Post them to our Community in the Agent Templates section, showing a demo of the Agent, and any help documentation you've created
  </Card>

  <Card title="Social media promotion" icon="share-nodes">
    Share them on your own social media platforms like YouTube, LinkedIn, TikTok, X (Twitter), etc.
  </Card>

  <Card title="Help others in the Community" icon="hands-helping">
    Help people out in the Community Questions section. If you see someone who would benefit from one of your listings, share it with them!
  </Card>

  <Card title="Create tutorials and demos" icon="video">
    Create video tutorials or written guides showing how to use your Agent. This builds trust and demonstrates value.
  </Card>
</CardGroup>

### Additional promotion strategies

<AccordionGroup>
  <Accordion title="Build a portfolio" icon="folder-open">
    Create a collection of related Agents that work together. Customers who buy one may be interested in others.
  </Accordion>

  <Accordion title="Gather testimonials" icon="quote-left">
    Ask satisfied users for feedback and testimonials. Display these in your listing descriptions or promotional materials.
  </Accordion>

  <Accordion title="Offer excellent support" icon="life-ring">
    Provide quick, helpful responses to questions about your Agents. Good support leads to positive reviews and word-of-mouth referrals.
  </Accordion>

  <Accordion title="Keep your Agents updated" icon="arrows-rotate">
    Regularly update your Agents with improvements and new features. This shows you're actively maintaining your listings.
  </Accordion>

  <Accordion title="Create documentation" icon="book">
    Provide comprehensive documentation for your Agents. This can be in the form of:

    * Public Notion pages
    * Video walkthroughs
    * Step-by-step guides
    * FAQ sections
  </Accordion>
</AccordionGroup>

## Payment and refund policies

### When do I receive payments?

Payments are processed through Stripe according to their standard payout schedule. Typically, funds are transferred to your bank account within 2-7 business days after a purchase, depending on your country and bank.

### How do refunds work?

Relevance has a 1 week, no questions asked, refund policy for purchased Marketplace listings. When a user refunds a listing, they will lose access to that listing and will not be able to use it. This money will be taken out of your Stripe account.

After 1 week, our policy is not to offer refunds. This means customers will need to assess within a week whether or not they wish to continue using your listing.

<Tip>
  To minimize refunds, make sure your Agent descriptions are accurate and comprehensive, and provide clear usage instructions. The better customers understand what they're getting, the less likely they are to request refunds.
</Tip>

## Tracking your earnings

You can track your earnings and sales through:

1. **Stripe Dashboard**: View detailed transaction history, payouts, and analytics
2. **Relevance Builders section**: See how many times your Agents have been cloned
3. **Community engagement**: Monitor discussions about your Agents to gauge interest and satisfaction

## Next steps

Learn about [supporting your Agents](/docs/get-started/marketplace/relevance-builders/supporting-agents) and engaging with the Relevance AI Community to build trust and increase sales.


# Set up your Relevance Builder profile
Source: https://relevanceai.com/docs/get-started/marketplace/relevance-builders/setup-builder-profile

Learn how to create your Relevance Builder profile and connect your Stripe account

<Note>
  This guide is for approved Relevance Builders. If you don't see the Relevance Builders section in your sidebar, you don't currently have builder access.
</Note>

Your Relevance Builders profile is where you'll manage your Marketplace listings and receive payments. This guide will walk you through setting up your profile and connecting your Stripe account.

## Setting up your profile

Your Relevance Builders profile will be connected to one of your projects — we recommend creating a new project, adding all of your Agents that you want to list to this project, and creating your profile from here.

<iframe />

<Steps>
  <Step title="Choose your project">
    Select the project you want to create the Agents you'll list on. We recommend creating a dedicated project for your Marketplace listings.
  </Step>

  <Step title="Navigate to Relevance Builders">
    Click on 'Relevance Builders' in the sidebar menu.
  </Step>

  <Step title="Access your profile">
    Head to the 'My Profile' tab.
  </Step>

  <Step title="Set your profile name">
    Enter your **profile name** — this will show up on anything you publish. Choose a name that represents you or your brand.
  </Step>

  <Step title="Create your profile">
    Click 'Create profile' to finish!
  </Step>
</Steps>

<Info>
  **Support for profile pictures and cover images is coming soon!**
</Info>

***

## Connecting to a Stripe account

If you wish to receive payments for paid listings, you'll need to connect to a Stripe account. If you only wish to add free Agents and Tools to the Marketplace, you do not need to follow this step.

<iframe />

<Steps>
  <Step title="Navigate to My Profile">
    Head to the 'My Profile' section of the 'Relevance Builders' section of Relevance AI, and click 'Get started' next to 'Link Stripe account'.
  </Step>

  <Step title="Select your details">
    Select your country and business type on the pop-up screen.
  </Step>

  <Step title="Generate Stripe link">
    Click 'Generate Stripe link' to proceed.
  </Step>

  <Step title="Complete Stripe setup">
    This will take you to Stripe to set up your Stripe account. You will need to fill in any required information for the country and business type you've selected.
  </Step>
</Steps>

<Note>
  Make sure to complete all required fields in Stripe to ensure you can receive payments. Stripe may require additional verification depending on your country and business type.
</Note>

## What information does Stripe require?

The information required by Stripe varies depending on your country and business type. Common requirements include:

<AccordionGroup>
  <Accordion title="Personal information" icon="user">
    * Full legal name
    * Date of birth
    * Address
    * Phone number
    * Email address
  </Accordion>

  <Accordion title="Business information (if applicable)" icon="building">
    * Business name
    * Business type (sole proprietorship, LLC, corporation, etc.)
    * Business address
    * Tax ID or EIN
  </Accordion>

  <Accordion title="Banking information" icon="bank">
    * Bank account number
    * Routing number (US) or equivalent
    * Bank account holder name
  </Accordion>

  <Accordion title="Identity verification" icon="id-card">
    Stripe may require you to upload identification documents such as:

    * Government-issued ID (passport, driver's license)
    * Proof of address (utility bill, bank statement)
  </Accordion>
</AccordionGroup>

## Troubleshooting

<AccordionGroup>
  <Accordion title="I can't see the Relevance Builders section">
    Make sure you have the appropriate permissions in your project. If you're part of an Enterprise organization with RBAC enabled, you may need elevated permissions. Contact your organization administrator for assistance.
  </Accordion>

  <Accordion title="My Stripe connection failed">
    Try the following:

    * Ensure you've completed all required fields in Stripe
    * Check that your browser isn't blocking pop-ups
    * Try using a different browser
    * Contact our support team if the issue persists
  </Accordion>

  <Accordion title="Can I change my profile name later?">
    Yes, you can update your profile name at any time from the 'My Profile' section. Note that this will update the name displayed on all your existing listings.
  </Accordion>

  <Accordion title="Can I use the same Stripe account for multiple projects?">
    Yes, you can connect the same Stripe account to multiple Relevance Builder profiles across different projects.
  </Accordion>
</AccordionGroup>

## Next steps

Now that your profile is set up, you're ready to start submitting Agents to the Marketplace! Head to the next page to learn about [how to submit Agents](/docs/get-started/marketplace/relevance-builders/submit-agents).


# Submit Agents to the Marketplace
Source: https://relevanceai.com/docs/get-started/marketplace/relevance-builders/submit-agents

Learn how to submit Agents and Tools to the Marketplace and meet approval requirements

Once your Relevance Builders profile is set up and ready to go, you can start submitting Agents!

<iframe />

## How to submit your Agent

To submit an Agent to the Marketplace:

<Steps>
  <Step title="Navigate to your Agent">
    Go to the Agent you want to submit in your project.
  </Step>

  <Step title="Open submission">
    Click the "Submit to Marketplace" button or navigate to the Relevance Builders section and select "Submit new listing".
  </Step>

  <Step title="Fill in listing details">
    Enter a title and description for your listing. Make sure these meet the requirements outlined below.
  </Step>

  <Step title="Set pricing">
    Choose whether your listing will be free or paid. If paid, set your price (up to \$1000 USD).
  </Step>

  <Step title="Submit for review">
    Click "Submit" to send your Agent for review by the Relevance team.
  </Step>
</Steps>

## Is there an approval process for submitted agents?

All agents submitted to the marketplace will be reviewed by the internal Relevance team. Agents must follow our agent building best practices to ensure all listed agents are usable, reliable, and valuable to the community. If your agent requires updating, the Relevance team will reach out with specific and actionable feedback to update your submission so that your agent can be approved the next time you submit it.

### Can I push changes and updates to your Agents and Tools once they are approved?

Yes you can!

To do this, head to the Agent / Workforce / Tool that has been submitted to Marketplace, make changes and then click 'Update Marketplace listing'.

The new changes will be sent for approval, and once approved, your Marketplace listing will point to the new version of your Agent / Workforce / Tool.

***

## Approval requirements for submitting marketplace Agents

You can use the below list as a checklist to tick off before submitting your agent, to make sure it meets the approval criteria.

### Do not include Sub-Agents

Agents that use Sub-Agents will not be approved, as Sub-Agents is now a legacy feature. Instead, we suggest that you wait until Workforces can be submitted (coming soon).

### Listing title and description

When you submit an Agent, you will enter a custom title and description.

<AccordionGroup>
  <Accordion title="Title requirements" icon="heading">
    The title of your Agent must clearly describe what the focus of the Agent is. This should not be longer than 50 characters. You can give the Agent a name if you like, but this is not required.

    **Examples**: "Atlas, the Account Analyst", "Google Calendar Meeting Notetaker"
  </Accordion>

  <Accordion title="Description requirements" icon="align-left">
    Your description should start with a short paragraph outlining exactly what your agent does and how it works. After the initial paragraph, you can add some more detail on how the agent works if you would like. You can also link to publicly accessible help docs (e.g. public Notion pages) if you would like - this will help with your agent's supportability and help get it approved.
  </Accordion>
</AccordionGroup>

### Agent core prompt

In order to help the user understand what the agent is doing, we recommend you leave comments throughout your core prompt to explain what the agent is doing. You can also leave comments detailing how the user can edit, change, or extend their agent, such as by adding knowledge or suggested triggers.

The agent core prompt should be optimised to ensure reliable accuracy and performance. This responsibility is up to you to ensure the agent is performing well and the core prompt following prompt engineering best practices.

You can also use comments to link to publicly accessible help docs, or video demos, about how to use the Agent.

### Tools that require OAuth

If your agent has tools that require OAuth to run, you *must* make sure these settings are surfaced as **required tool inputs**. This will allow the user to fill these out when the 'missing tool inputs' modal is shown while setting up the agent.

Agents that use tools that require OAuth within tool steps that are not surfaced as inputs will not be approved.

### Using API keys

If you have a tool that requires an API key to work, and we do not have a native OAuth solution for that API key, you will need to create a tool input of type 'Text', leave it empty, and provide instructions in the input description on where and how the user can create the API key. For example, you can include a link to relevant docs from the 3rd party tool explaining how to create an API key.

### Tool inputs

<AccordionGroup>
  <Accordion title="Excluded types" icon="ban">
    At this stage we do not accept tools which use:

    * The Knowledge feature (in the left side menu in tool builder)
    * Have knowledge table type input fields
  </Accordion>

  <Accordion title="Descriptions" icon="circle-info">
    Please make sure that the description fields for all tool inputs clearly describe exactly how a user or agent should fill in this input. These descriptions are used by the agent to know how to fill a field reliably, so make sure any critical requirements or guidelines are included in the input description.
  </Accordion>
</AccordionGroup>

### Agent variables

Agent variables will be copied over when a marketplace listing is cloned, as long as they are not knowledge tables, API keys or OAuth accounts.

It is best practice to save any information in the agent core prompt that can be customised for each user as a variable. Good examples include company name, company descriptions, output formatting guidelines, etc.

***

## What do I do if my agent is not approved?

If your agent is not approved, the Relevance team will reach out with specific and actionable feedback. You'll need to make the requested changes to your Agent and then resubmit it to the Marketplace for review.

***

## What doesn't get copied over when users clone your listing?

<AccordionGroup>
  <Accordion title="Knowledge" icon="brain">
    We currently do not copy over any knowledge tables along with your listing. We plan to add this soon, but in the meantime, we suggest creating a variable and add the knowledge content to that variable. You can then add that variable to the agent's core instructions.
  </Accordion>

  <Accordion title="Triggers" icon="bolt">
    Triggers will not get carried over, but we will suggest triggers for the user to set up based on the triggers attached to the agent when you submit it - so make sure you have triggers set up when you submit if you think the agent needs them.
  </Accordion>

  <Accordion title="Snippets" icon="scissors">
    Please use agent variables rather than global snippets, as global snippets don't copy over.
  </Accordion>

  <Accordion title="Secrets" icon="key">
    If you have an agent that uses secrets, these should be surfaced as tool inputs of type API key or OAuth so the user can fill them out when setting up the agent. Do not use secrets.
  </Accordion>

  <Accordion title="Default values for specific tool input / variable types" icon="sliders">
    The following tool input or variable types will not have their default values carried over:

    * Knowledge table (avoid using this input type)
    * API key (the user will have to fill this in)
    * OAuth account (the user will have to fill this in)
  </Accordion>
</AccordionGroup>

***

## Submission checklist

Before submitting your Agent, make sure you've completed the following:

* **Guide for using Agent**: Added a 'Guide for using agent' in Advanced Settings that tells users how to use the agent or provides examples to test with
* **No Sub-Agents**: Confirmed your Agent doesn't use Sub-Agents
* **Clear title and description**: Written a clear, descriptive title (under 50 characters) and created a comprehensive description with usage instructions
* **Optimized core prompt**: Added comments explaining what the Agent does, included instructions for customization, and followed prompt engineering best practices
* **OAuth and API keys**: Surfaced all OAuth requirements as required tool inputs and provided clear instructions for any API keys needed
* **Tool inputs**: Avoided using Knowledge or knowledge table inputs and written clear descriptions for all tool inputs
* **Variables**: Used agent variables for customizable information and avoided using global snippets or secrets

## Next steps

Once your Agent is approved, you can start earning from your listings! Learn more about [getting paid for your Agents](/docs/get-started/marketplace/relevance-builders/getting-paid).


# Supporting your Agents
Source: https://relevanceai.com/docs/get-started/marketplace/relevance-builders/supporting-agents

Learn about support expectations and engaging with the Relevance AI Community

As a Relevance Builder, understanding your support responsibilities and engaging with the community will help you build trust and increase the success of your Marketplace listings.

## Do I need to support my Agents / Tools?

You will not need to support your *free* Agents / Tools. Any free Agents / Tools that you add to the Marketplace will be supported by our Relevance AI support team, but if someone asks about them in the Community, we'd appreciate if you could help out and answer questions.

For paid Agents / Tools, our support team will attempt to troubleshoot with your customers first. However, our support team may reach out to you, or direct your customers to you, if they're unable to help your customers with your Agents.

<Note>
  **Best practice**: Even for free listings, being responsive to questions in the Community helps build your reputation and can lead to more downloads and purchases of your other listings.
</Note>

## How do refunds work?

Relevance has a 1 week, no questions asked, refund policy for purchased Marketplace listings. When a user refunds a listing, they will lose access to that listing and will not be able to use it. This money will be taken out of your Stripe account.

After 1 week, our policy is not to offer refunds. This means customers will need to assess within a week whether or not they wish to continue using your listing.

<Tip>
  To minimize refunds, provide clear and accurate descriptions of what your Agent does, include comprehensive usage instructions, respond quickly to support questions, and keep your Agents well-maintained and updated.
</Tip>

## Joining the Community

It is expected that Relevance Builders are also passionate members of our Relevance AI Community. Once you set up your Relevance Builder profile, you will be given a badge in the Community that shows your status as a Relevance Builder.

### Share your listings

You are encouraged to post any listings you submit to the Marketplace in our [Agent Templates](https://community.relevanceai.com/c/agent-templates) section. This will bring more attention to your listing, and allow you to chat with customers who download / purchase your listing directly. We also encourage you to share your listings on social media (LinkedIn, YouTube, TikTok, etc.).

### Answer questions

We would also appreciate you keeping an eye out for questions related to your listings in the [Questions](https://community.relevanceai.com/c/questions) section. We may tag you in questions if we believe they're about your listings if we can't solve these questions ourselves.

### Relevance Builders space

You'll also have access to a dedicated space for Relevance Builders in the Community. Join the Community to access this space where you can discuss with fellow Relevance Builders and the Relevance AI team, share your listings and learnings, and leave feedback for the Relevance AI team.

<Info>
  Relevance Builders who spend more time in the Community build more trust with our customer base, and therefore have more success in our Marketplace.
</Info>

## Best practices for supporting your Agents

<AccordionGroup>
  <Accordion title="Provide comprehensive documentation" icon="book">
    Create detailed documentation for your Agents including:

    * Step-by-step setup instructions
    * Common use cases and examples
    * Troubleshooting guides
    * FAQ sections

    You can host this on public Notion pages, GitHub, or your own website, and link to it from your listing description.
  </Accordion>

  <Accordion title="Respond promptly to questions" icon="clock">
    Try to respond to questions about your Agents within 24-48 hours. Quick responses show you're actively supporting your listings and build customer confidence.
  </Accordion>

  <Accordion title="Create video tutorials" icon="video">
    Video walkthroughs can be incredibly helpful for users. Consider creating:

    * Setup and configuration videos
    * Use case demonstrations
    * Troubleshooting guides

    Share these on YouTube and link to them from your listing descriptions.
  </Accordion>

  <Accordion title="Keep your Agents updated" icon="arrows-rotate">
    Regularly review and update your Agents to:

    * Fix any bugs or issues
    * Add new features based on user feedback
    * Ensure compatibility with platform updates
    * Improve performance and reliability
  </Accordion>

  <Accordion title="Be proactive about known issues" icon="triangle-exclamation">
    If you discover an issue with your Agent:

    * Communicate it clearly to users
    * Provide workarounds if available
    * Update the Agent as soon as possible
    * Notify users when the issue is resolved
  </Accordion>

  <Accordion title="Gather and act on feedback" icon="comments">
    Actively seek feedback from users and use it to improve your Agents. This shows you're committed to providing value and helps you create better listings.
  </Accordion>
</AccordionGroup>

## Community engagement strategies

<CardGroup>
  <Card title="Share your expertise" icon="graduation-cap">
    Answer questions in the Community, even if they're not directly about your listings. This establishes you as a knowledgeable builder.
  </Card>

  <Card title="Showcase use cases" icon="lightbulb">
    Post creative ways people are using your Agents. This inspires others and demonstrates the value of your listings.
  </Card>

  <Card title="Collaborate with other Builders" icon="handshake">
    Connect with other Relevance Builders to share knowledge, collaborate on projects, and cross-promote listings.
  </Card>

  <Card title="Participate in discussions" icon="message">
    Engage in broader platform discussions. Your insights help shape the community and increase your visibility.
  </Card>
</CardGroup>

## Support workflow for paid listings

When a customer has an issue with your paid listing:

<Steps>
  <Step title="Customer contacts Relevance support">
    The customer will typically reach out to our support team first.
  </Step>

  <Step title="Initial troubleshooting">
    Our support team will attempt to troubleshoot the issue using your documentation and their knowledge of the platform.
  </Step>

  <Step title="Escalation to Builder">
    If the issue requires your expertise, our support team will either:

    * Reach out to you directly for assistance
    * Direct the customer to contact you
    * Tag you in a Community post about the issue
  </Step>

  <Step title="Resolution">
    Work with the customer to resolve the issue. If it's a bug in your Agent, update the listing and push the changes.
  </Step>

  <Step title="Follow-up">
    After resolving the issue, follow up to ensure the customer is satisfied and consider if the issue reveals a need for better documentation or Agent improvements.
  </Step>
</Steps>

## Measuring your success

Track these metrics to understand how well you're supporting your Agents:

* **Clone count**: How many times your Agents have been downloaded
* **Refund rate**: Percentage of purchases that result in refunds
* **Community engagement**: Questions answered, posts made, discussions participated in
* **Customer feedback**: Reviews, testimonials, and direct feedback
* **Update frequency**: How often you improve and update your listings

<Tip>
  Builders with low refund rates, high community engagement, and regular updates tend to have the most successful Marketplace presence.
</Tip>

## Getting help

If you need assistance with supporting your Agents or have questions about your responsibilities as a Builder:

* **Relevance Builders space**: Ask questions in the private Builders community
* **Support team**: Reach out to our support team for guidance
* **Documentation**: Refer to this guide and other documentation resources

***

## Congratulations!

You've completed the Relevance Builders guide! You now have everything you need to:

* Set up your Builder profile
* Submit Agents to the Marketplace
* Price and promote your listings
* Support your customers effectively

Ready to start building? Head to the [Relevance Builders section](https://app.relevanceai.com) in your project and create your first Marketplace listing!

<Card title="Join the Community" icon="users" href="https://community.relevanceai.com">
  Connect with other Relevance Builders, share your listings, and get support from the community
</Card>


# Pricing
Source: https://relevanceai.com/docs/get-started/pricing

Plan tiers, Actions, Vendor Credits, and feature comparison for Relevance AI

Start delegating work to agents today. Grow as your AI Workforce expands.

<CardGroup>
  <Card title="Free" icon="gift">
    **\$0** / month

    Explore the platform.
  </Card>

  <Card title="Pro" icon="rocket">
    From **\$19** / month

    For solo GTM operators.
  </Card>

  <Card title="Team" icon="users">
    From **\$234** / month

    For teams building at scale.
  </Card>

  <Card title="Enterprise" icon="building">
    **Custom**

    For org-wide AI Workforces.
  </Card>
</CardGroup>

## Plan details

<AccordionGroup>
  <Accordion title="Free — $0 / month">
    <span />

    Explore the platform — build agents, clone from the marketplace, and test them out.

    **Includes:**

    * 200 Actions / month
    * \$2 bonus Vendor Credits (one-time)
    * Unlimited Agents & Tools
    * 1 Workforce
    * 1 User & 1 Project
    * 30-day Task History
    * Marketplace Access
    * Community Forum
    * SOC 2 & GDPR Compliance

    [Sign up today →](https://app.relevanceai.com/auth)
  </Accordion>

  <Accordion title="Pro — $19 / month annual, $29 / month monthly">
    <span />

    For GTM operators and engineers looking to start building agents and delegating them work. Annual billing saves 33% over the monthly rate.

    **Includes everything in Free, plus:**

    * 30,000 Actions / year (or 2,500 / month)
    * \$240 Vendor Credits / year (or \$20 / month)
    * Unlimited Workforces
    * 2 Build Users
    * Schedule Tasks
    * Chat Mode
    * Smart Escalations (Email & Slack)
    * Activity Center
    * Premium App Triggers (WhatsApp, LinkedIn, Telegram)
    * Bring Your Own LLM

    [Try for free →](https://app.relevanceai.com/auth)
  </Accordion>

  <Accordion title="Team — $234 / month annual, $349 / month monthly">
    <span />

    For teams building agents to handle large workloads or for many users. Annual billing saves 33% over the monthly rate.

    **Includes everything in Pro, plus:**

    * 84,000 Actions / year (or 7,000 / month)
    * \$840 Vendor Credits / year (or \$70 / month)
    * Unused Credits Rollover
    * 5 Build Users · 45 End Users
    * 5 Shared Projects
    * Calling & Meeting Agents
    * A/B Testing
    * Analytics Dashboard
    * Priority Support

    [Try for free →](https://app.relevanceai.com/auth)
  </Accordion>

  <Accordion title="Enterprise — Custom">
    <span />

    For companies looking to decouple growth from headcount via an AI Workforce.

    **Includes everything in Team, plus:**

    * Custom Actions and Vendor Credits
    * Unlimited Users & Projects
    * Enterprise App Triggers (Salesforce, Snowflake, Zendesk)
    * Agent Evaluations
    * Work Hour Controls
    * Multi-Org Management
    * Enterprise Security (SSO, RBAC, audit logs)
    * Dedicated Account Manager
    * Custom Implementation
    * Priority Early Access

    [Contact sales →](https://relevanceai.com/book-a-demo)
  </Accordion>
</AccordionGroup>

## Complete feature comparison

### Platform

| Feature                                                                                  | Free             | Pro       | Team      | Enterprise |
| ---------------------------------------------------------------------------------------- | ---------------- | --------- | --------- | ---------- |
| **[Actions](/docs/admin/subscriptions/plans) / month**                                        | 200              | 2,500     | 7,000     | Custom     |
| **[Vendor Credits](/docs/admin/subscriptions/plans) / month**                                 | 1,000 (one-time) | 10,000    | 35,000    | Custom     |
| **[Build Users](/docs/admin/project-management/add-members)**                                 | 1                | 2         | 5         | Unlimited  |
| **[End Users](/docs/admin/project-management/add-members)**                                   | —                | —         | 45        | Unlimited  |
| **[Shared Projects](/docs/admin/project-management/switch-projects)**                         | 1                | 1         | 5         | Unlimited  |
| **[Workforces](/docs/get-started/core-concepts/workforces)**                                  | 1                | Unlimited | Unlimited | Unlimited  |
| **[Agents](/docs/get-started/core-concepts/agents)**                                          | Unlimited        | Unlimited | Unlimited | Unlimited  |
| **[Tools](/docs/get-started/core-concepts/tools)**                                            | Unlimited        | Unlimited | Unlimited | Unlimited  |
| **[Integrations](/docs/integrations/introduction)**                                           | Unlimited        | Unlimited | Unlimited | Unlimited  |
| **[Concurrent Agent Tasks](/docs/admin/system-limits)**                                       | Less             | Standard  | More      | Custom     |
| **[Knowledge / Memory](/docs/get-started/core-concepts/knowledge)**                           | Less             | Standard  | More      | Custom     |
| **[Task History](/docs/build/agents/build-your-agent/agent-settings/task-view)**              | 30d              | 90d       | 90d       | Custom     |
| **[Scheduling Tasks](/docs/build/agents/build-your-agent/agent-triggers/scheduled-triggers)** | ✗                | ✓         | ✓         | ✓          |
| **[Escalations](/docs/build/agents/build-your-agent/alerts)**                                 | ✗                | ✓         | ✓         | ✓          |

### Agent modes

| Feature                                                               | Free | Pro | Team | Enterprise |
| --------------------------------------------------------------------- | ---- | --- | ---- | ---------- |
| **[Chat](/docs/get-started/chat/introduction)**                            | ✓    | ✓   | ✓    | ✓          |
| **[Calling](/docs/build/agents/build-your-agent/agent-modes/phone-calls)** | ✗    | ✗   | ✓    | ✓          |
| **[Meeting](/docs/build/agents/use-cases/meeting-agents-overview)**        | ✗    | ✗   | ✓    | ✓          |

### App integrations

| Feature                                                                          | Free | Pro | Team | Enterprise |
| -------------------------------------------------------------------------------- | ---- | --- | ---- | ---------- |
| **[2,000+ Apps](/docs/integrations/introduction)**                                    | ✓    | ✓   | ✓    | ✓          |
| **[Custom Apps & API Integrations](/docs/get-started/core-concepts/api-integration)** | ✓    | ✓   | ✓    | ✓          |

### Triggers

| Feature                                                                                                                | Free | Pro | Team | Enterprise |
| ---------------------------------------------------------------------------------------------------------------------- | ---- | --- | ---- | ---------- |
| **[1,000+ App Triggers](/docs/build/agents/build-your-agent/agent-triggers/integrations)**                                  | ✓    | ✓   | ✓    | ✓          |
| **[Premium Triggers](/docs/build/agents/build-your-agent/agent-triggers/integrations)** (WhatsApp, LinkedIn, Telegram)      | ✗    | ✓   | ✓    | ✓          |
| **[Build your own Triggers](/docs/build/agents/build-your-agent/triggers)**                                                 | ✗    | ✓   | ✓    | ✓          |
| **[Enterprise Triggers](/docs/build/agents/build-your-agent/agent-triggers/integrations)** (Salesforce, Snowflake, Zendesk) | ✗    | ✗   | ✗    | ✓          |

### AI management

| Feature                                                       | Free | Pro | Team | Enterprise |
| ------------------------------------------------------------- | ---- | --- | ---- | ---------- |
| **Use any LLMs with credits**                                 | ✓    | ✓   | ✓    | ✓          |
| **Bring your own LLM**                                        | ✗    | ✓   | ✓    | ✓          |
| **A/B Testing**                                               | ✗    | ✓   | ✓    | ✓          |
| **[Analytics Dashboard](/docs/enterprise/analytics)**              | ✗    | ✗   | ✓    | ✓          |
| **[Agent Evaluations](/docs/build/agents/build-your-agent/evals)** | ✗    | ✗   | ✗    | ✓          |
| **Work Hour Controls**                                        | ✗    | ✗   | ✗    | ✓          |

### Security & compliance

| Feature                                                              | Free | Pro | Team | Enterprise |
| -------------------------------------------------------------------- | ---- | --- | ---- | ---------- |
| **[SOC 2 & GDPR Compliance](/docs/admin/security)**                       | ✓    | ✓   | ✓    | ✓          |
| **[SSO (SAML)](/docs/enterprise/sso-setup)**                              | ✗    | ✗   | ✗    | ✓          |
| **[RBAC](/docs/enterprise/rbac)**                                         | ✗    | ✗   | ✗    | ✓          |
| **[Audit Logs](/docs/enterprise/streaming-events)**                       | ✗    | ✗   | ✗    | ✓          |
| **[Multi-Org Management](/docs/enterprise/org-structure-best-practices)** | ✗    | ✗   | ✗    | ✓          |

### Support

| Feature                                                         | Free | Pro | Team | Enterprise |
| --------------------------------------------------------------- | ---- | --- | ---- | ---------- |
| **[Community Forum](https://community.relevanceai.com)**        | ✓    | ✓   | ✓    | ✓          |
| **[Marketplace Access](/docs/get-started/marketplace/introduction)** | ✓    | ✓   | ✓    | ✓          |
| **Priority Support**                                            | ✗    | ✗   | ✓    | ✓          |
| **Dedicated Account Manager**                                   | ✗    | ✗   | ✗    | ✓          |
| **Custom Implementation**                                       | ✗    | ✗   | ✗    | ✓          |
| **Priority Early Access**                                       | ✗    | ✗   | ✗    | ✓          |

## Add-ons and top-ups

Top up your account with extra Actions or Vendor Credits at any time on a paid plan — no plan change required. Actions must be bought in increments of 1,000; Vendor Credits in increments of 10,000.

<CardGroup>
  <Card title="Extra Actions" icon="bolt">
    **\$80** per 1,000 Actions

    More capacity for your AI Workforce to send emails, update CRMs, and run workflows.
  </Card>

  <Card title="Extra Vendor Credits" icon="coins">
    **\$20** per 10,000 Vendor Credits

    More AI model and tool usage, passed through at wholesale with no markup.
  </Card>
</CardGroup>

<Note>
  Top-ups are only available on paid plans — Free users need to upgrade to Pro or Team to purchase. See [Plans and credits](/docs/admin/subscriptions/plans) for the in-platform purchase flow.
</Note>

### What rolls over and what resets at renewal

* **Vendor Credits roll over indefinitely** while you're subscribed — both your plan's included Vendor Credits and any top-ups you purchase. You won't lose unused Vendor Credits as long as your subscription stays active.
* **Plan Actions reset at each renewal** to your plan's default amount.
* **Action top-ups roll over to the next billing cycle**, so purchased Actions carry forward even though base plan Actions reset.

<Note>
  Rollover behavior above applies to Free, Pro, and Team plans. Enterprise customers should confirm rollover terms with their Account Manager.
</Note>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What is an Action?">
    An Action is counted when an agent runs a tool — whether it's a simple task like sending one email or a complex workflow with many steps.
  </Accordion>

  <Accordion title="What is a Vendor Credit?">
    Vendor Credits cover LLM and tool usage costs at wholesale, with no markup. They roll over indefinitely while you're subscribed, and you can bring your own API keys on paid plans to bypass them entirely.
  </Accordion>

  <Accordion title="Can I buy more Actions or Vendor Credits without upgrading my plan?">
    Yes — on any paid plan, you can purchase top-ups at $80 USD per 1,000 Actions and $20 USD per 10,000 Vendor Credits. Actions must be bought in increments of 1,000 and Vendor Credits in increments of 10,000. To purchase, click the credit & action counter in the bottom left of your home screen, then **Buy credits**. Free users need to upgrade to Pro or Team to add top-ups.
  </Accordion>

  <Accordion title="What rolls over and what resets at renewal?">
    Vendor Credits — both your plan's included Vendor Credits and any top-ups you purchase — roll over indefinitely while you're subscribed. Plan Actions reset to your plan's default amount at each renewal, but any Action top-ups you purchase roll over to your next billing cycle. Rollover behavior above applies to Free, Pro, and Team plans; Enterprise customers should confirm with their Account Manager.
  </Accordion>

  <Accordion title="How do I upgrade to Pro or Team?">
    You can upgrade in-platform by clicking the **Manage Plan** / credits counter in the bottom left of your dashboard in Relevance AI while logged in.
  </Accordion>

  <Accordion title="Is my subscription applied at an organization level?">
    Yes, for Free, Pro, and Team plans. Subscriptions are applied at the Organization level, not to individual users or projects. Everyone in the Organization shares the plan's Actions, Vendor Credits, and feature access.

    Enterprise customers can request additional Organizations under their subscription. If you're on an active Enterprise plan and need another Organization, speak to your account team.
  </Accordion>

  <Accordion title="Can I transfer my subscription to another organization?">
    No. Subscriptions cannot be transferred from one Organization to another.
  </Accordion>

  <Accordion title="I've run out of credits on Free, how do I continue using Relevance AI?">
    Vendor Credits on the Free plan are a one-time allocation (1,000 credits at sign-up) and do not renew. Once you've used them, you'll need to upgrade to a paid plan (Pro or Team) to get more — Free users can't purchase top-ups.

    Actions on Free reset to 200 each month, so basic Action-only usage continues each cycle.
  </Accordion>

  <Accordion title="I'm interested in setting up an Enterprise plan for my business. Where can I learn more?">
    Learn more about our Enterprise plans with our team. [Book a call](https://relevanceai.com/book-a-demo) to discuss your requirements.
  </Accordion>

  <Accordion title="Do I need to enter a credit card to try Relevance AI?">
    No, Relevance AI's free plan doesn't require a credit card.
  </Accordion>

  <Accordion title="How does Relevance AI protect my privacy?">
    Your Project Information is confidential, even from Relevance AI. We are SOC 2 Type II compliant. [Read more about Relevance AI Security](/docs/admin/security).
  </Accordion>
</AccordionGroup>

## Next steps

<CardGroup>
  <Card title="Plans and credits" icon="credit-card" href="/admin/subscriptions/plans">
    Learn how Actions, Vendor Credits, top-ups, rollovers, and renewals work in the platform.
  </Card>

  <Card title="Book a demo" icon="calendar" href="https://relevanceai.com/book-a-demo">
    Talk to our team about Enterprise plans, custom Actions and Vendor Credits, or curated multi-agent systems like the BDR agent.
  </Card>
</CardGroup>


# Quick Start
Source: https://relevanceai.com/docs/get-started/quick-start-guide

Build and test your first AI agent in under 5 minutes.

This guide gets you from zero to a working agent as fast as possible. Pick one of three paths, follow the steps, and see your agent in action.

These steps assume you're logged in and on the home screen. If you don't have an account yet, [sign up for free](https://auth.relevanceai.com/signup/).

## Step 1: Create your first Agent

There are three ways to get started. Pick the one that fits you best.

<CardGroup>
  <Card title="Invent an Agent" icon="wand-magic-sparkles" href="#option-1-invent-an-agent">
    Describe what you want and let AI build it for you
  </Card>

  <Card title="Clone from Marketplace" icon="clone" href="#option-2-clone-from-marketplace">
    Grab a pre-built agent and make it yours
  </Card>

  <Card title="Build from scratch" icon="hammer" href="#option-3-build-from-scratch">
    Full control over every detail
  </Card>
</CardGroup>

### Option 1: Invent an Agent

The fastest way to get started. Describe what you need and we'll generate a working agent for you.

<img alt="Describe what you want and Inventor builds your Agent for you." />

1. Click **Agents** in the left sidebar
2. Click **+ Create Agent**
3. Select **Invent**
4. Describe what you want your agent to do — be specific about the task, the output you expect, and any integrations it should use. For example: *"Create an agent that researches a company before a sales call. It should use web search to find recent news, funding, and key contacts, then output a one-page brief with company overview, decision makers, and talking points."*
5. Review the generated agent — Inventor will set up the prompt and suggest tools
6. Connect any integrations your tools need (e.g. Gmail, HubSpot, LinkedIn) in the Tools section of the Agent builder, or [add integrations](/docs/integrations/add-integrations) from the sidebar

<Tip>
  Best when you know what you want and can describe it clearly. Great for getting a concept off the ground quickly — Inventor handles the setup, you make the tweaks.
</Tip>

### Option 2: Clone from Marketplace

Start with an agent that's already been built and tested by domain experts.

1. Go to [Marketplace](https://marketplace.relevanceai.com/) or click **Marketplace** in the left sidebar
2. Browse or search for an agent that matches your use case
3. Click on the agent to view its details
4. Click **Clone** to add it to your workspace
5. A setup popup will walk you through connecting any integrations the agent needs — follow the prompts to link your accounts
6. Review the prompt, tools, and settings — adjust anything you need

<Tip>
  Great when you have a common use case — like lead qualification, meeting prep, or customer support — and want a proven starting point. Clone it, tweak it, ship it.
</Tip>

### Option 3: Build from scratch

Full control from the ground up. Best if you have a specific workflow in mind.

1. Click **Agents** in the left sidebar
2. Click **+ Create Agent**
3. Select **Start from scratch**
4. Write a [prompt](/docs/build/agents/build-your-agent/prompt) — tell the agent who it is, what it does, and how it should behave
5. Add [tools](/docs/build/agents/build-your-agent/tools) — give it the actions it needs (search the web, send emails, update your CRM, etc.)
6. Optionally add [knowledge](/docs/build/knowledge/create-knowledge) — upload docs or connect data sources for context

For a full breakdown of building with Relevance AI, see [Build with Relevance AI](/docs/build/introduction).

## Step 2: Test your Agent

<img alt="The Run tab — give your Agent a task and watch it work through it step by step." />

Head to the **Run** tab to give your agent a task. This is a real run — you can see exactly what tools it calls, what output it produces, and how it handles the task end-to-end.

<Tip>
  You can also use your agent in [Relevance Chat](/docs/get-started/chat/introduction) — a conversational interface where you and your team can @ mention any agent.
</Tip>

## Step 3: Refine

If the output isn't quite right, iterate:

* **Tighten the prompt** — be more specific about what you want, how it should be formatted, and what tone to use
* **Add or adjust tools** — make sure the agent has access to the right actions for the job
* **Add knowledge** — upload docs, connect data sources, or add context so the agent has what it needs

Test again after each change. A few rounds of this and you'll have a solid agent.

## Step 4: Share and deploy

Once you're happy with the output, put your agent to work:

* [Share your agent](/docs/build/agents/share-your-agent) with your team so others can use it
* Use [Relevance Chat](/docs/get-started/chat/introduction) to give your team instant access to your agents
* Set up [triggers](/docs/build/agents/build-your-agent/triggers) to run it automatically — on a schedule, from a webhook, or when something happens in your CRM
* Use [bulk schedule](/docs/build/agents/give-your-agent-tasks/bulk-schedule) to run it across a list of inputs
* Add [alerts](/docs/build/agents/build-your-agent/alerts) so you get notified when something needs attention, and the agent knows when to loop in a human

## Beyond building

Now that you've got an agent up and running, there's more to explore across the platform.

### Use your Agents in Chat

<img alt="Relevance Chat — @ mention any Agent and interact with it directly." />

[Relevance Chat](/docs/get-started/chat/introduction) is a conversational interface where you and your team interact with your agents directly. Type a message, @ mention any agent in your project, and it will handle the task using the tools and knowledge you've configured. You can switch between LLMs, combine multiple agents in a single conversation, save prompts your team uses often, and @ mention project files to insert their paths directly into your message. Chat also comes with built-in agents for things like building websites, generating images, deep research, and creating presentations.

Visit [chat.relevanceai.com](https://chat.relevanceai.com) to get started — it works on desktop and mobile.

### Browse the Marketplace

The [Marketplace](/docs/get-started/marketplace/introduction) is a curated library of pre-built agents and tools created by Relevance AI and the community. Instead of building from scratch, you can clone an agent that's already been designed and tested for common use cases — lead qualification, customer support, meeting prep, and more. Once cloned, you have full control to customize the agent's prompt, tools, and settings to fit your needs.

### Build from your AI coding environment

Prefer to work from your terminal? Connect your AI coding environment — Claude Code, Cursor, or any MCP-compatible client — to Relevance AI over MCP. Describe what you want in natural language and your AI assistant builds and edits your Agents, Tools, and Workforces directly. See [MCP & Plugins](/docs/get-started/core-concepts/mcp-plugins) to get started.

***

Ready to go deeper? Head to [Build with Relevance AI](/docs/build/introduction) for a full breakdown of how to configure agents, tools, knowledge, and multi-agent workforces.


# How to reach out to our support team
Source: https://relevanceai.com/docs/get-started/support

How to contact Relevance AI support for help and questions

If you need help or have questions, our support team is here to assist you! There are three ways to get support, either from our team or from fellow customers.

## Relevance AI Community

The [Relevance AI Community](https://community.relevanceai.com) is a platform to connect with fellow Relevance AI customers, Partners, AI experts and Rellies (Relevance AI team members).

If you're looking for support from your fellow users, including advice on how to build using our platform, we'd recommend reaching out to the Community and creating a question [here](https://community.relevanceai.com/c/questions/).

<Warning>
  Please note that this space is primarily monitored by our Community members, and is considered a peer to peer Community space. If your question in the Community has been left unanswered, you are welcome to reach out to our support team for help using the methods below.
</Warning>

## In-app chat ticket support

The fastest way to get in touch is through our in-app chat support:

1. Log in to [Relevance AI](https://app.relevanceai.com/)
2. Click the **Ask for Help** button, found in the bottom left corner of Relevance AI in the sidebar (or use the keyboard shortcut CMD + K)
3. Ask the AI Agent a question here, or click 'Create a support ticket' to start chatting with our support team
4. Then, click on **Start new chat** to start chatting with our support team

<div>
  <iframe title="Invite a user to a project in Relevance AI" />
</div>

## Email support

If you prefer, you can also email us at [**support@relevanceai.com**](mailto:support@relevanceai.com).

## Phone support

We only offer dedicated staff calls for **Enterprise customers**.

If you're on a **Free to Team plan**, we offer calls with **Harley**, an AI support agent who can walk through your concerns with you.

### How to request a call from Harley:

1. [Raise a support ticket](/docs/get-started/support#in-app-chat-ticket-support) first, including a description of your concern
2. Request a call in your ticket, providing:
   * Your phone number
   * A suitable time within our business hours (9am - 5pm Sydney time (AEST / AEDT) on weekdays)

<Warning>**Beta Service Notice:** Harley is currently in beta and may experience bugs or errors. This is an early-stage service, and we welcome constructive feedback to help us improve the experience.</Warning>

***

## Enterprise support

We offer priority support to Enterprise customers, including a dedicated Slack channel for support. We also offer custom implementation (agent building services and consultation) to our Enterprise customers, as well as a dedicated Account Manager and priority early access to new features.

You can compare our plans, including all the value Enterprise has to offer your organization, on our [pricing page](/docs/get-started/pricing).

If you'd like to discuss an upgrade to Enterprise, please [book a demo](https://relevanceai.com/book-a-demo). If you don't hear back from our sales team, please reach out to support via one of the methods above.

***

## Service Level Agreements (SLAs)

We offer the following first response times (FRT) SLAs, based on your subscription:

| Subscription   | SLA             |
| -------------- | --------------- |
| **Enterprise** | 1 business day  |
| **Team**       | 2 business days |
| **Pro**        | 3 business days |

While **Free** customers can raise tickets at this time, we do not offer SLAs for free customers.

<Info>Our team currently only offer support to Free, Pro and Team customers between 9am - 5pm Sydney time (AEST / AEDT) on weekdays at this time. We do not offer weekend support. Our team is actively growing, and we will expand our support team globally in the future.</Info>

***

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Do you use AI Agents in your support process?">
    Yes! We dogfood our own AI Agent platform across all teams, including support. We're AI-first and use AI Agents throughout our support process for answering customer questions, researching our platform, and drafting documentation.

    The level of support you receive depends on your subscription:

    * **Free tier:** AI Agent support and Community access
    * **Pro and Team tiers:** A mixture of human and AI support
    * **Enterprise tier:** Dedicated support including account managers, with AI assistance

    This approach allows us to provide faster, more consistent support while ensuring human experts are available for complex issues.
  </Accordion>

  <Accordion title="What are your communication guidelines?">
    We're committed to providing excellent support in a respectful and professional environment. We expect professional and respectful communication across all channels, including support tickets, email, and Community forums.

    Aggressive, antagonistic, or abusive language—including swearing—may result in our inability to provide support. Our team members deserve a safe and respectful work environment.

    All interactions are governed by our [Terms & Conditions](https://relevanceai.com/terms-and-conditions). We welcome constructive feedback about our products and services, and we encourage you to share it professionally.
  </Accordion>

  <Accordion title="I was on the Business plan — what support and SLAs are available to me?">
    The Business plan has been sunset, and all Business customers have automatically been moved to the Team plan. If you'd like to retain your dedicated Slack channel, please contact our sales team to discuss an upgrade to Enterprise.
  </Accordion>

  <Accordion title="I emailed a team member / AI Agent at Relevance AI for support and didn't get an answer - how do I get help?">
    Our team members and AI Agents aren't able to help answer questions. Please reach out using one of the methods above and we'll be able to help you with your question or issue!
  </Accordion>

  <Accordion title="What if I experience an outage outside of our supported hours?">
    Our engineering team is on-call 24/7, and we will respond to any outages that occur outside of our supported hours. You may not receive a response outside of our supported hours, but our engineering team will come online to address any alerts / outages and swiftly resolve them.
  </Accordion>

  <Accordion title="I raised a bug report in product -- why haven't I heard back?">
    Bug reports are sent to our engineering team for investigation, however, we don't respond to bug reports. If you need immediate help or want to follow up on a bug report, please reach out to support via one of the methods above.
  </Accordion>
</AccordionGroup>


# Troubleshooting agents not working
Source: https://relevanceai.com/docs/get-started/troubleshooting/troubleshooting-agents-not-working

Troubleshoot agents that are failing or not working properly

When your agents aren't working as expected, there are several common causes and solutions. This guide will help you identify and resolve the most frequent issues that prevent agents from functioning properly.

## Common causes of agent failures

### 1. Insufficient credits or actions

If your agent is failing, the most common cause is running out of credits or actions.

#### Understanding credits and actions

Relevance AI uses two types of consumption:

* **Actions**: Each time a tool runs, it counts as one action — whether it's a simple task like sending one email or running a complex workflow with many steps
* **Vendor Credits**: The cost of running the AI model (LLM costs) and the tools you use

#### How to check your usage

Learn how to monitor your credit and action usage at both the [organization level](/docs/admin/subscriptions/plans#monitoring-actions-and-vendor-credits-usage) and [individual agent level](/docs/admin/subscriptions/plans#at-an-individual-agent-level).

#### Solutions for insufficient credits/actions

* **Upgrade your plan**: Consider upgrading to a higher plan with more credits/actions
* **Purchase additional credits**: If you have a paid plan, you can purchase extra credits to use before your next renewal
* **Bring your own API keys**: Use your own API keys to bypass Vendor Credits entirely (available on paid plans only)

### 2. Tool failures

When agents fail, it's often because one of their tools is not working properly. Here's how to troubleshoot tool issues:

<Tip>
  Use the **Has active tool failures** filter in the Task list to quickly identify which Tasks had Tool failures — including cases where the Agent continued running despite a Tool error. See [Filtering Tasks](/docs/build/agents/give-your-agent-tasks/task-overview#filtering-tasks) for details.
</Tip>

#### Step 1: Test the tool independently

1. Go to the [Tools page](https://app.relevanceai.com/tools)
2. Find the tool your agent is using
3. Click on the tool and go to the "Use" tab
4. Run the tool with test inputs to see if it works on its own

If the tool fails independently, the issue is with the tool itself, not the agent. To fix tool issues:

1. Go to the "Build" tab of the tool
2. Run each tool step individually by clicking the play icon next to each step
3. Identify which specific step is failing
4. Check the step configuration and fix any issues:
   * Verify API keys are correct
   * Check input formats and data types
   * Review step settings and parameters
   * Remove or reconfigure problematic steps

#### Step 2: Check agent-to-tool communication

If the tool works independently but fails when used by the agent:

1. **Verify input data types**: Ensure the agent is sending the tool the correct data types (string, array, number, etc.)
2. **Check input format**: Make sure the agent is providing inputs in the expected format
3. **Review tool input descriptions**: Ensure each tool input has a clear description explaining what the agent should provide
4. **Review tool configuration**: Go to your agent's tools section and check:
   * Input configuration mode (Let agent decide, Set manually, or Tool output)
   * Whether the agent has the right context to use the tool
   * If approval settings are preventing tool execution

#### Common tool input issues

**Data type mismatches:**

* Tool expects a string but receives an array
* Tool expects a number but receives text
* Tool expects JSON but receives plain text

**Format issues:**

* Missing required fields
* Incorrect field names
* Wrong data structure

**Solution**: Review your agent's instructions and tool configuration to ensure the agent understands what data to send to each tool.

### 3. Agent configuration issues

#### Check agent settings

1. **Review agent prompt**: Ensure your agent has clear, specific instructions about when and how to use tools
2. **Verify tool approval settings**: Check if tools are set to "Auto Run", "Approval Required", or "Let Agent Decide"
3. **Check escalation settings**: Review retry settings and error handling behavior

#### Common configuration problems

* **Vague prompt**: Agent doesn't understand when to use tools
* **Wrong approval mode**: Tools require approval but agent doesn't ask
* **Missing context**: Agent lacks information needed to use tools effectively
* **Conflicting settings**: Multiple tools with overlapping purposes

### 4. Integration and API issues

#### Check integrations

1. Go to [Integrations & API Keys](https://app.relevanceai.com/integrations) in Relevance AI
2. Verify all required integrations are connected
3. Check if API keys are valid and have proper permissions
4. Test integration connections

#### Common integration problems

* **Expired API keys**: Update or refresh your API keys
* **Insufficient permissions**: Ensure API keys have the required scopes
* **Rate limiting**: Check if you've hit API rate limits
* **Service outages**: Verify the external service is operational

## Agent not working as expected

If your agent is running but not producing the results you want, this is often a prompt engineering issue.

### Understanding agent behavior

Agents are designed to make their own decisions and aren't end-to-end workflows. They use reasoning to determine the best approach to complete tasks, which means they may not always follow the exact path you expect.

### Improving your agent prompt

To get better results from your agent:

1. **Be as clear and specific as possible** in your agent prompt
2. **Provide detailed instructions** about what you want the agent to do
3. **Include examples** of good responses or behaviors
4. **Specify the format** you want outputs in
5. **Set clear boundaries** about what the agent should and shouldn't do

### Key prompt engineering principles

* **Be explicit**: Don't assume the agent will understand implicit requirements
* **Use clear language**: Avoid ambiguous terms and provide specific criteria
* **Provide context**: Give the agent relevant background information
* **Set expectations**: Clearly define what success looks like
* **Iterate and test**: Refine your prompt based on the agent's performance

<Info>
  For a deeper understanding of when to use AI agents vs workflows, read our co-founder's [comprehensive guide on LinkedIn](https://www.linkedin.com/feed/update/urn:li:activity:7336149777472024593/).
</Info>

### Model performance issues

If your agent isn't performing well, consider upgrading to a more capable language model.

#### When to upgrade your model

* **Complex reasoning tasks**: Advanced models handle multi-step reasoning better
* **Tool usage**: Some models are better at understanding when and how to use tools
* **Large context**: If you need to process large amounts of information
* **Specialized tasks**: Some models excel at specific types of work

#### Available models and their strengths

**OpenAI models:**

* Advanced conversational abilities and creative writing
* Broad general knowledge and versatility
* Best for: Versatile agents, customer support, brainstorming
* Learn more: [OpenAI LLM models](/docs/integrations/llm-integrations/openai)

**Google Gemini models:**

* Strong coding ability and complex task handling
* Excellent at processing multiple file types (PDF, images, audio, video)
* Best for: Software development agents, complex task execution
* Learn more: [Google's Gemini LLM models](/docs/integrations/llm-integrations/gemini)

**Anthropic Claude models:**

* Focused on safe, reliable, and ethical AI responses
* Excellent at reasoning and thoughtful tasks
* Best for: Detailed explanations, structured outputs, sensitive industries
* Learn more: [Anthropic LLM models](/docs/integrations/llm-integrations/anthropic)

<Warning>
  More advanced models are more expensive but often provide significantly better results. Consider your use case and budget when choosing a model.
</Warning>

<Warning>
  If you're experiencing configuration issues where your agent isn't working as expected, our support team has limited ability to provide guidance to customers on Team or below as this is considered implementation support, which we only offer to Enterprise customers.

  If you're interested in an Enterprise subscription with dedicated implementation support to build agents for your use cases, you can [book a demo](https://relevanceai.com/book-a-demo). You can also reach out to our [Partners](https://partners.relevanceai.com) for implementation support.
</Warning>

## Still need help?

If you've tried all the troubleshooting steps above and your agent is still not working, please [contact our support team](/docs/get-started/support) and include:

* Your agent configuration details
* Error messages you're seeing
* Steps you've already tried
* A description of the specific issue you're experiencing
* **Screen recording**: Attach a [Loom](https://loom.com) or [jam.dev](https://jam.dev) recording of the issue to help us understand the problem better


# Troubleshooting browser issues
Source: https://relevanceai.com/docs/get-started/troubleshooting/troubleshooting-browser-issues

Troubleshoot browser-related issues with Relevance AI

The browser is what you use to access Relevance AI on your desktop computer. This includes Chrome, Firefox, Safari, Edge, and other modern browsers. Because all browsers are different, sometimes you might experience issues with Relevance AI because of the browser you are using. These issues can range from login problems (such as blank pages or login loops) to unexpected behavior in Relevance AI once you log in, including slow loading times, broken features, or integration failures. This article will help you troubleshoot those issues and get back to using Relevance AI smoothly.

## Supported browsers and platforms

Relevance AI supports the following modern browsers on desktop:

* **Chrome**
* **Firefox**
* **Safari**
* **Edge**
* **Arc**
* **Dia**
* **Comet**

<Warning>
  **Do not use [app.relevanceai.com](https://app.relevanceai.com) on mobile devices.** The builder platform (where you create and configure Agents, Tools, and Workforces) is designed for desktop use only and will not function properly on mobile browsers. Please use a desktop computer to access [app.relevanceai.com](https://app.relevanceai.com).
</Warning>

<Info>
  While the builder platform is not available on mobile, Relevance Chat is available as a native app for [iOS](https://apps.apple.com/au/app/relevance/id6757512568) and [Android](https://play.google.com/store/apps/details?id=com.relevanceai.android), or in a mobile browser at [chat.relevanceai.com](https://chat.relevanceai.com).
</Info>

## Basic troubleshooting steps

If you're experiencing issues with Relevance AI, try the following steps in order:

### 1. Update your browser

Make sure you're using the latest version of your browser. Outdated browsers may not support all the features required by Relevance AI.

### 2. Disable browser extensions

Browser extensions can sometimes interfere with Relevance AI's functionality. Try disabling all extensions temporarily:

1. Open your browser in incognito/private mode
2. Navigate to [app.relevanceai.com](https://app.relevanceai.com)
3. Test if the issue persists

If the issue is resolved in incognito mode, one of your extensions is likely causing the problem.

### 3. Clear browser cache

Clearing your browser's cache can resolve many common issues:

* **Chrome**: Press `Ctrl+Shift+Delete` (Windows) or `Cmd+Shift+Delete` (Mac)
* **Firefox**: Press `Ctrl+Shift+Delete` (Windows) or `Cmd+Shift+Delete` (Mac)
* **Safari**: Go to Safari menu > Clear History and Website Data
* **Edge**: Press `Ctrl+Shift+Delete` (Windows) or `Cmd+Shift+Delete` (Mac)

### 4. Try a different browser

Test Relevance AI in a different browser to see if the issue is browser-specific.

### 5. Check your network connection

Try connecting to a different network to rule out network-related issues.

### 6. Check antivirus software

Make sure your antivirus software isn't blocking the following Relevance AI domains:

* `app.relevanceai.com`
* `chat.relevanceai.com`
* `marketplace.relevanceai.com`
* `api-REGIONID.stack.tryrelevanceai.com` (replace REGIONID with your [region ID](/docs/admin/project-management/ids))

## Troubleshooting specific issues

### Login problems

If you're having trouble logging in:

1. **Blank login page**: Clear your browser cache and try incognito mode
2. **Login loop**: Disable browser extensions and try incognito mode
3. **Invalid credentials**: Double-check your email and password, or [reset your password](/docs/admin/account-management/change-password) if needed
4. **SSO login issues**: If you're using Single Sign-On (SSO), contact your Enterprise admin for assistance with authentication problems

### Performance issues

If Relevance AI is running slowly:

1. **Check your internet connection**: Ensure you have a stable, fast connection
2. **Close unnecessary tabs**: Free up browser memory
3. **Disable resource-heavy extensions**: Ad blockers and privacy tools can slow down the platform
4. **Clear browser cache**: Accumulated cache can impact performance

### Integration connection problems

If integrations aren't working:

1. **Check pop-up blockers**: Some integrations require pop-ups to be allowed
2. **Verify domain permissions**: Ensure the integration domains are allowed
3. **Check antivirus settings**: Make sure your antivirus isn't blocking integration domains

## Browser console errors

If you're still experiencing issues, check your browser's console for error messages:

### Opening the console

* **Chrome**: Press `Ctrl+Shift+J` (Windows) or `Cmd+Opt+J` (Mac)
* **Firefox**: Press `Ctrl+Shift+K` (Windows) or `Cmd+Opt+K` (Mac)
* **Safari**: Enable Developer menu in Preferences > Advanced, then press `Cmd+Opt+C`
* **Edge**: Press `F12` and go to the 'Console' tab

### Common error messages

Look for these common error messages in the console:

* **CORS errors**: Usually related to domain restrictions
* **Network errors**: Check your internet connection
* **JavaScript errors**: May indicate browser compatibility issues
* **Authentication errors**: Check your login credentials

<Info>
  If you find error messages in the console, take a screenshot or copy the error text and include it when contacting support.
</Info>

## Still need help?

If you've tried all the troubleshooting steps above and are still experiencing issues, please [contact our support team](/docs/get-started/support) and include:

* Your browser type and version
* Any error messages from the console
* Steps you've already tried
* A description of the specific issue you're experiencing
* **Screen recording**: Attach a [Loom](https://loom.com) or [jam.dev](https://jam.dev) recording of the issue to help us understand the problem better


# Guides
Source: https://relevanceai.com/docs/guides

Practical, opinionated walkthroughs for putting Relevance AI to work in the function your team actually runs.

Each Guide takes a real go-to-market or operations function and shows you the shortest path from zero to a working AI implementation. They fork into three persona paths so the walkthrough matches how hands-on your team wants to be — no build, build visually, or build with AI.

## Pick a function

<CardGroup>
  <Card title="Sales" icon="chart-line" href="/guides/sales/getting-started">
    Research, outreach, lead scoring, CRM enrichment, meeting prep.
  </Card>

  <Card title="Marketing" icon="megaphone" href="/guides/marketing/getting-started">
    Lifecycle campaigns, content repurposing, campaign analytics.
  </Card>

  <Card title="Customer Success" icon="hand-holding-heart" href="/guides/customer-success/getting-started">
    Account health, QBR prep, renewal and expansion.
  </Card>

  <Card title="Support" icon="headset" href="/guides/customer-support/getting-started">
    Ticket triage, response drafting, knowledge base generation.
  </Card>

  <Card title="Solution Engineering" icon="screwdriver-wrench" href="/guides/solution-engineering/getting-started">
    RFP responses, demo prep, technical discovery summaries.
  </Card>

  <Card title="RevOps" icon="database" href="/guides/revops/getting-started">
    Pipeline hygiene, lead routing, data dedup.
  </Card>
</CardGroup>

## The four levels of AI autonomy

Every Guide climbs the same ladder — from human-led work to self-driving systems.

<CardGroup>
  <Card title="L1 — Assisted" icon="hand">
    Human asks, Agent acts, one task at a time. A sharper version of search — single request, single response.
  </Card>

  <Card title="L2 — Copilot" icon="map">
    Human kicks off a job, the Agent runs many actions, human reviews output before it ships. The Agent works ahead of you; you stay the gate.
  </Card>

  <Card title="L3 — Autopilot" icon="bolt">
    Events and signals trigger a workforce of Agents that collaborate end-to-end. Humans manage the fleet, not each task.
  </Card>

  <Card title="L4 — Self-driving" icon="infinity">
    Humans set business goals, Agents experiment and optimize autonomously. The system shapes itself against outcomes.
  </Card>
</CardGroup>

Pick a Guide above to see what each level looks like for your team.


# Account health and churn signals
Source: https://relevanceai.com/docs/guides/customer-success/account-health

Score every account on health with reasoning, not just a number. Surface risk before renewal so CSMs can act in time.

Health scores that are just numbers don't get used. A health Agent rates accounts across the signals that actually predict churn — usage drops, support spikes, exec turnover, deal-stage stalls — and surfaces the *why* so CSMs can do something about it.

## When this pays off

<CardGroup>
  <Card title="CSMs are reactive" icon="rotate-right">
    Churn risks are caught at renewal, not 90 days out when there's still time to fix them.
  </Card>

  <Card title="Health score is ignored" icon="eye-slash">
    The existing red/yellow/green field doesn't influence CSM action — nobody trusts the rollup.
  </Card>

  <Card title="Signals live in silos" icon="puzzle-piece">
    Usage data, support tickets, deal notes, exec changes — all in different tools, nobody synthesizes them.
  </Card>

  <Card title="Book of business is too big" icon="building-columns">
    Each CSM has 80+ accounts and there's no way to triage which ones need attention this week.
  </Card>
</CardGroup>

## The shape of this use case

A health Agent takes an account and returns a score, the reasoning behind it, and a recommended next action.

<CardGroup>
  <Card title="Inputs" icon="arrow-right-to-bracket">
    Account record, time window, segmentation cut.
  </Card>

  <Card title="Sources" icon="globe">
    CRM, support tickets, product usage / analytics, deal history, exec change tracking, your CS playbook.
  </Card>

  <Card title="Output" icon="file-lines">
    A health score with cited signals, a short summary of the risk picture, and a recommended outreach action.
  </Card>

  <Card title="Delivery" icon="paper-plane">
    Written back to the CRM as a field plus a note, posted in [Slack](/docs/integrations/popular-integrations/slack) for at-risk accounts, surfaced in a CSM digest.
  </Card>
</CardGroup>

## Where to start

Two ways in, depending on whether you want something running today or built to your exact spec.

<CardGroup>
  <Card title="Clone a pre-built Agent" icon="copy">
    Open the **[Customer Success Manager](https://marketplace.relevanceai.com/listing/b9b5e05c-4512-4749-8f66-4623ab895e21)**. More in the [Marketplace](/docs/get-started/marketplace/introduction).
  </Card>

  <Card title="Build your own" icon="hammer">
    Start from scratch in the [builder](/docs/build/introduction), or by describing it in Claude Code or Cursor with [MCP & Plugins](/docs/get-started/core-concepts/mcp-plugins).
  </Card>
</CardGroup>

Either way, these are prompts your CSMs can use on day one:

* *"How is Acme Corp doing right now? Usage, tickets, last contact, anything I should worry about?"*
* *"Which 5 accounts in my book look highest-risk for Q1 renewal?"*
* *"Compare Globex's engagement this quarter vs. last — what changed?"*

## Where to take it

Once it's running, deepen it in three moves:

<CardGroup>
  <Card title="Give it a playbook" icon="book">
    Shape it with a [prompt](/docs/build/agents/build-your-agent/prompt), your CS playbook in [Knowledge](/docs/build/knowledge/create-knowledge), and [Bulk Schedule](/docs/build/agents/give-your-agent-tasks/bulk-schedule).
  </Card>

  <Card title="Automate it on signals" icon="bolt">
    Wrap it in a [workflow](/docs/build/workforces/create-a-workforce) that fires on a [trigger](/docs/build/agents/build-your-agent/triggers).
  </Card>

  <Card title="Let it improve" icon="arrows-rotate">
    Pipe churn outcomes back into the Agent's [evals](/docs/build/agents/build-your-agent/evals) so scoring tracks what predicted cancellation.
  </Card>
</CardGroup>

## Common pitfalls

<AccordionGroup>
  <Accordion title="Score without reasoning" icon="magnifying-glass">
    A number alone is unactionable. CSMs need to see *which* signals moved and *why* the Agent flagged the account — otherwise they ignore it.
  </Accordion>

  <Accordion title="Stale signal definitions" icon="clock-rotate-left">
    Your churn drivers shift over time. Review the prompt's scoring criteria each quarter against actual churned accounts.
  </Accordion>

  <Accordion title="Overweighting product usage" icon="scale-unbalanced">
    Some segments use the product weekly and never churn; others log in monthly and stay loyal. Weight signals by segment, not globally.
  </Accordion>

  <Accordion title="No feedback loop from churned accounts" icon="arrows-rotate">
    Without piping churn outcomes back to the Agent's evals, the scoring can't learn. Connect closed-lost / churned status to evaluation runs.
  </Accordion>
</AccordionGroup>


# Customer Success
Source: https://relevanceai.com/docs/guides/customer-success/getting-started

Put AI Agents to work across the post-sale journey — health, renewals, expansion — so CSMs spend time on the accounts that need them.

CSMs are stretched across too many accounts. AI Agents take the watchful work — tracking signals, prepping QBRs, surfacing renewal risks — so CSMs focus on the conversations that actually move retention and expansion.

## Start with a Marketplace template

The Marketplace ships pre-built Agents you can clone in one click. Adapt them to your segments, your renewal cadence, your playbook. Or [build a custom Agent from scratch](/docs/build/introduction).

<CardGroup>
  <Card title="Atlas, the Account Analyst" icon="magnifying-glass-chart" href="https://marketplace.relevanceai.com/listing/678c9cf9-9b64-467a-8c51-e2d3db409fb1">
    In-depth research on an account, producing a structured report — useful before QBRs and renewal conversations.
  </Card>

  <Card title="Apla, the Account Planning Agent" icon="clipboard-list" href="https://marketplace.relevanceai.com/listing/fd712abc-66e4-4507-8f16-c4404cf5f5e5">
    Researches an account and writes the plan back to your CRM — handy for expansion across the book.
  </Card>

  <Card title="Browse the Marketplace" icon="store" href="/get-started/marketplace/introduction">
    The full catalog of Agents — health scoring, renewals, QBR prep, customer comms.
  </Card>
</CardGroup>

Click **Clone Agent** on any listing, connect the systems it needs (CRM, support, product analytics), and run it on a real account.

## Pick your path

How hands-on do you want to be? Pick a tab — the rest of the page is written for you.

<div>
  <Tabs>
    <Tab title="No build">
      You don't need to touch the builder. Open the cloned Agent above, ask in plain English, and let it work across the systems it's connected to.

      ## Sharpen the answers with knowledge

      Upload your CS playbooks, segmentation rules, renewal use case docs, and product positioning. Every conversation references them — sharper output, no new setup.

      <Card title="Add knowledge" icon="book" href="/build/knowledge/create-knowledge">
        Upload context once. Every conversation references it.
      </Card>

      ## Make it run on its own

      When you find yourself summarizing the same account state every Monday, set a trigger — every CSM gets a weekly book-of-business digest, every at-risk account auto-flags. One toggle, no code, no builder.

      <Card title="Triggers" icon="bolt" href="/build/agents/build-your-agent/triggers">
        Schedule runs, fire on CRM events, listen to webhooks.
      </Card>

      ## Ready to go further?

      When prompting and templates stop being enough — when you want the Agent to follow your team's playbook word for word and orchestrate cross-system signal monitoring — pick the **Build visually** tab above.
    </Tab>

    <Tab title="Build visually">
      You're comfortable in product UIs. You don't want to wire up an Agent from scratch, but you do want it to follow your CS playbook account by account. Open the cloned Agent in the builder and shape every piece.

      ## Connect the tools it needs

      CS Agents work best when they can read from and write to the systems your team already uses.

      * **CRM** — [HubSpot](/docs/integrations/popular-integrations/hubspot) or [Salesforce](/docs/integrations/popular-integrations/salesforce) for account data, deal stage, and renewal timing
      * **Support** — [Zendesk](/docs/integrations/popular-integrations/zendesk) or Intercom for ticket activity and CSAT
      * **Product analytics** — your usage data source for engagement signals
      * **Comms** — [Gmail](/docs/integrations/popular-integrations/gmail), [Slack](/docs/integrations/popular-integrations/slack) for outreach and internal alerts

      See more about [Integrations](/docs/integrations/introduction).

      ## Shape every surface

      Each piece below changes how the Agent behaves. Run it on a real account between every change — small iterations beat big rewrites.

      <CardGroup>
        <Card title="Prompt" icon="message" href="/build/agents/build-your-agent/prompt">
          Edit the system prompt to match your segmentation, escalation rules, and renewal use case.
        </Card>

        <Card title="Tools" icon="wrench" href="/build/agents/build-your-agent/tools">
          Add or remove the actions the Agent can take — pull usage data, query support tickets, draft renewal emails.
        </Card>

        <Card title="Knowledge" icon="book" href="/build/knowledge/create-knowledge">
          Upload your CS playbook, customer journey docs, segmentation rules. The Agent draws on these whenever it works.
        </Card>

        <Card title="Triggers" icon="bolt" href="/build/agents/build-your-agent/triggers">
          Run the Agent automatically — on a schedule, when usage drops, when a renewal date approaches.
        </Card>

        <Card title="Alerts" icon="bell" href="/build/agents/build-your-agent/alerts">
          Get notified when an account crosses a health threshold or when a renewal slips.
        </Card>

        <Card title="Memory" icon="brain" href="/build/agents/build-your-agent/memory">
          Persistent context per account — last interactions, prior escalations, what worked at renewal time.
        </Card>
      </CardGroup>

      ## Chain Agents into a workforce

      When a single Agent isn't enough — score health, then draft outreach, then schedule, then log to the CRM — wire them together on the Workforce canvas.

      <Card title="Build a workforce" icon="diagram-project" href="/build/workforces/create-a-workforce">
        Drag, connect, and run multi-Agent pipelines without code.
      </Card>

      ## Test before you trust

      Set up evals so health scoring or QBR-prep regressions get caught before they reach the CSM.

      <Card title="Evals" icon="vial" href="/build/agents/build-your-agent/evals">
        Define test cases in the UI. Run them on every change.
      </Card>

      ## Ready to go further?

      When clicking through the builder starts to feel slow — when you'd rather have your AI coding assistant build and edit Agents for you, drop into custom Python or JS code blocks, or call Agents from your own app via the SDK — pick the **Build with AI** tab above.
    </Tab>

    <Tab title="Build with AI">
      You already work in an AI coding environment. Clicking through a UI to edit health-scoring rules feels slow. Have your AI assistant build, shape, and orchestrate Agents directly from your editor.

      ## MCP & Plugins

      Connect your AI coding environment to Relevance AI over MCP. Describe what you want in natural language; your AI assistant reads and writes your Relevance workspace — system prompts, Tools, knowledge, triggers, workforces, evals.

      <CardGroup>
        <Card title="Claude Code plugin" icon="terminal" href="/integrations/mcp/claude-code">
          The fastest setup. Install the Relevance AI plugin and build Agents from your terminal.
        </Card>

        <Card title="MCP server" icon="plug" href="/integrations/mcp/mcp-server">
          Connect from Claude Desktop, Cursor, VS Code, ChatGPT, OpenAI Codex, or any MCP client.
        </Card>

        <Card title="Agent skills" icon="graduation-cap" href="/integrations/mcp/agent-skills">
          Clone the skills repo so your coding assistant has built-in knowledge of Relevance patterns.
        </Card>

        <Card title="MCP & Plugins overview" icon="book" href="/get-started/core-concepts/mcp-plugins">
          The full reference for what you can do programmatically.
        </Card>
      </CardGroup>
    </Tab>
  </Tabs>
</div>

## The four levels in customer success

How the [autonomy ladder](/docs/guides) plays out across your book of business:

<CardGroup>
  <Card title="L1 — Assisted" icon="hand">
    A CSM asks the Agent to summarize Acme's last quarter before a check-in — the Agent returns a snapshot with usage, tickets, and any deal-stage shifts.
  </Card>

  <Card title="L2 — Copilot" icon="map">
    A CSM asks for renewal plans across 12 accounts — the Agent drafts each plan, the CSM reviews and queues outreach.
  </Card>

  <Card title="L3 — Autopilot" icon="bolt">
    Product usage drops on a key account → health score updates → the CSM is alerted with a draft outreach plan, all without anyone touching a dashboard.
  </Card>

  <Card title="L4 — Self-driving" icon="infinity">
    You set the objective — "reduce gross churn" — and the Workforce scores health, drafts outreach, and logs to the CRM, with its risk-signal criteria tuned through evals as churn outcomes flow back.
  </Card>
</CardGroup>

## Zoom into a single use case

Each use case below is a working entry point — clone a template, then run it, shape it visually, or build with AI depending on which tab fit you.

<CardGroup>
  <Card title="Account health & churn signals" icon="heart-pulse" href="/guides/customer-success/account-health">
    Score every account on health with reasoning, not just a number. Surface risk before renewal.
  </Card>

  <Card title="QBR & business review prep" icon="presentation-screen" href="/guides/customer-success/qbr-prep">
    Auto-generate exec-ready QBR decks from CRM, support, and usage data — minutes, not hours.
  </Card>

  <Card title="Renewal & expansion outreach" icon="handshake" href="/guides/customer-success/renewal-expansion">
    Draft account-aware renewal and expansion plays for every account, every quarter.
  </Card>
</CardGroup>

## See it in production

<Frame>
  <iframe title="Canva x Relevance AI — Agents@Work" />
</Frame>


# QBR and business review prep
Source: https://relevanceai.com/docs/guides/customer-success/qbr-prep

Auto-generate exec-ready QBR decks from CRM, support, and usage data — minutes, not hours.

A QBR is supposed to take an hour to prep and instead takes half a day. The CSM pulls usage data, support history, deal notes, and an exec summary into a slide deck — and does that for 12 accounts a quarter. A QBR Agent does the assembly so the CSM spends time on the *insight*, not the data hunting.

## When this pays off

<CardGroup>
  <Card title="QBRs slip or shrink" icon="calendar-xmark">
    Reviews get canceled or shortened because nobody had time to prep — the customer notices.
  </Card>

  <Card title="Inconsistent across CSMs" icon="scale-unbalanced">
    Senior CSMs ship excellent QBRs; newer ones ship rough ones — the customer experience varies wildly.
  </Card>

  <Card title="Manual data pulling" icon="hammer">
    CSMs are exporting from HubSpot, screenshotting Looker, copying from Zendesk — the same way every time.
  </Card>

  <Card title="Strategic content is rushed" icon="hourglass">
    The "next quarter recommendation" slide always gets written last, briefly, because data prep ate the time.
  </Card>
</CardGroup>

## The shape of this use case

A QBR Agent takes an account + time window and returns a structured review with sourced data.

<CardGroup>
  <Card title="Inputs" icon="arrow-right-to-bracket">
    Account, review period, audience (operational vs exec), template / brand to follow.
  </Card>

  <Card title="Sources" icon="globe">
    CRM activity, support ticket history, product analytics, deal notes, your QBR template, segment playbooks.
  </Card>

  <Card title="Output" icon="file-lines">
    A draft QBR — usage summary, support summary, business outcomes, recommended next quarter, cited data — in your team's format.
  </Card>

  <Card title="Delivery" icon="paper-plane">
    Posted to a [Notion](/docs/integrations/popular-integrations/notion) or Google Doc for the CSM to refine, emailed as a draft, dropped into a [Slack](/docs/integrations/popular-integrations/slack) thread for review.
  </Card>
</CardGroup>

## Where to start

Two ways in, depending on whether you want something running today or built to your exact spec.

<CardGroup>
  <Card title="Clone a pre-built Agent" icon="copy">
    Open the **[Customer Success Manager](https://marketplace.relevanceai.com/listing/b9b5e05c-4512-4749-8f66-4623ab895e21)**. More in the [Marketplace](/docs/get-started/marketplace/introduction).
  </Card>

  <Card title="Build your own" icon="hammer">
    Start from scratch in the [builder](/docs/build/introduction), or by describing it in Claude Code or Cursor with [MCP & Plugins](/docs/get-started/core-concepts/mcp-plugins).
  </Card>
</CardGroup>

Either way, these are prompts your CSMs can use on day one:

* *"Draft a QBR for Acme Corp covering last quarter — usage, support summary, business wins, recommended next steps."*
* *"What did Globex achieve this year vs. their stated goals? Pull from CRM notes and product usage."*
* *"Give me the operational summary slide for Initech's QBR — usage trends and support trend only."*

## Where to take it

Once it's running, deepen it in three moves:

<CardGroup>
  <Card title="Give it a playbook" icon="book">
    Shape it with a [prompt](/docs/build/agents/build-your-agent/prompt), your QBR template in [Knowledge](/docs/build/knowledge/create-knowledge), and a connected [CRM](/docs/integrations/popular-integrations/hubspot), support, and analytics.
  </Card>

  <Card title="Automate it on signals" icon="bolt">
    Wrap it in a [workflow](/docs/build/workforces/create-a-workforce) that fires on a [trigger](/docs/build/agents/build-your-agent/triggers).
  </Card>

  <Card title="Let it improve" icon="arrows-rotate">
    Feed back which QBR sections landed into the Agent's [evals](/docs/build/agents/build-your-agent/evals) so the format tracks what's used.
  </Card>
</CardGroup>

## Common pitfalls

<AccordionGroup>
  <Accordion title="Data without narrative" icon="chart-line">
    A QBR that lists metrics without telling a story is just a report. Force the prompt to summarize what changed and why it matters.
  </Accordion>

  <Accordion title="Same QBR for every segment" icon="clone">
    Enterprise QBRs and mid-market QBRs should look different. Branch the prompt on segment so the depth and audience tone shift accordingly.
  </Accordion>

  <Accordion title="Auto-sending without CSM review" icon="paper-plane">
    Tempting to skip the CSM read. Don't — a misread on a strategic account is a relationship cost. The Agent drafts, the CSM ships.
  </Accordion>

  <Accordion title="Stale or wrong attribution" icon="database">
    If the Agent pulls product usage but doesn't filter by account properly, you get the wrong numbers in front of an exec. Validate the data sources for the top 10 accounts before going live.
  </Accordion>
</AccordionGroup>


# Renewal and expansion outreach
Source: https://relevanceai.com/docs/guides/customer-success/renewal-expansion

Draft account-aware renewal and expansion plays for every account, every quarter — so no opportunity slips.

The accounts most likely to expand and the accounts most likely to churn both want personalized attention at renewal. With 80 accounts per CSM, "personalized" usually means "whoever has the loudest email this week." A renewal Agent drafts the playbook touch for every account so nothing is on autopilot — and nothing falls through.

## When this pays off

<CardGroup>
  <Card title="Renewals are scattergun" icon="shuffle">
    Some accounts get a renewal plan three months out, others get a check-in email two weeks before.
  </Card>

  <Card title="Expansion is missed" icon="chart-line">
    Accounts that look healthy don't get expansion conversations because nobody flagged them.
  </Card>

  <Card title="Renewal emails sound canned" icon="comments">
    Every renewal email reads the same — the customer can tell.
  </Card>

  <Card title="QBR action items don't follow through" icon="hourglass-half">
    Promises made at QBR aren't being tracked into the renewal cycle.
  </Card>
</CardGroup>

## The shape of this use case

A renewal Agent takes an account + renewal window and returns a tailored plan with drafted outreach.

<CardGroup>
  <Card title="Inputs" icon="arrow-right-to-bracket">
    Account, renewal date, health score, segmentation, last touch, prior QBR commitments.
  </Card>

  <Card title="Sources" icon="globe">
    CRM, prior QBRs, support history, usage data, your renewal playbook, expansion criteria.
  </Card>

  <Card title="Output" icon="file-lines">
    A renewal plan — recommended action sequence, drafted outreach for each step, expansion talking points where they apply.
  </Card>

  <Card title="Delivery" icon="paper-plane">
    Plan saved to the CRM, outreach drafts queued in [Gmail](/docs/integrations/popular-integrations/gmail), expansion intel posted to the AE in [Slack](/docs/integrations/popular-integrations/slack).
  </Card>
</CardGroup>

## Where to start

Two ways in, depending on whether you want something running today or built to your exact spec.

<CardGroup>
  <Card title="Clone a pre-built Agent" icon="copy">
    Open **[Apla, the Account Planning Agent](https://marketplace.relevanceai.com/listing/fd712abc-66e4-4507-8f16-c4404cf5f5e5)**. More in the [Marketplace](/docs/get-started/marketplace/introduction).
  </Card>

  <Card title="Build your own" icon="hammer">
    Start from scratch in the [builder](/docs/build/introduction), or by describing it in Claude Code or Cursor with [MCP & Plugins](/docs/get-started/core-concepts/mcp-plugins).
  </Card>
</CardGroup>

Either way, these are prompts your CSMs can use on day one:

* *"Acme Corp renews in 90 days — give me a 3-touch renewal plan with drafted emails, plus any expansion angle worth raising."*
* *"What did we promise Globex at the last QBR? Draft a follow-up confirming we delivered before their renewal."*
* *"Is Initech worth a price increase conversation this renewal? Pull usage and engagement signals to make the case."*

## Where to take it

Once it's running, deepen it in three moves:

<CardGroup>
  <Card title="Give it a playbook" icon="book">
    Shape it with a [prompt](/docs/build/agents/build-your-agent/prompt), your renewal playbook in [Knowledge](/docs/build/knowledge/create-knowledge), and [Bulk Schedule](/docs/build/agents/give-your-agent-tasks/bulk-schedule).
  </Card>

  <Card title="Automate it on signals" icon="bolt">
    Wrap it in a [workflow](/docs/build/workforces/create-a-workforce) that fires on a [trigger](/docs/build/agents/build-your-agent/triggers).
  </Card>

  <Card title="Let it improve" icon="arrows-rotate">
    Feed renewal outcomes back into the Agent's [evals](/docs/build/agents/build-your-agent/evals) so it leans on the plays that saved accounts.
  </Card>
</CardGroup>

## Common pitfalls

<AccordionGroup>
  <Accordion title="Same template for every account" icon="copy">
    Without per-account context (last QBR commitments, segment, usage pattern), every renewal email reads identical. Pull from past account interactions, not just static templates.
  </Accordion>

  <Accordion title="Auto-sending the renewal pitch" icon="paper-plane">
    Renewal outreach to a strategic account that goes out without a human read is how trust gets burned. Keep CSM-in-the-loop until you've watched it for several cycles.
  </Accordion>

  <Accordion title="Missing the expansion moment" icon="bullseye">
    A healthy renewal that doesn't ask about expansion is a missed quarter. Build expansion checks into the same use case, not as a separate pass.
  </Accordion>

  <Accordion title="No tracking of what's been said" icon="list-check">
    The renewal pitch contradicts what got said at QBR. Have the Agent reference prior commitments from CRM notes before drafting.
  </Accordion>
</AccordionGroup>


# Customer Support
Source: https://relevanceai.com/docs/guides/customer-support/getting-started

Put AI Agents to work across the support queue — triage, drafting, knowledge — so agents handle the cases that need a human.

Support teams are buried in repetitive volume that's faster to answer than to escalate. AI Agents take the routine — triaging tickets, drafting first replies, generating knowledge base articles from resolved cases — so human agents handle the cases that need judgment.

## Start with a Marketplace template

The Marketplace ships pre-built Agents you can clone in one click. Adapt them to your products, your tone of voice, your escalation rules. Or [build a custom Agent from scratch](/docs/build/introduction).

<CardGroup>
  <Card title="Customer Query Handler" icon="message" href="https://marketplace.relevanceai.com/listing/5dc6f8ea-a087-4f7f-a053-5870d9292891">
    Answers customer questions from your knowledge base — useful for first-response drafts and self-serve flows.
  </Card>

  <Card title="Customer Support Categorization" icon="tags" href="https://marketplace.relevanceai.com/listing/bffa15a4-5d7c-4408-82a0-6a09fa3d02a9">
    Classifies tickets into billing / technical / account / feature / bug and assigns P1–P4 priority on the way in.
  </Card>

  <Card title="Browse the Marketplace" icon="store" href="/get-started/marketplace/introduction">
    The full catalog of Agents — ticket triage, response drafting, KB management, escalation routing.
  </Card>
</CardGroup>

Click **Clone Agent** on any listing, connect your support tools (Zendesk / Intercom / Salesforce Service Cloud), and run on a real ticket.

## Pick your path

How hands-on do you want to be? Pick a tab — the rest of the page is written for you.

<div>
  <Tabs>
    <Tab title="No build">
      You don't need to touch the builder. Open the cloned Agent above, ask in plain English, and let it work across the systems it's connected to.

      ## Sharpen the answers with knowledge

      Upload your support playbook, tone-of-voice doc, escalation rules, and knowledge base articles. Every conversation references them — sharper output, no new setup.

      <Card title="Add knowledge" icon="book" href="/build/knowledge/create-knowledge">
        Upload context once. Every conversation references it.
      </Card>

      ## Make it run on its own

      When the same kinds of tickets keep landing, set a trigger — every billing question gets first-touch drafted, every refund request gets routed to the right team. One toggle, no code, no builder.

      <Card title="Triggers" icon="bolt" href="/build/agents/build-your-agent/triggers">
        Schedule runs, fire on ticket events, listen to webhooks.
      </Card>

      ## Ready to go further?

      When prompting and templates stop being enough — when you want the Agent to follow your playbook word for word and orchestrate triage across channels — pick the **Build visually** tab above.
    </Tab>

    <Tab title="Build visually">
      You're comfortable in product UIs. You don't want to wire up an Agent from scratch, but you do want it to follow your support playbook ticket by ticket. Open the cloned Agent in the builder and shape every piece.

      ## Connect the tools it needs

      Support Agents work best when they can read from and write to the systems your team already uses.

      * **Ticketing** — [Zendesk](/docs/integrations/popular-integrations/zendesk), Intercom, or Salesforce Service Cloud for tickets and case history
      * **Knowledge base** — your KB system for citation-grounded responses
      * **CRM** — [HubSpot](/docs/integrations/popular-integrations/hubspot) or [Salesforce](/docs/integrations/popular-integrations/salesforce) for customer context
      * **Comms** — [Gmail](/docs/integrations/popular-integrations/gmail), [Slack](/docs/integrations/popular-integrations/slack) for follow-up and internal escalation

      See more about [Integrations](/docs/integrations/introduction).

      ## Shape every surface

      Each piece below changes how the Agent behaves. Run it on a real ticket between every change — small iterations beat big rewrites.

      <CardGroup>
        <Card title="Prompt" icon="message" href="/build/agents/build-your-agent/prompt">
          Edit the system prompt to match your tone, your escalation rules, your tier-1 boundary.
        </Card>

        <Card title="Tools" icon="wrench" href="/build/agents/build-your-agent/tools">
          Add or remove the actions the Agent can take — search the KB, query the CRM, draft replies, route tickets.
        </Card>

        <Card title="Knowledge" icon="book" href="/build/knowledge/create-knowledge">
          Upload your KB articles, support playbook, and tone-of-voice guides. The Agent draws on these in every reply.
        </Card>

        <Card title="Triggers" icon="bolt" href="/build/agents/build-your-agent/triggers">
          Run the Agent automatically — on every new ticket, on tag updates, on SLA breaches.
        </Card>

        <Card title="Alerts" icon="bell" href="/build/agents/build-your-agent/alerts">
          Get notified when CSAT drops, when an SLA is at risk, when an escalation pattern emerges.
        </Card>

        <Card title="Memory" icon="brain" href="/build/agents/build-your-agent/memory">
          Persistent context per customer — past tickets, known account context, prior resolutions.
        </Card>
      </CardGroup>

      ## Chain Agents into a workforce

      When a single Agent isn't enough — triage, then draft, then update the CRM, then notify the team — wire them on the Workforce canvas.

      <Card title="Build a workforce" icon="diagram-project" href="/build/workforces/create-a-workforce">
        Drag, connect, and run multi-Agent pipelines without code.
      </Card>

      ## Test before you trust

      Set up evals so regressions in tone or accuracy get caught before the Agent replies to a customer.

      <Card title="Evals" icon="vial" href="/build/agents/build-your-agent/evals">
        Define test cases in the UI. Run them on every change.
      </Card>

      ## Ready to go further?

      When clicking through the builder starts to feel slow — when you'd rather have your AI coding assistant build and edit Agents for you, drop into custom Python or JS code blocks, or call Agents from your own app via the SDK — pick the **Build with AI** tab above.
    </Tab>

    <Tab title="Build with AI">
      You already work in an AI coding environment. Clicking through a UI to edit response templates feels slow. Have your AI assistant build, shape, and orchestrate Agents directly from your editor.

      ## MCP & Plugins

      Connect your AI coding environment to Relevance AI over MCP. Describe what you want in natural language; your AI assistant reads and writes your Relevance workspace — system prompts, Tools, knowledge, triggers, workforces, evals.

      <CardGroup>
        <Card title="Claude Code plugin" icon="terminal" href="/integrations/mcp/claude-code">
          The fastest setup. Install the Relevance AI plugin and build Agents from your terminal.
        </Card>

        <Card title="MCP server" icon="plug" href="/integrations/mcp/mcp-server">
          Connect from Claude Desktop, Cursor, VS Code, ChatGPT, OpenAI Codex, or any MCP client.
        </Card>

        <Card title="Agent skills" icon="graduation-cap" href="/integrations/mcp/agent-skills">
          Clone the skills repo so your coding assistant has built-in knowledge of Relevance patterns.
        </Card>

        <Card title="MCP & Plugins overview" icon="book" href="/get-started/core-concepts/mcp-plugins">
          The full reference for what you can do programmatically.
        </Card>
      </CardGroup>
    </Tab>
  </Tabs>
</div>

## The four levels in support

How the [autonomy ladder](/docs/guides) plays out across the support queue:

<CardGroup>
  <Card title="L1 — Assisted" icon="hand">
    An agent opens a ticket and asks the Agent to summarize the customer's history and suggest a reply — the Agent returns context with a draft response.
  </Card>

  <Card title="L2 — Copilot" icon="map">
    An agent reviews a batch of new tickets — the Agent has already drafted replies and triage suggestions, the agent approves and sends.
  </Card>

  <Card title="L3 — Autopilot" icon="bolt">
    A new ticket lands → triaged, tagged, routed, first-reply drafted and auto-sent on simple cases, all without an agent touching the queue.
  </Card>

  <Card title="L4 — Self-driving" icon="infinity">
    You set the objective — "cut time-to-first-response" — and the Workforce triages, drafts, and routes, with its triage and template criteria tuned through evals as CSAT outcomes flow back.
  </Card>
</CardGroup>

## Zoom into a single use case

Each use case below is a working entry point — clone a template, then run it, shape it visually, or build with AI depending on which tab fit you.

<CardGroup>
  <Card title="Ticket triage and routing" icon="route" href="/guides/customer-support/ticket-triage">
    Classify, tag, and route every inbound ticket so the right agent picks it up first.
  </Card>

  <Card title="Response drafting from KB" icon="reply" href="/guides/customer-support/response-drafting">
    First-draft replies grounded in your knowledge base, with citations so the agent can spot-check.
  </Card>

  <Card title="KB article generation" icon="book-open" href="/guides/customer-support/kb-generation">
    Turn resolved tickets into knowledge base articles — close the loop without writing them by hand.
  </Card>
</CardGroup>

## See it in production

<Frame>
  <iframe title="Canva x Relevance AI — Agents@Work" />
</Frame>


# KB article generation from resolved tickets
Source: https://relevanceai.com/docs/guides/customer-support/kb-generation

Turn resolved tickets into knowledge base articles — close the loop without writing them by hand.

Every support team has the same problem: the KB is the most leveraged thing in the operation, and it's the thing nobody has time to write. A KB Agent watches resolved tickets, spots the patterns worth documenting, and drafts the articles — so the next ticket of that shape gets a faster answer.

## When this pays off

<CardGroup>
  <Card title="KB hasn't grown in months" icon="calendar-xmark">
    New product features ship every sprint; the KB last got an article six months ago.
  </Card>

  <Card title="Same questions, same drafting" icon="rotate">
    Agents answer the same niche question 30 times a quarter — and never have time to write the article.
  </Card>

  <Card title="Tribal knowledge in agent heads" icon="user-group">
    Senior agents know the answers; new hires don't, because nobody documented them.
  </Card>

  <Card title="Self-serve is underperforming" icon="user-magnifying-glass">
    Customer self-serve traffic is low because there's no article matching their search.
  </Card>
</CardGroup>

## The shape of this use case

A KB Agent takes a resolved ticket (or cluster of tickets) and returns a publishable article.

<CardGroup>
  <Card title="Inputs" icon="arrow-right-to-bracket">
    Resolved ticket(s), category, prior similar resolutions, your KB style guide.
  </Card>

  <Card title="Sources" icon="globe">
    Ticket conversations, agent resolution notes, product changelog, your KB style and structure guides.
  </Card>

  <Card title="Output" icon="file-lines">
    A drafted KB article — title, summary, steps, screenshots placeholders, related-articles links — ready for review.
  </Card>

  <Card title="Delivery" icon="paper-plane">
    Posted to your KB platform as a draft, dropped in [Notion](/docs/integrations/popular-integrations/notion) for review, queued in a "needs editor" workflow.
  </Card>
</CardGroup>

## Where to start

Two ways in, depending on whether you want something running today or built to your exact spec.

<CardGroup>
  <Card title="Clone a pre-built Agent" icon="copy">
    Open **[Tim, Technical Documentation Generator](https://marketplace.relevanceai.com/listing/214b9ba5-7e83-4c54-8494-dcaed6557aeb)**. More in the [Marketplace](/docs/get-started/marketplace/introduction).
  </Card>

  <Card title="Build your own" icon="hammer">
    Start from scratch in the [builder](/docs/build/introduction), or by describing it in Claude Code or Cursor with [MCP & Plugins](/docs/get-started/core-concepts/mcp-plugins).
  </Card>
</CardGroup>

Either way, these are prompts your team can use on day one:

* *"Look at the last 50 resolved tickets in the 'billing' category — what 3 articles should we write that we don't have?"*
* *"Turn this ticket into a KB article. Customer hit the SSO bug last week and we resolved with workaround X."*
* *"What's the most common search query on our help site this month that has no matching article?"*

## Where to take it

Once it's running, deepen it in three moves:

<CardGroup>
  <Card title="Give it a playbook" icon="book">
    Shape it with a [prompt](/docs/build/agents/build-your-agent/prompt), your style guide in [Knowledge](/docs/build/knowledge/create-knowledge), and [Bulk Schedule](/docs/build/agents/give-your-agent-tasks/bulk-schedule).
  </Card>

  <Card title="Automate it on signals" icon="bolt">
    Wrap it in a [workflow](/docs/build/workforces/create-a-workforce) that fires on a [trigger](/docs/build/agents/build-your-agent/triggers).
  </Card>

  <Card title="Let it improve" icon="arrows-rotate">
    Feed read and deflection data back into the Agent's [evals](/docs/build/agents/build-your-agent/evals) so it drafts the formats that help.
  </Card>
</CardGroup>

## Common pitfalls

<AccordionGroup>
  <Accordion title="Articles that paraphrase one ticket" icon="clone">
    Drafting from a single ticket gives you an article that only solves that exact case. Cluster similar tickets first so the article generalizes.
  </Accordion>

  <Accordion title="Stale or contradictory content" icon="clock-rotate-left">
    Auto-drafting without checking existing articles produces overlap and contradiction. Have the Agent search the KB first and propose an *edit* if an article already covers it.
  </Accordion>

  <Accordion title="Auto-publishing without review" icon="eye">
    KB articles are externally visible — bad ones cost trust. Always gate publishing with a human editor until you've watched it for two quarters.
  </Accordion>

  <Accordion title="No tie to actual search demand" icon="magnifying-glass">
    Drafting articles agents *would have used* is fine, but missing the bigger win: articles for what customers *search and can't find*. Connect the Agent to KB search analytics.
  </Accordion>
</AccordionGroup>


# Response drafting from KB
Source: https://relevanceai.com/docs/guides/customer-support/response-drafting

First-draft replies grounded in your knowledge base, with citations so the agent can spot-check before sending.

The fastest support team isn't the one with the most agents — it's the one whose agents don't start replies from a blank page. A response Agent drafts the first reply for every ticket, pulling the right KB articles, the right account context, the right tone. The agent edits, sends, and moves on.

## When this pays off

<CardGroup>
  <Card title="Repetitive tickets eat hours" icon="rotate">
    The same 20 question types make up 70% of inbound. Agents type the same answer with slight variation, all day.
  </Card>

  <Card title="KB underused" icon="book">
    You've got a knowledge base but tickets get answered from agent memory — and the KB grows stale.
  </Card>

  <Card title="Tone drift across agents" icon="comments">
    Tone varies wildly between new hires and tenured agents — customer experience is uneven.
  </Card>

  <Card title="Onboarding takes too long" icon="user-clock">
    New agents take weeks to become productive because there's no fast path to "right answer + right voice".
  </Card>
</CardGroup>

## The shape of this use case

A response Agent takes a ticket and returns a drafted reply with citations.

<CardGroup>
  <Card title="Inputs" icon="arrow-right-to-bracket">
    Ticket body, customer record, prior tickets, channel-specific format hints.
  </Card>

  <Card title="Sources" icon="globe">
    Knowledge base, prior resolutions, tone-of-voice guide, product status / changelog, account history.
  </Card>

  <Card title="Output" icon="file-lines">
    A drafted reply with cited KB articles and a confidence indicator — ready for agent review and send.
  </Card>

  <Card title="Delivery" icon="paper-plane">
    Inserted as a draft in [Zendesk](/docs/integrations/popular-integrations/zendesk) / Intercom for agent review, posted in a comment for visibility, attached as a macro suggestion.
  </Card>
</CardGroup>

## Where to start

Two ways in, depending on whether you want something running today or built to your exact spec.

<CardGroup>
  <Card title="Clone a pre-built Agent" icon="copy">
    Open the **[Customer Query Handler](https://marketplace.relevanceai.com/listing/5dc6f8ea-a087-4f7f-a053-5870d9292891)**. More in the [Marketplace](/docs/get-started/marketplace/introduction).
  </Card>

  <Card title="Build your own" icon="hammer">
    Start from scratch in the [builder](/docs/build/introduction), or by describing it in Claude Code or Cursor with [MCP & Plugins](/docs/get-started/core-concepts/mcp-plugins).
  </Card>
</CardGroup>

Either way, these are prompts your team can use on day one:

* *"Draft a reply to this ticket — the customer can't activate their account and we've seen this before with their email domain."*
* *"What's our standard response when someone asks about the new billing cycle? Cite the KB articles."*
* *"Customer is asking for a refund — what's the playbook and what should I check first?"*

## Where to take it

Once it's running, deepen it in three moves:

<CardGroup>
  <Card title="Give it a playbook" icon="book">
    Shape it with a [prompt](/docs/build/agents/build-your-agent/prompt), your KB in [Knowledge](/docs/build/knowledge/create-knowledge), and [Tools](/docs/build/agents/build-your-agent/tools) for KB and CRM search.
  </Card>

  <Card title="Automate it on signals" icon="bolt">
    Wrap it in a [workflow](/docs/build/workforces/create-a-workforce) that fires on a [trigger](/docs/build/agents/build-your-agent/triggers).
  </Card>

  <Card title="Let it improve" icon="arrows-rotate">
    Feed back which drafts shipped untouched, plus CSAT, into the Agent's [evals](/docs/build/agents/build-your-agent/evals) so it matches what your team sends.
  </Card>
</CardGroup>

## Common pitfalls

<AccordionGroup>
  <Accordion title="Drafts agents have to rewrite" icon="pen">
    If agents are rewriting every draft, the Agent isn't actually saving time — it's adding a step. Watch the edit-rate; rebuild the prompt or Knowledge if it climbs.
  </Accordion>

  <Accordion title="Hallucinated KB references" icon="link-slash">
    The Agent cites articles that don't exist or fabricates policy. Force citations to KB IDs from your actual catalog and fail visibly when there's no match.
  </Accordion>

  <Accordion title="Tone gets robotic" icon="robot">
    Without a strong tone-of-voice doc, every draft sounds like the same chatbot. Upload your best past replies as examples in Knowledge.
  </Accordion>

  <Accordion title="Auto-send before you're ready" icon="paper-plane">
    Auto-sending replies before you've watched the Agent on real tickets for a month is how you ship a bad answer at scale. Start with drafts only.
  </Accordion>
</AccordionGroup>


# Ticket triage and routing
Source: https://relevanceai.com/docs/guides/customer-support/ticket-triage

Classify, tag, and route every inbound ticket so the right agent picks it up first.

The first 30 seconds of a ticket's life shapes its whole trajectory. If it's mis-tagged, mis-routed, or lands in the wrong queue, response time blows out. A triage Agent reads each inbound ticket the moment it lands and tags, prioritizes, and routes it so the right human picks it up.

## When this pays off

<CardGroup>
  <Card title="Manual triage backlog" icon="layer-group">
    A lead or senior agent spends an hour every morning reading and routing the overnight backlog.
  </Card>

  <Card title="Mis-routed tickets" icon="route">
    Tickets sit in the wrong queue for hours before someone re-routes them.
  </Card>

  <Card title="Inconsistent tagging" icon="tags">
    Tags differ across agents, so reporting and trend analysis are unreliable.
  </Card>

  <Card title="VIP escalations are missed" icon="crown">
    High-value account issues sit in the general queue because nobody flagged them.
  </Card>
</CardGroup>

## The shape of this use case

A triage Agent takes an inbound ticket and returns a structured classification.

<CardGroup>
  <Card title="Inputs" icon="arrow-right-to-bracket">
    Ticket body, subject, channel (email / chat / form), customer record, prior tickets.
  </Card>

  <Card title="Sources" icon="globe">
    CRM for account context, prior tickets for pattern matching, your taxonomy / tag library, SLA / VIP rules.
  </Card>

  <Card title="Output" icon="file-lines">
    Classification (category, priority, urgency), tags, recommended queue and assignee, with reasoning.
  </Card>

  <Card title="Delivery" icon="paper-plane">
    Applied directly in [Zendesk](/docs/integrations/popular-integrations/zendesk) / Intercom / Service Cloud, posted to [Slack](/docs/integrations/popular-integrations/slack) for VIP escalations, written to a triage log for review.
  </Card>
</CardGroup>

## Where to start

Two ways in, depending on whether you want something running today or built to your exact spec.

<CardGroup>
  <Card title="Clone a pre-built Agent" icon="copy">
    Open **[Customer Support Categorization](https://marketplace.relevanceai.com/listing/bffa15a4-5d7c-4408-82a0-6a09fa3d02a9)**. More in the [Marketplace](/docs/get-started/marketplace/introduction).
  </Card>

  <Card title="Build your own" icon="hammer">
    Start from scratch in the [builder](/docs/build/introduction), or by describing it in Claude Code or Cursor with [MCP & Plugins](/docs/get-started/core-concepts/mcp-plugins).
  </Card>
</CardGroup>

Either way, these are prompts your team can use on day one:

* *"How would you triage this ticket? Customer is on Enterprise tier, ticket mentions a billing error and a feature outage."*
* *"Walk me through these 10 overnight tickets — what should we pick up first and why?"*
* *"Is this a P1? Customer says 'urgent' but the issue is a typo in our marketing site."*

## Where to take it

Once it's running, deepen it in three moves:

<CardGroup>
  <Card title="Give it a playbook" icon="book">
    Shape it with a [prompt](/docs/build/agents/build-your-agent/prompt), [Knowledge](/docs/build/knowledge/create-knowledge), and [Bulk Schedule](/docs/build/agents/give-your-agent-tasks/bulk-schedule).
  </Card>

  <Card title="Automate it on signals" icon="bolt">
    Wrap it in a [workflow](/docs/build/workforces/create-a-workforce) that fires on a [trigger](/docs/build/agents/build-your-agent/triggers).
  </Card>

  <Card title="Let it improve" icon="arrows-rotate">
    Feed resolved-ticket outcomes back into the Agent's [evals](/docs/build/agents/build-your-agent/evals) so its triage tracks what actually held up.
  </Card>
</CardGroup>

## Common pitfalls

<AccordionGroup>
  <Accordion title="Tag drift" icon="tag">
    The Agent invents new tags rather than using your existing taxonomy. Force the prompt to choose from a fixed Knowledge list — and fail closed if no tag fits.
  </Accordion>

  <Accordion title="Overconfident priority decisions" icon="scale-unbalanced">
    The Agent marks every "urgent"-flagged email as P1. Train it to read the actual issue against your priority matrix, not the customer's adjective.
  </Accordion>

  <Accordion title="Routing without account context" icon="route">
    Tickets get routed by topic only and miss the account dimension — a P3 from a churning enterprise account isn't a P3. Pull CRM context into the triage prompt.
  </Accordion>

  <Accordion title="Auto-applying without review" icon="robot">
    A wrong tag at scale corrupts reporting and routing. Start with Agent-as-suggester (writes to a "suggested\_priority" field) and graduate to direct application after a quarter of audit.
  </Accordion>
</AccordionGroup>


# Campaign performance analytics
Source: https://relevanceai.com/docs/guides/marketing/campaign-analytics

Pull cross-channel campaign reports together and surface what's working — without hours of spreadsheet wrangling.

Performance data lives in seven tools. The weekly campaign review takes someone half a day to pull together and another half day to translate into insight. A campaign analytics Agent does the pulling and the translating — so the review meeting starts with answers, not data hunting.

## When this pays off

<CardGroup>
  <Card title="Reports take half a day" icon="hourglass-half">
    Pulling Google Ads + HubSpot + LinkedIn + GA into one slide takes the same person, the same way, every Monday.
  </Card>

  <Card title="Channel performance is opaque" icon="eye-slash">
    You know spend by channel but not what each channel is actually doing for the funnel.
  </Card>

  <Card title="Campaign comparisons are manual" icon="scale-balanced">
    Was this campaign better than the last one? Nobody can answer without an hour of cross-referencing.
  </Card>

  <Card title="Leadership wants weekly digests" icon="newspaper">
    The CMO wants a Friday "what worked, what didn't" — and it always slips because it's not anyone's job.
  </Card>
</CardGroup>

## The shape of this use case

A campaign analytics Agent takes a time window + campaign set and returns a report with reasoning.

<CardGroup>
  <Card title="Inputs" icon="arrow-right-to-bracket">
    Time window, campaign set (specific campaigns or all), channels to include, segmentation cuts.
  </Card>

  <Card title="Sources" icon="globe">
    Marketing automation ([HubSpot](/docs/integrations/popular-integrations/hubspot), [Marketo](/docs/integrations/popular-integrations/marketo)), ad platforms (Google, Meta, [LinkedIn](/docs/integrations/popular-integrations/linkedin)), web analytics, CRM, your own benchmark docs.
  </Card>

  <Card title="Output" icon="file-lines">
    A digest — top performers, underperformers, key shifts vs. last period, suggested actions — with cited data and reasoning.
  </Card>

  <Card title="Delivery" icon="paper-plane">
    Posted to [Slack](/docs/integrations/popular-integrations/slack), written to a [Notion](/docs/integrations/popular-integrations/notion) page for the team review, emailed to the CMO Friday morning.
  </Card>
</CardGroup>

## Where to start

Two ways in, depending on whether you want something running today or built to your exact spec.

<CardGroup>
  <Card title="Clone a pre-built Agent" icon="copy">
    Open the **[Email Marketing Strategist](https://marketplace.relevanceai.com/listing/28783038-03f0-4aec-9c76-4b28452b6e33)**. More in the [Marketplace](/docs/get-started/marketplace/introduction).
  </Card>

  <Card title="Build your own" icon="hammer">
    Start from scratch in the [builder](/docs/build/introduction), or by describing it in Claude Code or Cursor with [MCP & Plugins](/docs/get-started/core-concepts/mcp-plugins).
  </Card>
</CardGroup>

Either way, these are prompts your team can use on day one:

* *"Pull last week's campaign performance from HubSpot and Google Ads — top 3 winners, top 3 losers, with reasons."*
* *"Compare the Q3 onboarding sequence against Q2 — what shifted in open and click rates?"*
* *"Which channels drove qualified pipeline last month? Cite the data."*

## Where to take it

Once it's running, deepen it in three moves:

<CardGroup>
  <Card title="Give it a playbook" icon="book">
    Shape it with a [prompt](/docs/build/agents/build-your-agent/prompt), KPI definitions in [Knowledge](/docs/build/knowledge/create-knowledge), and [Tools](/docs/build/agents/build-your-agent/tools) to query each platform.
  </Card>

  <Card title="Automate it on signals" icon="bolt">
    Wrap it in a [workflow](/docs/build/workforces/create-a-workforce) that fires on a [trigger](/docs/build/agents/build-your-agent/triggers).
  </Card>

  <Card title="Let it improve" icon="arrows-rotate">
    Weight the Agent's [evals](/docs/build/agents/build-your-agent/evals) toward the metrics that actually predicted pipeline.
  </Card>
</CardGroup>

## Common pitfalls

<AccordionGroup>
  <Accordion title="Pulling data without interpreting it" icon="magnifying-glass">
    A wall of numbers in Slack isn't useful. Force the Agent to explain *why* metrics moved and *what to do* — not just *what they are*.
  </Accordion>

  <Accordion title="Inconsistent attribution" icon="scale-unbalanced">
    HubSpot's "campaign" and your ad platform's "campaign" probably aren't the same thing. Define attribution rules in Knowledge once and reference them every time.
  </Accordion>

  <Accordion title="No benchmark for 'good'" icon="bullseye">
    Without baselines, every CTR looks normal. Upload your historical benchmarks to Knowledge so the Agent flags actual outliers.
  </Accordion>

  <Accordion title="Stale data sources" icon="clock-rotate-left">
    Ad platforms can lag 24-48 hours. Have the Agent disclose data freshness in the report so the team knows what's preliminary.
  </Accordion>
</AccordionGroup>


# Content creation and repurposing
Source: https://relevanceai.com/docs/guides/marketing/content-repurposing

Turn one piece of content into many — blog posts, social snippets, ad copy, email teasers — without losing the source voice.

Every long-form asset is supposed to become a quarter's worth of derivative content. In practice, the blog goes up, the webinar drops, and the LinkedIn posts never get written. A content Agent turns one source asset into channel-specific drafts — pulling your brand voice and past top-performers from knowledge, then queuing each derivative for review before it ships.

## When this pays off

<CardGroup>
  <Card title="Content backlog never ships" icon="boxes-stacked">
    Long-form is published; the social/email/ad versions are sitting in a Slack channel labeled "todo".
  </Card>

  <Card title="Channel-specific rewriting" icon="repeat">
    Every channel needs its own voice — copying the headline into Twitter doesn't work. Manual rewrites take hours per asset.
  </Card>

  <Card title="Webinar / podcast assets sit idle" icon="microphone">
    Hours of recorded content with no repurposing — clips, summaries, follow-up emails are all theoretically going to happen.
  </Card>

  <Card title="Big launch coming up" icon="rocket">
    A product launch needs 30+ derivative pieces and the content team has time for 5.
  </Card>
</CardGroup>

## The shape of this use case

A content Agent takes a source asset + target formats and returns ready-to-publish derivatives.

<CardGroup>
  <Card title="Inputs" icon="arrow-right-to-bracket">
    Source asset (blog, webinar transcript, video, whitepaper), target channels and formats, campaign goal.
  </Card>

  <Card title="Sources" icon="globe">
    Brand voice docs, channel-specific style guides, past top-performing posts per channel.
  </Card>

  <Card title="Output" icon="file-lines">
    A bundle of derivatives — blog posts, LinkedIn carousels, X threads, email teasers, ad copy variations — each tuned to its channel.
  </Card>

  <Card title="Delivery" icon="paper-plane">
    Posted to [Notion](/docs/integrations/popular-integrations/notion) or a content doc for review, drafted into your scheduler (Buffer / Hootsuite), pushed to ad platforms via integration.
  </Card>
</CardGroup>

## Where to start

Two ways in, depending on whether you want something running today or built to your exact spec.

<CardGroup>
  <Card title="Clone a pre-built Agent" icon="copy">
    Open the **[Simple Content Repurposer](https://marketplace.relevanceai.com/listing/870991de-99fd-4453-83bf-084c872e7cb2)**. More in the [Marketplace](/docs/get-started/marketplace/introduction).
  </Card>

  <Card title="Build your own" icon="hammer">
    Start from scratch in the [builder](/docs/build/introduction), or by describing it in Claude Code or Cursor with [MCP & Plugins](/docs/get-started/core-concepts/mcp-plugins).
  </Card>
</CardGroup>

Either way, these are prompts your team can use on day one:

* *"Turn this blog post into 5 LinkedIn posts, 3 X threads, and a teaser email. Match our voice — direct, no hype."*
* *"Pull the 4 strongest quotes from this webinar transcript and write them as standalone social posts."*
* *"Take this product launch one-pager and give me 10 ad headline variations for paid social."*

## Where to take it

Once it's running, deepen it in three moves:

<CardGroup>
  <Card title="Give it a playbook" icon="book">
    Shape it with a [prompt](/docs/build/agents/build-your-agent/prompt), your brand voice in [Knowledge](/docs/build/knowledge/create-knowledge), and [Bulk Schedule](/docs/build/agents/give-your-agent-tasks/bulk-schedule).
  </Card>

  <Card title="Automate it on signals" icon="bolt">
    Wrap it in a [workflow](/docs/build/workforces/create-a-workforce) that fires on a [trigger](/docs/build/agents/build-your-agent/triggers).
  </Card>

  <Card title="Let it improve" icon="arrows-rotate">
    Feed channel engagement back into the Agent's [evals](/docs/build/agents/build-your-agent/evals) so it leans on what performed.
  </Card>
</CardGroup>

## Common pitfalls

<AccordionGroup>
  <Accordion title="Same voice across every channel" icon="clone">
    LinkedIn voice, X voice, and email voice are different. Upload per-channel style guides and force the prompt to reference them, otherwise everything ends up sounding LinkedIn-flavored.
  </Accordion>

  <Accordion title="Auto-publishing without review" icon="paper-plane">
    Generated content that goes live without a human read is how brand misfires happen. Keep marketing review gates until you've watched a quarter of output.
  </Accordion>

  <Accordion title="Generic repurposing" icon="copy">
    "Turn this blog into a tweet thread" produces something forgettable without strong style references. Show the Agent your best past threads, not just the source asset.
  </Accordion>

  <Accordion title="Losing the source voice" icon="comment">
    Heavy rewriting can flatten the original author's perspective. If the source is a guest post or executive POV, prompt the Agent to preserve the speaker's voice in derivatives.
  </Accordion>
</AccordionGroup>


# Marketing
Source: https://relevanceai.com/docs/guides/marketing/getting-started

Put AI Agents to work across the marketing funnel — content, campaigns, analytics — without losing your brand voice.

Marketing teams are stretched across channels. AI Agents take the high-volume execution work — drafting nurture sequences, repurposing content, pulling cross-channel reports — so marketers can spend their time on strategy and storytelling.

## Start with a Marketplace template

The Marketplace ships pre-built Agents you can clone in one click. Adapt them to your brand voice, your channels, your funnel. Or [build a custom Agent from scratch](/docs/build/introduction).

<CardGroup>
  <Card title="Email Marketing Strategist" icon="envelope-open-text" href="https://marketplace.relevanceai.com/listing/28783038-03f0-4aec-9c76-4b28452b6e33">
    Analyzes campaign performance, optimizes subject lines, segments audiences, drafts copy, and plans A/B tests.
  </Card>

  <Card title="Simple Content Repurposer" icon="recycle" href="https://marketplace.relevanceai.com/listing/870991de-99fd-4453-83bf-084c872e7cb2">
    Paste a YouTube link, podcast episode, or blog URL → get platform-native LinkedIn and X posts drafted in your voice.
  </Card>

  <Card title="Browse the Marketplace" icon="store" href="/get-started/marketplace/introduction">
    The full catalog of Marketing-relevant Agents — content, social, analytics, lifecycle.
  </Card>
</CardGroup>

Click **Clone Agent** on any listing, connect the channels it needs (email, social, web analytics), and run it on real campaign data.

## Pick your path

How hands-on do you want to be? Pick a tab — the rest of the page is written for you.

<div>
  <Tabs>
    <Tab title="No build">
      You don't need to touch the builder. Open the cloned Agent above, ask in plain English, and let it work across the channels it's connected to.

      ## Sharpen the answers with knowledge

      Upload your brand guidelines, tone-of-voice doc, ICP definitions, and messaging frameworks. Every conversation references them — sharper output, no new setup.

      <Card title="Add knowledge" icon="book" href="/build/knowledge/create-knowledge">
        Upload context once. Every conversation references it.
      </Card>

      ## Make it run on its own

      When you find yourself drafting the same campaign every month, set a trigger — every new MQL gets the onboarding sequence drafted, every product launch triggers a content repurposing pass. One toggle, no code, no builder.

      <Card title="Triggers" icon="bolt" href="/build/agents/build-your-agent/triggers">
        Schedule runs, fire on CRM events, listen to webhooks.
      </Card>

      ## Ready to go further?

      When prompting and templates stop being enough — when you want the Agent to follow your brand voice word for word and orchestrate multi-step campaign automation — pick the **Build visually** tab above.
    </Tab>

    <Tab title="Build visually">
      You're comfortable in product UIs. You don't want to wire up an Agent from scratch, but you do want it to sound like your brand and follow your campaign cadence. Open the cloned Agent in the builder and shape every piece.

      ## Connect the tools it needs

      Marketing Agents work best when they can read from and write to the systems your team already uses.

      * **Marketing automation** — [HubSpot](/docs/integrations/popular-integrations/hubspot) or [Marketo](/docs/integrations/popular-integrations/marketo) for campaigns and lifecycle
      * **Email** — [Gmail](/docs/integrations/popular-integrations/gmail) or [Outlook](/docs/integrations/popular-integrations/outlook) for outbound
      * **Content & social** — [Notion](/docs/integrations/popular-integrations/notion), [Canva](/docs/integrations/popular-integrations/canva), [Slack](/docs/integrations/popular-integrations/slack) for content workflows
      * **Analytics** — [Google Sheets](/docs/integrations/popular-integrations/google-sheets) or your BI tool for performance pulls

      See more about [Integrations](/docs/integrations/introduction).

      ## Shape every surface

      Each piece below changes how the Agent behaves. Run the Agent on a real campaign between every change — small iterations beat big rewrites.

      <CardGroup>
        <Card title="Prompt" icon="message" href="/build/agents/build-your-agent/prompt">
          Edit the system prompt to match your brand voice, segmentation logic, and approval rules.
        </Card>

        <Card title="Tools" icon="wrench" href="/build/agents/build-your-agent/tools">
          Add or remove the actions the Agent can take — pull campaign data, draft email copy, post to Slack.
        </Card>

        <Card title="Knowledge" icon="book" href="/build/knowledge/create-knowledge">
          Upload brand guidelines, persona docs, top-performing past campaigns. The Agent draws on these whenever it works.
        </Card>

        <Card title="Triggers" icon="bolt" href="/build/agents/build-your-agent/triggers">
          Run the Agent automatically — on a schedule, when an MQL enters HubSpot, when a campaign launches.
        </Card>

        <Card title="Alerts" icon="bell" href="/build/agents/build-your-agent/alerts">
          Get notified when CTR drops, when copy fails a brand check, when send caps approach.
        </Card>

        <Card title="Memory" icon="brain" href="/build/agents/build-your-agent/memory">
          Persistent context across campaigns — past performance, which segments responded to which hooks.
        </Card>
      </CardGroup>

      ## Chain Agents into a workforce

      When a single Agent isn't enough — research the audience, write the copy, schedule the send, post a recap — wire them together on the Workforce canvas.

      <Card title="Build a workforce" icon="diagram-project" href="/build/workforces/create-a-workforce">
        Drag, connect, and run multi-Agent pipelines without code.
      </Card>

      ## Test before you trust

      Set up evals so brand-voice regressions get caught before they ship to your list.

      <Card title="Evals" icon="vial" href="/build/agents/build-your-agent/evals">
        Define test cases in the UI. Run them on every change.
      </Card>

      ## Ready to go further?

      When clicking through the builder starts to feel slow — when you'd rather have your AI coding assistant build and edit Agents for you, drop into custom Python or JS code blocks, or call Agents from your own app via the SDK — pick the **Build with AI** tab above.
    </Tab>

    <Tab title="Build with AI">
      You already work in an AI coding environment. Clicking through a UI to edit copy guidelines feels slow. Have your AI assistant build, shape, and orchestrate Agents directly from your editor.

      ## MCP & Plugins

      Connect your AI coding environment to Relevance AI over MCP. Describe what you want in natural language; your AI assistant reads and writes your Relevance workspace — system prompts, Tools, knowledge, triggers, workforces, evals.

      <CardGroup>
        <Card title="Claude Code plugin" icon="terminal" href="/integrations/mcp/claude-code">
          The fastest setup. Install the Relevance AI plugin and build Agents from your terminal.
        </Card>

        <Card title="MCP server" icon="plug" href="/integrations/mcp/mcp-server">
          Connect from Claude Desktop, Cursor, VS Code, ChatGPT, OpenAI Codex, or any MCP client.
        </Card>

        <Card title="Agent skills" icon="graduation-cap" href="/integrations/mcp/agent-skills">
          Clone the skills repo so your coding assistant has built-in knowledge of Relevance patterns.
        </Card>

        <Card title="MCP & Plugins overview" icon="book" href="/get-started/core-concepts/mcp-plugins">
          The full reference for what you can do programmatically.
        </Card>
      </CardGroup>
    </Tab>
  </Tabs>
</div>

## The four levels in marketing

How the [autonomy ladder](/docs/guides) plays out across your funnel:

<CardGroup>
  <Card title="L1 — Assisted" icon="hand">
    A marketer asks the Agent for three subject-line options for tomorrow's drop — the Agent returns options with reasoning.
  </Card>

  <Card title="L2 — Copilot" icon="map">
    A marketer asks for a 5-step onboarding sequence for new free-trial users — the Agent drafts all 5, the marketer reviews and queues.
  </Card>

  <Card title="L3 — Autopilot" icon="bolt">
    A new MQL lands in HubSpot — segmentation, nurture-sequence drafting, and Slack alerts to the AE all happen automatically.
  </Card>

  <Card title="L4 — Self-driving" icon="infinity">
    You set the objective — "increase trial-to-paid conversion" — and the Workforce runs onboarding sequences, with subject lines, send times, and CTAs tuned through evals as conversion data flows back.
  </Card>
</CardGroup>

## Zoom into a single use case

Each use case below is a working entry point — clone a template, then run it, shape it visually, or build with AI depending on which tab fit you.

<CardGroup>
  <Card title="Lifecycle email campaigns" icon="envelope" href="/guides/marketing/lifecycle-campaigns">
    Draft nurture, re-engagement, and onboarding sequences in your brand voice at the moments that matter.
  </Card>

  <Card title="Content creation & repurposing" icon="copy" href="/guides/marketing/content-repurposing">
    Turn one asset into blog posts, social snippets, ad copy — without losing the source voice.
  </Card>

  <Card title="Campaign performance analytics" icon="chart-line" href="/guides/marketing/campaign-analytics">
    Pull cross-channel reports together; surface what's working and what isn't.
  </Card>
</CardGroup>

## See it in production

<Frame>
  <iframe title="Autodesk x Relevance AI — Agents@Work" />
</Frame>

<Frame>
  <iframe title="Lightspeed x Relevance AI — Agents@Work" />
</Frame>


# Lifecycle email campaigns
Source: https://relevanceai.com/docs/guides/marketing/lifecycle-campaigns

Draft nurture, re-engagement, and onboarding sequences in your brand voice at the moments that matter.

Lifecycle email is the workhorse of B2B marketing — and the place that scales worst with headcount. A lifecycle Agent drafts the sequences that match each segment, lifecycle stage, and trigger so you can run more campaigns without copying templates into Google Docs.

## When this pays off

<CardGroup>
  <Card title="Lifecycle is templated" icon="files">
    You're copy-pasting "Welcome Series v3" docs between every product launch instead of running per-launch sequences.
  </Card>

  <Card title="Trigger moments are missed" icon="bell-slash">
    Trial ending, contract renewing, feature unused — these moments deserve sequences and currently get nothing.
  </Card>

  <Card title="Brand voice drift" icon="comments">
    Different team members write each campaign and the voice has slipped across them.
  </Card>

  <Card title="Re-engagement is overdue" icon="hourglass">
    A dormant-list re-engagement campaign keeps getting deprioritized because the copy work is the bottleneck.
  </Card>
</CardGroup>

## The shape of this use case

A lifecycle Agent takes a segment + trigger and returns a multi-step email sequence in your voice.

<CardGroup>
  <Card title="Inputs" icon="arrow-right-to-bracket">
    Segment definition, lifecycle stage, triggering event, campaign goal.
  </Card>

  <Card title="Sources" icon="globe">
    Brand voice docs, past-campaign copy, persona definitions, product positioning, your marketing automation platform.
  </Card>

  <Card title="Output" icon="file-lines">
    A drafted sequence — subject lines, body copy, send timing, CTA per step — ready for review.
  </Card>

  <Card title="Delivery" icon="paper-plane">
    Drafted into [HubSpot](/docs/integrations/popular-integrations/hubspot) or [Marketo](/docs/integrations/popular-integrations/marketo) for review, written to a doc for marketing approval, queued in your ESP.
  </Card>
</CardGroup>

## Where to start

Two ways in, depending on whether you want something running today or built to your exact spec.

<CardGroup>
  <Card title="Clone a pre-built Agent" icon="copy">
    Open the **[Email Marketing Strategist](https://marketplace.relevanceai.com/listing/28783038-03f0-4aec-9c76-4b28452b6e33)**. More in the [Marketplace](/docs/get-started/marketplace/introduction).
  </Card>

  <Card title="Build your own" icon="hammer">
    Start from scratch in the [builder](/docs/build/introduction), or by describing it in Claude Code or Cursor with [MCP & Plugins](/docs/get-started/core-concepts/mcp-plugins).
  </Card>
</CardGroup>

Either way, these are prompts your team can use on day one:

* *"Draft a 4-step welcome series for new free-trial users in HubSpot — friendly tone, focus on activating the dashboard feature."*
* *"Write a re-engagement sequence for trial users who haven't logged in for 30 days."*
* *"Give me a 3-touch renewal sequence for accounts in the 90-day renewal window."*

## Where to take it

Once it's running, deepen it in three moves:

<CardGroup>
  <Card title="Give it a playbook" icon="book">
    Shape it with a [prompt](/docs/build/agents/build-your-agent/prompt), your brand voice in [Knowledge](/docs/build/knowledge/create-knowledge), and [Bulk Schedule](/docs/build/agents/give-your-agent-tasks/bulk-schedule).
  </Card>

  <Card title="Automate it on signals" icon="bolt">
    Wrap it in a [workflow](/docs/build/workforces/create-a-workforce) that fires on a [trigger](/docs/build/agents/build-your-agent/triggers).
  </Card>

  <Card title="Let it improve" icon="arrows-rotate">
    Feed opens, clicks, and conversion back into the Agent's [evals](/docs/build/agents/build-your-agent/evals) so drafts track what converts.
  </Card>
</CardGroup>

## Common pitfalls

<AccordionGroup>
  <Accordion title="Sequences that all sound the same" icon="clone">
    Without persona-specific Knowledge, every sequence reads like a watered-down version of the welcome email. Upload distinct persona docs and force the prompt to reference them.
  </Accordion>

  <Accordion title="Trigger drift" icon="route">
    The HubSpot trigger fires on the wrong field and the sequence drafts the wrong campaign. Gate triggers tightly and review what actually causes them to fire in week one.
  </Accordion>

  <Accordion title="Brand-voice regressions" icon="clock-rotate-left">
    A prompt tweak changes tone everywhere. Use [evals](/docs/build/agents/build-your-agent/evals) on a few canonical drafts to catch drift before it ships to subscribers.
  </Accordion>

  <Accordion title="Auto-sending without review" icon="paper-plane">
    Tempting to skip marketing approval and queue directly. Don't — one bad subject line in production costs trust. Gate L3 with a Slack approval until you've watched it run for a quarter.
  </Accordion>
</AccordionGroup>


# Data dedup and cleanup
Source: https://relevanceai.com/docs/guides/revops/data-dedup

Find and merge duplicate records, fix inconsistent fields, surface the records that need a human.

CRMs accrete duplicate records the way oceans accrete plastic. The exact-match dedup is easy; the fuzzy cases — "Acme Corp" vs "Acme Corporation" vs "Acme Inc" — eat hours of RevOps time. A dedup Agent does the fuzzy matching with reasoning, merges what's safe, and surfaces the ambiguous cases for a human to decide.

## When this pays off

<CardGroup>
  <Card title="Duplicate accounts everywhere" icon="copy">
    Sales operations reports keep getting flagged because the same company exists under three names.
  </Card>

  <Card title="Manual dedup is hours" icon="hourglass">
    Each quarter, someone spends a day or two going through dedup candidates and merging by hand.
  </Card>

  <Card title="Conflicting fields after merge" icon="scale-unbalanced">
    Even when records get merged, field-level conflicts (which address? which industry?) get resolved arbitrarily.
  </Card>

  <Card title="Bad merge erodes trust" icon="triangle-exclamation">
    A wrong merge wipes deal history; teams stop trusting the dedup process and the queue grows.
  </Card>
</CardGroup>

## The shape of this use case

A dedup Agent takes a candidate pair (or a list of candidates) and returns a merge decision with reasoning.

<CardGroup>
  <Card title="Inputs" icon="arrow-right-to-bracket">
    Candidate records, full field-level data, related activity / deal history.
  </Card>

  <Card title="Sources" icon="globe">
    CRM, your dedup heuristics / playbook, parent-child hierarchy, third-party data for ground-truth checks.
  </Card>

  <Card title="Output" icon="file-lines">
    A decision (merge / keep separate / human review) with confidence, plus field-level conflict resolution recommendations.
  </Card>

  <Card title="Delivery" icon="paper-plane">
    High-confidence merges applied directly (with audit log); ambiguous cases queued in a review tool or posted to RevOps [Slack](/docs/integrations/popular-integrations/slack).
  </Card>
</CardGroup>

## Where to start

Two ways in, depending on whether you want something running today or built to your exact spec.

<CardGroup>
  <Card title="Clone a pre-built Agent" icon="copy">
    Open the **[CRM Agent](https://marketplace.relevanceai.com/listing/f3e18700-27e2-477d-8eaa-4c6fa04282bf)**. More in the [Marketplace](/docs/get-started/marketplace/introduction).
  </Card>

  <Card title="Build your own" icon="hammer">
    Start from scratch in the [builder](/docs/build/introduction), or by describing it in Claude Code or Cursor with [MCP & Plugins](/docs/get-started/core-concepts/mcp-plugins).
  </Card>
</CardGroup>

Either way, these are prompts your team can use on day one:

* *"These two HubSpot accounts both look like Acme Corp — are they the same company? Compare addresses, websites, contact overlap."*
* *"Run dedup on the leads imported from this trade-show CSV — flag the high-confidence matches and the ones I should look at."*
* *"Did we already have a record for this contact? Check across CRM and prior tickets."*

## Where to take it

Once it's running, deepen it in three moves:

<CardGroup>
  <Card title="Give it a playbook" icon="book">
    Shape it with a [prompt](/docs/build/agents/build-your-agent/prompt), your matching rules in [Knowledge](/docs/build/knowledge/create-knowledge), and [Bulk Schedule](/docs/build/agents/give-your-agent-tasks/bulk-schedule).
  </Card>

  <Card title="Automate it on signals" icon="bolt">
    Wrap it in a [workflow](/docs/build/workforces/create-a-workforce) that fires on a [trigger](/docs/build/agents/build-your-agent/triggers).
  </Card>

  <Card title="Let it improve" icon="arrows-rotate">
    Feed merge undos and disputes back into the Agent's [evals](/docs/build/agents/build-your-agent/evals) so its confidence thresholds track what's reliable.
  </Card>
</CardGroup>

## Common pitfalls

<AccordionGroup>
  <Accordion title="Auto-merging without audit" icon="clock-rotate-left">
    A wrong merge loses deal history. Always log what got merged and offer one-click undo for the first quarter of running.
  </Accordion>

  <Accordion title="Confidence thresholds set too loose" icon="gauge-high">
    Aggressive auto-merge merges records that should have stayed separate. Start with high thresholds and loosen as you watch outcomes.
  </Accordion>

  <Accordion title="Ignoring field-level conflicts" icon="scale-unbalanced">
    Merging accounts without a rule for which address / industry / size wins produces garbled records. Document field priority rules in Knowledge.
  </Accordion>

  <Accordion title="No relationship to ongoing deals" icon="link-slash">
    Merging an account with an active opportunity into one without can break attribution and rep ownership. Have the Agent factor open-deal status into the decision.
  </Accordion>
</AccordionGroup>


# RevOps
Source: https://relevanceai.com/docs/guides/revops/getting-started

Put AI Agents to work across revenue operations — pipeline hygiene, routing, forecasting — so the GTM machine actually runs on clean data.

RevOps owns the systems that everyone else depends on. Data hygiene, routing rules, pipeline reviews, forecast accuracy — all of it gets done manually because the rules are too edge-casey to hard-code. AI Agents handle the judgment-y middle ground: clear enough to automate, fuzzy enough that hard rules break.

## Start with a Marketplace template

The Marketplace ships pre-built Agents you can clone in one click. Adapt them to your CRM, your segmentation, your pipeline definitions. Or [build a custom Agent from scratch](/docs/build/introduction).

<CardGroup>
  <Card title="Elli, the Enrichment Agent" icon="database" href="https://marketplace.relevanceai.com/listing/88ab63e7-f357-48ad-84c6-463be16d850b">
    Fills in firmographics, contact details, and job-function classification on every record — the foundation for everything else.
  </Card>

  <Card title="Lia, the LinkedIn Lead Researcher & Qualifier" icon="user-magnifying-glass" href="https://marketplace.relevanceai.com/listing/ba230945-e61a-4fb0-8c47-7e240f18456d">
    Takes a LinkedIn profile or CRM record, researches the company, and returns a qualification read you can wire into routing.
  </Card>

  <Card title="Browse the Marketplace" icon="store" href="/get-started/marketplace/introduction">
    The full catalog of Agents — enrichment, routing, forecasting, data hygiene.
  </Card>
</CardGroup>

Click **Clone Agent** on any listing, connect your CRM and data sources, and run on real records.

## Pick your path

How hands-on do you want to be? Pick a tab — the rest of the page is written for you.

<div>
  <Tabs>
    <Tab title="No build">
      You don't need to touch the builder. Open the cloned Agent above, ask in plain English, and let it work across the systems it's connected to.

      ## Sharpen the answers with knowledge

      Upload your ICP definitions, segmentation rules, routing logic, and pipeline definitions. Every conversation references them — sharper output, no new setup.

      <Card title="Add knowledge" icon="book" href="/build/knowledge/create-knowledge">
        Upload context once. Every conversation references it.
      </Card>

      ## Make it run on its own

      When you're answering the same data-quality or routing question every week, set a trigger — every new account auto-enriches, every stale record auto-flags. One toggle, no code, no builder.

      <Card title="Triggers" icon="bolt" href="/build/agents/build-your-agent/triggers">
        Schedule runs, fire on CRM events, listen to webhooks.
      </Card>

      ## Ready to go further?

      When prompting and templates stop being enough — when you want the Agent to enforce your routing rules across the whole pipeline and orchestrate cross-system workflows — pick the **Build visually** tab above.
    </Tab>

    <Tab title="Build visually">
      You're comfortable in product UIs. You don't want to wire up an Agent from scratch, but you do want it to enforce your routing logic record by record. Open the cloned Agent in the builder and shape every piece.

      ## Connect the tools it needs

      RevOps Agents work best when they can read from and write to the systems your team already uses.

      * **CRM** — [HubSpot](/docs/integrations/popular-integrations/hubspot) or [Salesforce](/docs/integrations/popular-integrations/salesforce) — the system of record
      * **Data sources** — [Apollo](/docs/integrations/popular-integrations/apollo), [ZoomInfo](/docs/integrations/popular-integrations/zoominfo), web search for enrichment
      * **BI / analytics** — [Google Sheets](/docs/integrations/popular-integrations/google-sheets) or your warehouse for pipeline reporting
      * **Comms** — [Slack](/docs/integrations/popular-integrations/slack), [Gmail](/docs/integrations/popular-integrations/gmail) for alerts and ops notifications

      See more about [Integrations](/docs/integrations/introduction).

      ## Shape every surface

      Each piece below changes how the Agent behaves. Run it on a small record set between every change — small iterations beat big rewrites.

      <CardGroup>
        <Card title="Prompt" icon="message" href="/build/agents/build-your-agent/prompt">
          Edit the system prompt to match your ICP, segmentation, routing rules, and pipeline definitions.
        </Card>

        <Card title="Tools" icon="wrench" href="/build/agents/build-your-agent/tools">
          Add or remove the actions the Agent can take — query CRM, enrich from third-party data, write back fields, post to Slack.
        </Card>

        <Card title="Knowledge" icon="book" href="/build/knowledge/create-knowledge">
          Upload ICP, territory definitions, routing logic, pipeline schemas. The Agent draws on these whenever it works.
        </Card>

        <Card title="Triggers" icon="bolt" href="/build/agents/build-your-agent/triggers">
          Run the Agent automatically — on record create, on stage change, on a freshness schedule.
        </Card>

        <Card title="Alerts" icon="bell" href="/build/agents/build-your-agent/alerts">
          Get notified when routing rules conflict, when forecast accuracy drifts, when dedup confidence is low.
        </Card>

        <Card title="Memory" icon="brain" href="/build/agents/build-your-agent/memory">
          Persistent context per record — past routing decisions, prior enrichment sources, change history.
        </Card>
      </CardGroup>

      ## Chain Agents into a workforce

      When a single Agent isn't enough — enrich, then score, then route, then notify — wire them together on the Workforce canvas.

      <Card title="Build a workforce" icon="diagram-project" href="/build/workforces/create-a-workforce">
        Drag, connect, and run multi-Agent pipelines without code.
      </Card>

      ## Test before you trust

      Set up evals so routing or enrichment regressions get caught before they propagate across thousands of records.

      <Card title="Evals" icon="vial" href="/build/agents/build-your-agent/evals">
        Define test cases in the UI. Run them on every change.
      </Card>

      ## Ready to go further?

      When clicking through the builder starts to feel slow — when you'd rather have your AI coding assistant build and edit Agents for you, drop into custom Python or JS code blocks, or call Agents from your own app via the SDK — pick the **Build with AI** tab above.
    </Tab>

    <Tab title="Build with AI">
      You already work in an AI coding environment. Clicking through a UI to edit routing rules feels slow. Have your AI assistant build, shape, and orchestrate Agents directly from your editor.

      ## MCP & Plugins

      Connect your AI coding environment to Relevance AI over MCP. Describe what you want in natural language; your AI assistant reads and writes your Relevance workspace — system prompts, Tools, knowledge, triggers, workforces, evals.

      <CardGroup>
        <Card title="Claude Code plugin" icon="terminal" href="/integrations/mcp/claude-code">
          The fastest setup. Install the Relevance AI plugin and build Agents from your terminal.
        </Card>

        <Card title="MCP server" icon="plug" href="/integrations/mcp/mcp-server">
          Connect from Claude Desktop, Cursor, VS Code, ChatGPT, OpenAI Codex, or any MCP client.
        </Card>

        <Card title="Agent skills" icon="graduation-cap" href="/integrations/mcp/agent-skills">
          Clone the skills repo so your coding assistant has built-in knowledge of Relevance patterns.
        </Card>

        <Card title="MCP & Plugins overview" icon="book" href="/get-started/core-concepts/mcp-plugins">
          The full reference for what you can do programmatically.
        </Card>
      </CardGroup>
    </Tab>
  </Tabs>
</div>

## The four levels in revenue operations

How the [autonomy ladder](/docs/guides) plays out across your pipeline:

<CardGroup>
  <Card title="L1 — Assisted" icon="hand">
    A RevOps analyst asks the Agent which accounts in the new MQL batch look like a fit for the enterprise team — the Agent returns a list with reasoning.
  </Card>

  <Card title="L2 — Copilot" icon="map">
    A RevOps analyst asks the Agent to dedupe a list of 1,200 imported leads — the Agent flags clusters with merge recommendations, the analyst approves the merges.
  </Card>

  <Card title="L3 — Autopilot" icon="bolt">
    A new lead enters the CRM → enriched, scored, routed to the right rep, posted to Slack, all without anyone touching it.
  </Card>

  <Card title="L4 — Self-driving" icon="infinity">
    You set the objective — "improve forecast accuracy" — and the Workforce enriches, scores, and routes, with its deal-signal criteria tuned through evals as close-vs-slip outcomes flow back.
  </Card>
</CardGroup>

## Zoom into a single use case

Each use case below is a working entry point — clone a template, then run it, shape it visually, or build with AI depending on which tab fit you.

<CardGroup>
  <Card title="Pipeline forecasting & hygiene" icon="chart-line" href="/guides/revops/pipeline-hygiene">
    Spot stalled deals, missing fields, and at-risk forecast slippage before pipeline review.
  </Card>

  <Card title="Lead routing & territory" icon="route" href="/guides/revops/lead-routing">
    Apply your routing rules — plus the judgment edge cases — to every new lead automatically.
  </Card>

  <Card title="Data dedup & cleanup" icon="layer-group" href="/guides/revops/data-dedup">
    Find and merge duplicates, fix inconsistent fields, surface the records that need a human.
  </Card>
</CardGroup>

For account-level enrichment, see the [CRM data enrichment](/docs/guides/sales/crm-data-enrichment) use case in Sales — same shape, applies cleanly to RevOps use. For scoring inbound, see [Lead scoring](/docs/guides/sales/lead-scoring).

## See it in production

<Frame>
  <iframe title="KPMG x Relevance AI — Agents@Work" />
</Frame>


# Lead routing and territory assignment
Source: https://relevanceai.com/docs/guides/revops/lead-routing

Apply your routing rules — plus the judgment edge cases — to every new lead automatically.

Routing rules in Salesforce or HubSpot handle 80% of inbound leads. The other 20% are the edge cases that eat hours — accounts spanning multiple territories, ambiguous industries, parent-child company tangles, region overrides. A routing Agent reads each lead, applies your rules where they're clear, and handles the judgment where they aren't.

## When this pays off

<CardGroup>
  <Card title="Routing rules are too brittle" icon="route">
    Existing logic handles the easy 80%; the rest end up in a queue waiting for RevOps to manually route.
  </Card>

  <Card title="Region overlaps cause disputes" icon="map">
    Reps argue over who owns a lead because firmographic data disagrees about region or segment.
  </Card>

  <Card title="Routing latency" icon="hourglass">
    Hours pass between lead landing and lead getting picked up because manual routing is the bottleneck.
  </Card>

  <Card title="Parent-child confusion" icon="sitemap">
    A new contact at a subsidiary lands without recognizing the parent already has an open opportunity.
  </Card>
</CardGroup>

## The shape of this use case

A routing Agent takes a new lead and returns an assignment decision with reasoning.

<CardGroup>
  <Card title="Inputs" icon="arrow-right-to-bracket">
    Lead record, account context, territory boundaries, current rep workload.
  </Card>

  <Card title="Sources" icon="globe">
    CRM, territory / segmentation rules, rep capacity data, parent-child account hierarchy, your routing playbook.
  </Card>

  <Card title="Output" icon="file-lines">
    A routing decision (rep / team, reasoning, confidence), with flags for edge cases that need human review.
  </Card>

  <Card title="Delivery" icon="paper-plane">
    Applied directly in the CRM as owner, posted to [Slack](/docs/integrations/popular-integrations/slack) on edge cases for RevOps to confirm, logged with reasoning for audit.
  </Card>
</CardGroup>

## Where to start

Two ways in, depending on whether you want something running today or built to your exact spec.

<CardGroup>
  <Card title="Clone a pre-built Agent" icon="copy">
    Open the **[CRM Agent](https://marketplace.relevanceai.com/listing/f3e18700-27e2-477d-8eaa-4c6fa04282bf)**. More in the [Marketplace](/docs/get-started/marketplace/introduction).
  </Card>

  <Card title="Build your own" icon="hammer">
    Start from scratch in the [builder](/docs/build/introduction), or by describing it in Claude Code or Cursor with [MCP & Plugins](/docs/get-started/core-concepts/mcp-plugins).
  </Card>
</CardGroup>

Either way, these are prompts your team can use on day one:

* *"Where should this lead go? Acme Corp, contact at the German subsidiary, parent account owned by the US enterprise team."*
* *"This batch of 50 leads imported from a trade show — apply our routing rules and tell me which ones need a human decision."*
* *"Is this lead a re-engagement of an existing opportunity or a new prospect? Check parent-child and prior history."*

## Where to take it

Once it's running, deepen it in three moves:

<CardGroup>
  <Card title="Give it a playbook" icon="book">
    Shape it with a [prompt](/docs/build/agents/build-your-agent/prompt), your territory rules in [Knowledge](/docs/build/knowledge/create-knowledge), and [Bulk Schedule](/docs/build/agents/give-your-agent-tasks/bulk-schedule).
  </Card>

  <Card title="Automate it on signals" icon="bolt">
    Wrap it in a [workflow](/docs/build/workforces/create-a-workforce) that fires on a [trigger](/docs/build/agents/build-your-agent/triggers).
  </Card>

  <Card title="Let it improve" icon="arrows-rotate">
    Feed back which routing held up into the Agent's [evals](/docs/build/agents/build-your-agent/evals) so edge-case handling tracks real outcomes.
  </Card>
</CardGroup>

## Common pitfalls

<AccordionGroup>
  <Accordion title="Overconfident decisions on ambiguous leads" icon="bullseye">
    Force the Agent to flag low-confidence decisions for human review rather than making a guess. Wrong assignments are expensive to undo.
  </Accordion>

  <Accordion title="Missing parent-child relationships" icon="sitemap">
    Without hierarchy data in Knowledge, the Agent treats subsidiaries as net-new accounts. Document the parent-child structure explicitly.
  </Accordion>

  <Accordion title="Ignoring rep capacity" icon="users">
    Routing on rules alone overloads top reps. Have the Agent factor current pipeline volume into the routing decision.
  </Accordion>

  <Accordion title="No audit trail" icon="list-check">
    Reps dispute routing decisions. Log every Agent decision with its reasoning so you can review later.
  </Accordion>
</AccordionGroup>


# Pipeline forecasting and hygiene
Source: https://relevanceai.com/docs/guides/revops/pipeline-hygiene

Spot stalled deals, missing fields, and at-risk forecast slippage before pipeline review.

The Monday pipeline review depends on data that's only as fresh as the rep updated it on Friday — which is to say, not very. A pipeline Agent reads the whole pipeline state, flags what's stalled, what's missing, what's likely to slip, and surfaces the gaps so the review actually starts with reality.

## When this pays off

<CardGroup>
  <Card title="Stale pipeline" icon="hourglass">
    Half the deals in pipeline haven't been touched in 30 days; reps update only when forced.
  </Card>

  <Card title="Forecast accuracy poor" icon="chart-line-down">
    Commit vs. close-rate gap grows quarter over quarter — leadership stops trusting the forecast.
  </Card>

  <Card title="Pipeline review is data hunting" icon="binoculars">
    The first 20 minutes of every pipeline review is reps explaining missing data.
  </Card>

  <Card title="Slipping deals caught late" icon="alarm-clock">
    Deals slip the week before the quarter ends — when it's too late to do anything.
  </Card>
</CardGroup>

## The shape of this use case

A pipeline Agent takes a pipeline snapshot and returns a structured analysis with flags.

<CardGroup>
  <Card title="Inputs" icon="arrow-right-to-bracket">
    Pipeline state (deals, stages, dates, amounts), historical close-rate benchmarks, your forecast definitions.
  </Card>

  <Card title="Sources" icon="globe">
    CRM, prior quarter close data, deal activity logs, your sales methodology / MEDDICC framework.
  </Card>

  <Card title="Output" icon="file-lines">
    A pipeline brief — stalled deals, missing critical fields, slippage risk per deal, forecast vs. commit gap with reasoning.
  </Card>

  <Card title="Delivery" icon="paper-plane">
    Posted to [Slack](/docs/integrations/popular-integrations/slack) ahead of pipeline review, written to a [Notion](/docs/integrations/popular-integrations/notion) ops doc, surfaced as deal-level alerts to AEs.
  </Card>
</CardGroup>

## Where to start

Two ways in, depending on whether you want something running today or built to your exact spec.

<CardGroup>
  <Card title="Clone a pre-built Agent" icon="copy">
    Open the **[Deal Health Analyst](https://marketplace.relevanceai.com/listing/5ecd3c3f-73ba-4f08-877c-bb68dde05387)**. More in the [Marketplace](/docs/get-started/marketplace/introduction).
  </Card>

  <Card title="Build your own" icon="hammer">
    Start from scratch in the [builder](/docs/build/introduction), or by describing it in Claude Code or Cursor with [MCP & Plugins](/docs/get-started/core-concepts/mcp-plugins).
  </Card>
</CardGroup>

Either way, these are prompts your team can use on day one:

* *"What deals look at-risk for Q3? Pull stalled deals, missing close dates, and weak MEDDICC scores."*
* *"Compare this week's commit vs. last quarter's commit-vs-close at the same point in the cycle — are we tracking?"*
* *"Which 10 deals should the team focus on this week to hit number?"*

## Where to take it

Once it's running, deepen it in three moves:

<CardGroup>
  <Card title="Give it a playbook" icon="book">
    Shape it with a [prompt](/docs/build/agents/build-your-agent/prompt), your methodology in [Knowledge](/docs/build/knowledge/create-knowledge), and [Tools](/docs/build/agents/build-your-agent/tools) to query the CRM.
  </Card>

  <Card title="Automate it on signals" icon="bolt">
    Wrap it in a [workflow](/docs/build/workforces/create-a-workforce) that fires on a [trigger](/docs/build/agents/build-your-agent/triggers).
  </Card>

  <Card title="Let it improve" icon="arrows-rotate">
    Feed back which at-risk signals predicted slippage into the Agent's [evals](/docs/build/agents/build-your-agent/evals) so flagging sharpens.
  </Card>
</CardGroup>

## Common pitfalls

<AccordionGroup>
  <Accordion title="Noise overwhelms signal" icon="bell-slash">
    If the Agent flags every deal as "at risk", the team stops reading. Tighten criteria to what actually predicts slippage.
  </Accordion>

  <Accordion title="No outcome feedback" icon="arrows-rotate">
    Without piping won/lost / slipped outcomes back to the Agent's evals, the criteria can't improve. Pipe deal outcomes to evaluations.
  </Accordion>

  <Accordion title="Missing the qualitative" icon="comment">
    Pure field-based analysis misses what AEs hear on calls. Have the Agent read recent activity notes and pick up sentiment signals.
  </Accordion>

  <Accordion title="Auto-changing CRM fields" icon="pen">
    Letting the Agent overwrite stage or amount fields without rep review breaks trust and reporting. Surface recommendations; the rep changes the field.
  </Accordion>
</AccordionGroup>


# Competitive intelligence
Source: https://relevanceai.com/docs/guides/sales/competitive-intelligence

Track competitor moves and keep your team's positioning sharp.

Deals get lost when reps can't articulate how you compare. A competitive intelligence Agent monitors what competitors are doing and keeps your team's battle cards current — so reps walk into positioning conversations ready, not improvising.

## When this pays off

<CardGroup>
  <Card title="Competitors shipping fast" icon="rocket">
    The market is moving faster than your battle cards can keep up.
  </Card>

  <Card title="Out-of-date positioning" icon="calendar-xmark">
    Battle cards exist but were last updated two quarters ago and nobody trusts them.
  </Card>

  <Card title="Losing winnable deals" icon="chart-line-down">
    Win/loss reviews keep flagging the same competitor as the reason — and the same gaps.
  </Card>

  <Card title="Weekly digests requested" icon="newspaper">
    Leadership wants a regular pulse on competitive activity for the GTM team.
  </Card>
</CardGroup>

## The shape of this use case

A competitive intelligence Agent monitors a defined competitor set and surfaces what matters.

<CardGroup>
  <Card title="Inputs" icon="arrow-right-to-bracket">
    Competitor list, products to compare, market segments you operate in.
  </Card>

  <Card title="Sources" icon="globe">
    Competitor websites, news, job boards, social media, your own win/loss notes.
  </Card>

  <Card title="Output" icon="file-lines">
    Updated battle cards, product comparison summaries, weekly digests.
  </Card>

  <Card title="Delivery" icon="paper-plane">
    Posted to [Slack](/docs/integrations/popular-integrations/slack), written to your wiki or [Notion](/docs/integrations/popular-integrations/notion), attached to deal records when a competitor shows up.
  </Card>
</CardGroup>

## Where to start

Two ways in, depending on whether you want something running today or built to your exact spec.

<CardGroup>
  <Card title="Clone a pre-built Agent" icon="copy">
    Open the **[Sales Battlecard Creator](https://marketplace.relevanceai.com/listing/ca29a93b-cca6-4238-95d0-84159c692241)**. More in the [Marketplace](/docs/get-started/marketplace/introduction).
  </Card>

  <Card title="Build your own" icon="hammer">
    Start from scratch in the [builder](/docs/build/introduction), or by describing it in Claude Code or Cursor with [MCP & Plugins](/docs/get-started/core-concepts/mcp-plugins).
  </Card>
</CardGroup>

Either way, these are prompts your reps can use on day one:

* *"What did Acme Corp announce last quarter? They came up on a call this morning."*
* *"Pull a quick comparison of our pricing vs. Globex for an exec on this deal."*
* *"Summarize where we stand against Initech for a renewal conversation tomorrow."*

## Where to take it

Once it's running, deepen it in three moves:

<CardGroup>
  <Card title="Give it a playbook" icon="book">
    Shape it with a [prompt](/docs/build/agents/build-your-agent/prompt), [Knowledge](/docs/build/knowledge/create-knowledge), and [Firecrawl](/docs/build/tools/tool-steps/firecrawl) for competitor sites.
  </Card>

  <Card title="Automate it on signals" icon="bolt">
    Wrap it in a [workflow](/docs/build/workforces/create-a-workforce) that fires on a [trigger](/docs/build/agents/build-your-agent/triggers).
  </Card>

  <Card title="Let it improve" icon="arrows-rotate">
    Feed win/loss notes back into the Agent's [evals](/docs/build/agents/build-your-agent/evals) so it tracks the signals that matter.
  </Card>
</CardGroup>

## Common pitfalls

<AccordionGroup>
  <Accordion title="Tracking too many competitors" icon="bell-slash">
    Twenty competitors equals alert fatigue and nothing gets read. Limit to the 5–10 that actually surface in deals.
  </Accordion>

  <Accordion title="Battle cards that read like marketing copy" icon="comment">
    Reps can't use a battle card mid-call if it's a webpage. Write them as short objection-handling scripts the rep can read in one breath.
  </Accordion>

  <Accordion title="Intel that never reaches a deal" icon="route">
    A weekly digest is interesting. Resurfacing the right battle card the moment a competitor shows up in a deal is useful. Wire the Agent to deal records, not just channels.
  </Accordion>

  <Accordion title="Sources that are too easy" icon="magnifying-glass">
    Press releases only show you what competitors want you to see. Add hiring patterns, job postings, and customer-side social signals to catch real direction.
  </Accordion>
</AccordionGroup>


# CRM data enrichment
Source: https://relevanceai.com/docs/guides/sales/crm-data-enrichment

Fill in missing CRM fields automatically so reps stop wasting time looking up basics.

Incomplete CRM data slows everything down — reps waste time looking up basics, lead routing breaks, and reporting becomes unreliable. An enrichment Agent fills in the gaps automatically so your data stays clean without manual effort.

## When this pays off

<CardGroup>
  <Card title="Routing rules can't fire" icon="route">
    Lead-routing logic depends on fields that are blank for half the records.
  </Card>

  <Card title="Unreliable reporting" icon="chart-pie">
    Industry and company-size data is inconsistent, so segmentation reports are useless.
  </Card>

  <Card title="LinkedIn URL as research" icon="link">
    Reps are pasting LinkedIn profile links into the CRM and calling it enrichment.
  </Card>

  <Card title="New ICP rollout" icon="rotate">
    You're redefining your ICP and need to re-evaluate every existing record against the new lens.
  </Card>
</CardGroup>

## The shape of this use case

An enrichment Agent takes a partial CRM record and fills in missing fields.

<CardGroup>
  <Card title="Inputs" icon="arrow-right-to-bracket">
    Existing CRM record (contact or account) with partial data.
  </Card>

  <Card title="Sources" icon="globe">
    Web search, [LinkedIn](/docs/integrations/popular-integrations/linkedin), third-party data vendors, your own CRM history.
  </Card>

  <Card title="Output" icon="file-lines">
    Filled-in fields — firmographics, contact details, tech stack, social profiles — with source attribution.
  </Card>

  <Card title="Delivery" icon="paper-plane">
    Written back to the CRM in the correct fields, with sourcing notes attached.
  </Card>
</CardGroup>

## Where to start

Two ways in, depending on whether you want something running today or built to your exact spec.

<CardGroup>
  <Card title="Clone a pre-built Agent" icon="copy">
    Open **[Elli, the Enrichment Agent](https://marketplace.relevanceai.com/listing/88ab63e7-f357-48ad-84c6-463be16d850b)**. More in the [Marketplace](/docs/get-started/marketplace/introduction).
  </Card>

  <Card title="Build your own" icon="hammer">
    Start from scratch in the [builder](/docs/build/introduction), or by describing it in Claude Code or Cursor with [MCP & Plugins](/docs/get-started/core-concepts/mcp-plugins).
  </Card>
</CardGroup>

Either way, these are prompts your reps can use on day one:

* *"Enrich Acme Corp — fill in industry, employee count, and any recent news."*
* *"Find the head of RevOps at Initech and add them as a contact in HubSpot."*
* *"What tech stack is Globex running? Web search and check LinkedIn."*

## Where to take it

Once it's running, deepen it in three moves:

<CardGroup>
  <Card title="Give it a playbook" icon="book">
    Shape it with a [prompt](/docs/build/agents/build-your-agent/prompt), a connected [CRM](/docs/integrations/popular-integrations/hubspot) for read/write, and rules in [Knowledge](/docs/build/knowledge/create-knowledge).
  </Card>

  <Card title="Automate it on signals" icon="bolt">
    Wrap it in a [workflow](/docs/build/workforces/create-a-workforce) that fires on a [trigger](/docs/build/agents/build-your-agent/triggers).
  </Card>

  <Card title="Let it improve" icon="arrows-rotate">
    Tune it through [evals](/docs/build/agents/build-your-agent/evals) so it maintains the fields your team actually uses.
  </Card>
</CardGroup>

## Common pitfalls

<AccordionGroup>
  <Accordion title="Enriching every record on every event" icon="coins">
    Credit-burn at scale. Gate enrichment on lifecycle stage and a freshness window — don't re-enrich records that were updated last week.
  </Accordion>

  <Accordion title="Overwriting human-entered data" icon="user-shield">
    A rep's hand-typed value usually beats whatever the web returns. Preserve human overrides, or write to a parallel "enriched\_\*" field that reports can choose to use.
  </Accordion>

  <Accordion title="No source attribution" icon="link-slash">
    Reps can't trust a value if they can't see where it came from. Have the Agent attach a source URL and timestamp to every enriched field.
  </Accordion>

  <Accordion title="Stale data without a freshness check" icon="clock-rotate-left">
    Pulling 2019 data into a 2026 record looks like enrichment but isn't. Filter sources by recency where it matters — leadership, headcount, funding.
  </Accordion>
</AccordionGroup>


# Sales
Source: https://relevanceai.com/docs/guides/sales/getting-started

Put AI Agents to work across your sales pipeline. Start wherever you are — Relevance grows with you.

Hand the research, prep, and outreach grind to AI. Spend your time on the conversations that close deals.

## Start with a Marketplace template

The Marketplace ships pre-built sales Agents you can clone in one click. They're the starting point for every path below — run them as-is, shape them in the builder, or hand them to your AI coding assistant. Or skip the templates and [build a custom Agent from scratch](/docs/build/introduction).

<CardGroup>
  <Card title="Sales Researcher" icon="magnifying-glass" href="https://marketplace.relevanceai.com/listing/4b454347-0656-4fee-9c15-8fb2f2713424">
    Builds detailed call-prep briefs covering company, contacts, news, and pain points.
  </Card>

  <Card title="LinkedIn Outreach & Follow up" icon="linkedin" href="https://marketplace.relevanceai.com/listing/5508c7f6-f3a7-4593-8faf-22fbecc4e0d4">
    Writes personalized connection requests and follow-ups based on prospect research.
  </Card>

  <Card title="Browse the Marketplace" icon="store" href="/get-started/marketplace/introduction">
    The full catalog of pre-built Agents across every use case and vertical.
  </Card>
</CardGroup>

Click **Clone Agent** on any listing, then connect the integrations it needs (CRM, email, web search). Run it on a real account to see it work.

## Pick your path

How hands-on do you want to be? Pick a tab — the rest of the page is written for you.

<div>
  <Tabs>
    <Tab title="No build">
      You don't need to touch the builder. Open the cloned Agent above, ask in plain English, and let it work across the integrations it's connected to.

      ## Sharpen the answers with knowledge

      Upload ICP definitions, personas, and battle cards. Every conversation references them — sharper output, no new setup.

      <Card title="Add knowledge" icon="book" href="/build/knowledge/create-knowledge">
        Upload context once. Every conversation references it.
      </Card>

      ## Make it run on its own

      When you find yourself asking for the same thing every day, set a trigger on a cloned Agent — research every new lead, draft a brief before every meeting. One toggle, no code, no builder.

      <Card title="Triggers" icon="bolt" href="/build/agents/build-your-agent/triggers">
        Schedule runs, fire on CRM events, listen to webhooks.
      </Card>

      ## Ready to go further?

      When prompting and templates stop being enough — when you want the Agent to follow your team's exact playbook word for word — pick the **Build visually** tab above.
    </Tab>

    <Tab title="Build visually">
      You're comfortable in product UIs. You don't want to wire up an Agent from scratch, but you do want it to sound like your team and follow your process. Open the cloned Agent above in the builder and shape every piece.

      ## Connect the tools it needs

      Sales Agents work best when they can read from and write to the systems your team already uses.

      * **CRM** — [HubSpot](/docs/integrations/popular-integrations/hubspot) or [Salesforce](/docs/integrations/popular-integrations/salesforce) for account data and activity logging
      * **Email** — [Gmail](/docs/integrations/popular-integrations/gmail) or [Outlook](/docs/integrations/popular-integrations/outlook) for sending outreach
      * **Prospecting** — [Apollo](/docs/integrations/popular-integrations/apollo) or [ZoomInfo](/docs/integrations/popular-integrations/zoominfo) for contact and firmographic data

      See more about [Integrations](/docs/integrations/introduction). Start with your CRM and one outreach channel; add the rest as the workflow grows.

      ## Shape every surface

      Each piece below changes how the Agent behaves. Run the Agent on a real account between every change — small iterations beat big rewrites.

      <CardGroup>
        <Card title="Prompt" icon="message" href="/build/agents/build-your-agent/prompt">
          The system prompt is the Agent's playbook. Edit it to match your tone, your qualification criteria, your escalation rules.
        </Card>

        <Card title="Tools" icon="wrench" href="/build/agents/build-your-agent/tools">
          Add or remove the actions the Agent can take — search the web, query your CRM, send emails, run another Agent.
        </Card>

        <Card title="Knowledge" icon="book" href="/build/knowledge/create-knowledge">
          Upload ICP definitions, personas, battle cards, product one-pagers. The Agent references these whenever it works.
        </Card>

        <Card title="Triggers" icon="bolt" href="/build/agents/build-your-agent/triggers">
          Run the Agent automatically — on a schedule, when a CRM record updates, when a webhook fires.
        </Card>

        <Card title="Alerts" icon="bell" href="/build/agents/build-your-agent/alerts">
          Get notified when the Agent hits a threshold or runs into an issue. Use the prompt to instruct it which cases to flag for human review.
        </Card>

        <Card title="Memory" icon="brain" href="/build/agents/build-your-agent/memory">
          Give the Agent persistent context across conversations — past interactions, account history, learned preferences.
        </Card>
      </CardGroup>

      ## Chain Agents into a workforce

      When a single Agent isn't enough — research, then enrich, then write outreach, then send — wire Agents together on the Workforce canvas.

      <Card title="Build a workforce" icon="diagram-project" href="/build/workforces/create-a-workforce">
        Drag, connect, and run multi-Agent pipelines without code.
      </Card>

      ## Test before you trust

      Once you're shaping prompts and tools, set up evals so you catch regressions before customers do.

      <Card title="Evals" icon="vial" href="/build/agents/build-your-agent/evals">
        Define test cases in the UI. Run them on every change.
      </Card>

      ## Ready to go further?

      When clicking through the builder starts to feel slow — when you'd rather have your AI coding assistant build and edit Agents for you, drop into custom Python or JS code blocks, or call Agents from your own app via the SDK — pick the **Build with AI** tab above.
    </Tab>

    <Tab title="Build with AI">
      You already work in an AI coding environment. Clicking through a UI to edit a system prompt feels slow. Have your AI assistant build, shape, and orchestrate Agents directly from your editor.

      ## MCP & Plugins

      Connect your AI coding environment to Relevance AI over MCP. Describe what you want in natural language; your AI assistant reads and writes your Relevance workspace — system prompts, Tools, knowledge, triggers, workforces, evals.

      <CardGroup>
        <Card title="Claude Code plugin" icon="terminal" href="/integrations/mcp/claude-code">
          The fastest setup. Install the Relevance AI plugin and build Agents from your terminal.
        </Card>

        <Card title="MCP server" icon="plug" href="/integrations/mcp/mcp-server">
          Connect from Claude Desktop, Cursor, VS Code, ChatGPT, OpenAI Codex, or any MCP client.
        </Card>

        <Card title="Agent skills" icon="graduation-cap" href="/integrations/mcp/agent-skills">
          Clone the skills repo so your coding assistant has built-in knowledge of Relevance patterns.
        </Card>

        <Card title="MCP & Plugins overview" icon="book" href="/get-started/core-concepts/mcp-plugins">
          The full reference for what you can do programmatically.
        </Card>
      </CardGroup>
    </Tab>
  </Tabs>
</div>

## The four levels in sales

How the [autonomy ladder](/docs/guides) plays out across your pipeline:

<CardGroup>
  <Card title="L1 — Assisted" icon="hand">
    A rep asks the Agent before a call — "research Acme Corp" — and gets a brief back.
  </Card>

  <Card title="L2 — Copilot" icon="map">
    A rep asks for 12 personalized LinkedIn messages — the Agent drafts all 12, the rep approves and sends.
  </Card>

  <Card title="L3 — Autopilot" icon="bolt">
    A new lead lands in HubSpot — research, enrichment, outbound, and CRM logging all happen automatically. The rep sees the result, not the queue.
  </Card>

  <Card title="L4 — Self-driving" icon="infinity">
    You set the objective — "increase qualified meetings from inbound" — and the Workforce researches, enriches, drafts outbound, and logs to the CRM, with its copy and qualification thresholds tuned through evals as results flow back.
  </Card>
</CardGroup>

## Zoom into a single use case

Each use case below is a working entry point — clone the template, then run it, shape it visually, or build with AI depending on which tab fit you.

<CardGroup>
  <Card title="Prospect research" icon="building" href="/guides/sales/prospect-research-use-case">
    Automate company and contact research before calls.
  </Card>

  <Card title="Personalized outbound" icon="paper-plane" href="/guides/sales/personalized-outbound">
    Generate tailored emails and LinkedIn messages at scale.
  </Card>

  <Card title="Lead scoring" icon="chart-bar" href="/guides/sales/lead-scoring">
    Qualify and prioritize inbound leads automatically.
  </Card>

  <Card title="CRM data enrichment" icon="table" href="/guides/sales/crm-data-enrichment">
    Keep CRM records accurate and complete.
  </Card>

  <Card title="Meeting briefs" icon="clipboard" href="/guides/sales/meeting-briefs">
    Generate structured call prep for every meeting.
  </Card>

  <Card title="Competitive intelligence" icon="chess" href="/guides/sales/competitive-intelligence">
    Monitor competitors and surface insights for reps.
  </Card>
</CardGroup>

## See it in production

<Frame>
  <iframe title="Lightspeed x Relevance AI" />
</Frame>


# Lead scoring
Source: https://relevanceai.com/docs/guides/sales/lead-scoring

Qualify and prioritize leads so reps spend their time on the accounts most likely to close.

Not every lead deserves the same effort. A scoring Agent evaluates incoming leads against your ICP and buying signals so reps focus on the accounts most likely to close — and disqualify the ones that won't.

## When this pays off

<CardGroup>
  <Card title="High inbound volume" icon="inbox">
    Leads are arriving faster than reps can manually triage them.
  </Card>

  <Card title="Equal time on tire-kickers" icon="scale-unbalanced">
    Reps are spending the same hour on a Fortune 500 prospect as a free-trial signup with no fit.
  </Card>

  <Card title="MQLs losing momentum" icon="hourglass-half">
    Marketing-qualified leads are sitting unrouted while marketing complains the funnel is broken.
  </Card>

  <Card title="ICP exists on paper only" icon="file">
    You have an ICP doc but the team isn't applying it consistently across new leads.
  </Card>
</CardGroup>

## The shape of this use case

A scoring Agent takes a lead record and returns a qualification assessment.

<CardGroup>
  <Card title="Inputs" icon="arrow-right-to-bracket">
    CRM lead record, form-fill data, intent signals, web activity.
  </Card>

  <Card title="Sources" icon="globe">
    CRM, marketing automation, ICP and qualification framework in Knowledge, web search for missing context.
  </Card>

  <Card title="Output" icon="file-lines">
    A score with a reasoning summary — not just a number — plus a recommended action.
  </Card>

  <Card title="Delivery" icon="paper-plane">
    Written back to the CRM as a field, posted to [Slack](/docs/integrations/popular-integrations/slack) for hot leads, used to auto-route to the right rep.
  </Card>
</CardGroup>

## Where to start

Two ways in, depending on whether you want something running today or built to your exact spec.

<CardGroup>
  <Card title="Clone a pre-built Agent" icon="copy">
    Open **[Lia, the LinkedIn Lead Researcher & Qualifier](https://marketplace.relevanceai.com/listing/ba230945-e61a-4fb0-8c47-7e240f18456d)**. More in the [Marketplace](/docs/get-started/marketplace/introduction).
  </Card>

  <Card title="Build your own" icon="hammer">
    Start from scratch in the [builder](/docs/build/introduction), or by describing it in Claude Code or Cursor with [MCP & Plugins](/docs/get-started/core-concepts/mcp-plugins).
  </Card>
</CardGroup>

Either way, these are prompts your reps can use on day one:

* *"Score this lead — they came in from the pricing page yesterday. Is it worth a first call?"*
* *"Why did we lose Globex last quarter, and is this new account similar?"*
* *"Walk me through the top 10 new leads from this week and what to do with each."*

## Where to take it

Once it's running, deepen it in three moves:

<CardGroup>
  <Card title="Give it a playbook" icon="book">
    Shape it with a [prompt](/docs/build/agents/build-your-agent/prompt), your ICP in [Knowledge](/docs/build/knowledge/create-knowledge), and [Bulk Schedule](/docs/build/agents/give-your-agent-tasks/bulk-schedule) to backfill.
  </Card>

  <Card title="Automate it on signals" icon="bolt">
    Wrap it in a [workflow](/docs/build/workforces/create-a-workforce) that fires on a [trigger](/docs/build/agents/build-your-agent/triggers).
  </Card>

  <Card title="Let it improve" icon="arrows-rotate">
    Feed closed-won and closed-lost outcomes back into the Agent's [evals](/docs/build/agents/build-your-agent/evals) so "qualified" tracks what closes.
  </Card>
</CardGroup>

## Common pitfalls

<AccordionGroup>
  <Accordion title="Scoring without enrichment first" icon="database">
    Thin CRM data produces unreliable scores. Enrich the record before you ask the Agent to judge it.
  </Accordion>

  <Accordion title="Treating the score as truth" icon="gauge-high">
    Reps stop using their own judgment if the number is treated as final. The score is an input — surface the reasoning so reps can override it.
  </Accordion>

  <Accordion title="Stale ICP definitions" icon="calendar-xmark">
    If the ICP in Knowledge hasn't been touched in 18 months, the Agent is judging against last year's reality. Review it quarterly against actual close-won data.
  </Accordion>

  <Accordion title="No feedback loop from outcomes" icon="arrows-rotate">
    Without won/lost deals flowing back into the scoring criteria, the model can't learn. Pipe deal outcomes back to the Agent's evals.
  </Accordion>
</AccordionGroup>


# Meeting briefs
Source: https://relevanceai.com/docs/guides/sales/meeting-briefs

Generate structured call prep so reps walk into every meeting already up to speed.

Reps shouldn't be scrambling for context five minutes before a call. A meeting brief Agent pulls together everything relevant about the account — recent conversations, deal history, company updates — into a snapshot a rep can scan in under a minute.

## When this pays off

<CardGroup>
  <Card title="Back-to-back calendars" icon="calendar-days">
    Reps run meetings in sequence with no buffer to prep between them.
  </Card>

  <Card title="Context scattered everywhere" icon="puzzle-piece">
    Account history lives across CRM, Slack, support tickets, and email — and key details get missed.
  </Card>

  <Card title="New reps inheriting accounts" icon="user-plus">
    Reps joining mid-deal need to come up to speed without reading every email thread.
  </Card>

  <Card title="Exec drop-ins on deals" icon="user-tie">
    Leadership joins late-stage calls and needs the picture in two minutes, not twenty.
  </Card>
</CardGroup>

## The shape of this use case

A meeting brief Agent compiles a one-page snapshot of the account ahead of a scheduled call.

<CardGroup>
  <Card title="Inputs" icon="arrow-right-to-bracket">
    Calendar event — account, attendees, deal stage, meeting time.
  </Card>

  <Card title="Sources" icon="globe">
    CRM activity, past meeting notes, email threads, support tickets, company news.
  </Card>

  <Card title="Output" icon="file-lines">
    A brief covering deal context, recent activity, company updates, and suggested talking points.
  </Card>

  <Card title="Delivery" icon="paper-plane">
    Emailed to the rep before the meeting, attached to the calendar event, or posted in [Slack](/docs/integrations/popular-integrations/slack).
  </Card>
</CardGroup>

## Where to start

Two ways in, depending on whether you want something running today or built to your exact spec.

<CardGroup>
  <Card title="Clone a pre-built Agent" icon="copy">
    Open the **[Pre-Meeting Prepper](https://marketplace.relevanceai.com/listing/e2bc9eb0-6afd-4e24-b2ba-16160259981a)**. More in the [Marketplace](/docs/get-started/marketplace/introduction).
  </Card>

  <Card title="Build your own" icon="hammer">
    Start from scratch in the [builder](/docs/build/introduction), or by describing it in Claude Code or Cursor with [MCP & Plugins](/docs/get-started/core-concepts/mcp-plugins).
  </Card>
</CardGroup>

Either way, these are prompts your reps can use on day one:

* *"What do I need to know about Acme Corp before my 3pm? Pull the last three emails and recent deal activity."*
* *"Summarize where we are with Globex — last call notes, open objections, next steps."*
* *"Who from Initech is on my calendar tomorrow and what should I prep on each of them?"*

## Where to take it

Once it's running, deepen it in three moves:

<CardGroup>
  <Card title="Give it a playbook" icon="book">
    Shape it with a [prompt](/docs/build/agents/build-your-agent/prompt), [Knowledge](/docs/build/knowledge/create-knowledge), and a connected CRM, email, and [calendar](/docs/integrations/popular-integrations/google-calendar).
  </Card>

  <Card title="Automate it on signals" icon="bolt">
    Wrap it in a [workflow](/docs/build/workforces/create-a-workforce) that fires on a [trigger](/docs/build/agents/build-your-agent/triggers).
  </Card>

  <Card title="Let it improve" icon="arrows-rotate">
    Feed back which sections reps actually use into the Agent's [evals](/docs/build/agents/build-your-agent/evals) so briefs sharpen over time.
  </Card>
</CardGroup>

## Common pitfalls

<AccordionGroup>
  <Accordion title="Briefs that are too long" icon="hourglass-half">
    If the brief takes more than a minute to scan, reps stop reading. Cap section lengths in the prompt and force prioritization.
  </Accordion>

  <Accordion title="Static account data with no recency" icon="clock-rotate-left">
    A brief that just regurgitates the CRM record isn't useful. The Agent has to surface what's *new* — last contact, last news, last ticket.
  </Accordion>

  <Accordion title="Same depth for every meeting" icon="layer-group">
    A discovery call and a renewal call need different briefs. Branch the prompt by meeting type or deal stage.
  </Accordion>

  <Accordion title="Raw data instead of insight" icon="wand-magic-sparkles">
    "2 open tickets" is data. "Two open critical tickets opened this week — both about billing" is a talking point. Have the Agent interpret, not just enumerate.
  </Accordion>
</AccordionGroup>


# Personalized outbound
Source: https://relevanceai.com/docs/guides/sales/personalized-outbound

Draft tailored emails and LinkedIn messages from real prospect context, then queue them for the rep to approve and send.

Generic outreach gets ignored. Hand-crafted messages get replies — but they don't scale to a list of 500 prospects. An outbound Agent reads each prospect's CRM and LinkedIn context, drafts a personalized message, and queues it for rep approval.

## When this pays off

<CardGroup>
  <Card title="High outbound volume" icon="paper-plane">
    SDRs running 100+ touches a week and personalization is collapsing into copy-paste.
  </Card>

  <Card title="Flatlining reply rates" icon="chart-line-down">
    The team is writing more emails and getting fewer replies — the messages stopped sounding different.
  </Card>

  <Card title="Top-10% personalization" icon="medal">
    Reps only hand-personalize their A-list. Everyone else gets a template they're not happy with.
  </Card>

  <Card title="Multi-channel sequences" icon="shuffle">
    Email + LinkedIn + follow-up timing is breaking down because no human is orchestrating it.
  </Card>
</CardGroup>

## The shape of this use case

An outbound Agent takes a prospect and produces a message ready to send.

<CardGroup>
  <Card title="Inputs" icon="arrow-right-to-bracket">
    Prospect name, company, role, recent activity, CRM history.
  </Card>

  <Card title="Sources" icon="globe">
    CRM, [LinkedIn](/docs/integrations/popular-integrations/linkedin), web and news, your messaging playbooks, ICP and persona docs.
  </Card>

  <Card title="Output" icon="file-lines">
    A drafted email or LinkedIn message tuned to that prospect, in your team's voice.
  </Card>

  <Card title="Delivery" icon="paper-plane">
    Queued in Gmail/Outlook, posted to LinkedIn, written to the CRM, or dropped in a rep's queue for review.
  </Card>
</CardGroup>

## Where to start

Two ways in, depending on whether you want something running today or built to your exact spec.

<CardGroup>
  <Card title="Clone a pre-built Agent" icon="copy">
    Open **[LinkedIn Outreach & Follow up](https://marketplace.relevanceai.com/listing/5508c7f6-f3a7-4593-8faf-22fbecc4e0d4)** — or **[Outbound Composer](https://marketplace.relevanceai.com/listing/42f2cab0-71d6-41ca-b71e-fbaa55dfc53a)** for multi-variant email. More in the [Marketplace](/docs/get-started/marketplace/introduction).
  </Card>

  <Card title="Build your own" icon="hammer">
    Start from scratch in the [builder](/docs/build/introduction), or by describing it in Claude Code or Cursor with [MCP & Plugins](/docs/get-started/core-concepts/mcp-plugins).
  </Card>
</CardGroup>

Either way, these are prompts your reps can use on day one:

* *"Draft a LinkedIn DM to Sarah at Initech — she just got promoted to VP of Sales last month."*
* *"Write a follow-up email to Acme Corp referencing our last call notes in HubSpot."*
* *"Give me three subject line options for a re-engagement email to dormant trial users."*

## Where to take it

Once it's running, deepen it in three moves:

<CardGroup>
  <Card title="Give it a playbook" icon="book">
    Shape it with a [prompt](/docs/build/agents/build-your-agent/prompt), [Knowledge](/docs/build/knowledge/create-knowledge), and [Bulk Schedule](/docs/build/agents/give-your-agent-tasks/bulk-schedule).
  </Card>

  <Card title="Automate it on signals" icon="bolt">
    Wrap it in a [workflow](/docs/build/workforces/create-a-workforce) that fires on a [trigger](/docs/build/agents/build-your-agent/triggers).
  </Card>

  <Card title="Let it improve" icon="arrows-rotate">
    Feed reply and meeting rates back into the Agent's [evals](/docs/build/agents/build-your-agent/evals) so drafts work from what converts.
  </Card>
</CardGroup>

## Common pitfalls

<AccordionGroup>
  <Accordion title="Personalization that's surface-level" icon="eye">
    Mentioning the company name doesn't count. The opening line has to reference something the prospect would only believe you noticed if you actually paid attention — a post, a launch, a role change.
  </Accordion>

  <Accordion title="Sending too fast" icon="paper-plane">
    Volume plus zero warm-up equals the spam folder. Use sending caps and a warm-up schedule on new domains.
  </Accordion>

  <Accordion title="Every message reads the same" icon="clone">
    If your reply rates flatline as volume grows, the Agent has converged on a template. Force structural variation in the prompt — different opens, different CTA placements.
  </Accordion>

  <Accordion title="No human gate on exec accounts" icon="user-check">
    Use [Alerts](/docs/build/agents/build-your-agent/alerts) to route messages for high-value accounts through human review before they send.
  </Accordion>
</AccordionGroup>


# Prospect research
Source: https://relevanceai.com/docs/guides/sales/prospect-research-use-case

Automate company and contact research so reps walk into every call already prepared.

Reps spend hours digging through LinkedIn, company websites, and news to prep for a single call. Relevance can produce the same brief in seconds — at whatever level of automation you're ready for.

## When this pays off

<CardGroup>
  <Card title="High call volume" icon="phone">
    Reps run more than five discovery calls a week and prep is eating their selling time.
  </Card>

  <Card title="Uneven prep" icon="scale-unbalanced">
    Some calls get deeply prepped, others get a quick scan — output quality is inconsistent.
  </Card>

  <Card title="Inbound triage" icon="filter">
    Inbound leads need quick context before a rep is assigned.
  </Card>

  <Card title="Baseline context" icon="layer-group">
    Every meeting starts with the same baseline brief, not whatever the rep happens to find.
  </Card>
</CardGroup>

## The shape of this use case

A prospect research workflow takes a company name or contact and returns a structured brief.

<CardGroup>
  <Card title="Inputs" icon="arrow-right-to-bracket">
    Company domain, contact email, or CRM record.
  </Card>

  <Card title="Sources" icon="globe">
    Web search, [LinkedIn](/docs/integrations/popular-integrations/linkedin), news, your CRM, and any internal knowledge — ICP, personas, battle cards.
  </Card>

  <Card title="Output" icon="file-lines">
    A brief covering company overview, recent news, key contacts, likely pain points, and conversation hooks.
  </Card>

  <Card title="Delivery" icon="paper-plane">
    Returned inline, posted to [Slack](/docs/integrations/popular-integrations/slack), written back to the CRM, or attached to a calendar event.
  </Card>
</CardGroup>

## Where to start

Two ways in, depending on whether you want something running today or built to your exact spec.

<CardGroup>
  <Card title="Clone a pre-built Agent" icon="copy">
    Open the **[Sales Researcher](https://marketplace.relevanceai.com/listing/4b454347-0656-4fee-9c15-8fb2f2713424)** — or **[Lia](https://marketplace.relevanceai.com/listing/ba230945-e61a-4fb0-8c47-7e240f18456d)** for LinkedIn-first. More in the [Marketplace](/docs/get-started/marketplace/introduction).
  </Card>

  <Card title="Build your own" icon="hammer">
    Start from scratch in the [builder](/docs/build/introduction), or by describing it in Claude Code or Cursor with [MCP & Plugins](/docs/get-started/core-concepts/mcp-plugins).
  </Card>
</CardGroup>

Either way, these are prompts your reps can use on day one:

* *"Research Acme Corp ahead of my 2pm call. Pull recent news, the buyer's LinkedIn, and any open deals in HubSpot."*
* *"Summarize the last three meetings with Globex and what's still open."*
* *"Who at Initech should I be talking to if we're selling into RevOps?"*

## Where to take it

Once it's running, deepen it in three moves:

<CardGroup>
  <Card title="Give it a playbook" icon="book">
    Shape it with a [prompt](/docs/build/agents/build-your-agent/prompt), [Knowledge](/docs/build/knowledge/create-knowledge), and [Bulk Schedule](/docs/build/agents/give-your-agent-tasks/bulk-schedule).
  </Card>

  <Card title="Automate it on signals" icon="bolt">
    Wrap it in a [workflow](/docs/build/workforces/create-a-workforce) that fires on a [trigger](/docs/build/agents/build-your-agent/triggers).
  </Card>

  <Card title="Let it improve" icon="arrows-rotate">
    Feed won-deal signals back into the Agent's [evals](/docs/build/agents/build-your-agent/evals) so it tracks what closes.
  </Card>
</CardGroup>

<Frame>
  <iframe title="Build an AI sales research agent" />
</Frame>

## Common pitfalls

<AccordionGroup>
  <Accordion title="Researching accounts you already own" icon="user-check">
    Gate the trigger on lifecycle stage so you don't burn credits re-researching closed-won customers.
  </Accordion>

  <Accordion title="Generic briefs" icon="copy">
    Without an ICP or persona in knowledge, the output reads like a Wikipedia summary. Add the lens.
  </Accordion>

  <Accordion title="Stale data" icon="hourglass-half">
    Set a freshness window — e.g. re-research only if the last brief is more than 30 days old — instead of regenerating on every interaction.
  </Accordion>

  <Accordion title="Over-trusting recency" icon="magnifying-glass">
    Web search returns whatever ranks, not what's true. Cite sources in the brief so reps can spot-check.
  </Accordion>
</AccordionGroup>


# Demo prep and pre-call briefs
Source: https://relevanceai.com/docs/guides/solution-engineering/demo-prep

Account-aware technical briefs the SE can read in five minutes before any call.

SEs walk into demos blind more often than they'd admit — they know the prospect's name and the deal size, not their tech stack, integration patterns, or what their team has been Googling. A demo-prep Agent builds the brief the SE wishes they had, every time.

## When this pays off

<CardGroup>
  <Card title="Back-to-back demos" icon="calendar-days">
    SEs run 4+ demos a day and prep blurs into the prior one.
  </Card>

  <Card title="Technical context is shallow" icon="layer-group">
    The CRM has firmographics; the SE needs tech stack, scale, and likely integration patterns.
  </Card>

  <Card title="Demos feel generic" icon="copy">
    SEs use the same demo flow for every prospect because there's no time to tailor.
  </Card>

  <Card title="Tribal knowledge in handoffs" icon="user-group">
    AE-to-SE deal handoffs lose detail. The SE rediscovers what the AE already learned.
  </Card>
</CardGroup>

## The shape of this use case

A demo-prep Agent takes a calendar event + opportunity and returns a structured technical brief.

<CardGroup>
  <Card title="Inputs" icon="arrow-right-to-bracket">
    Calendar event, opportunity record, attendees, deal stage, prior conversation notes.
  </Card>

  <Card title="Sources" icon="globe">
    CRM, prior call notes, public tech stack signals, past similar deals, your demo playbook.
  </Card>

  <Card title="Output" icon="file-lines">
    A one-page brief — account context, likely tech stack, integration angles, suggested demo flow, anticipated objections.
  </Card>

  <Card title="Delivery" icon="paper-plane">
    Emailed to the SE before the call, attached to the calendar event, posted in [Slack](/docs/integrations/popular-integrations/slack) to the deal channel.
  </Card>
</CardGroup>

## Where to start

Two ways in, depending on whether you want something running today or built to your exact spec.

<CardGroup>
  <Card title="Clone a pre-built Agent" icon="copy">
    Open the **[Pre-Meeting Prepper](https://marketplace.relevanceai.com/listing/e2bc9eb0-6afd-4e24-b2ba-16160259981a)**. More in the [Marketplace](/docs/get-started/marketplace/introduction).
  </Card>

  <Card title="Build your own" icon="hammer">
    Start from scratch in the [builder](/docs/build/introduction), or by describing it in Claude Code or Cursor with [MCP & Plugins](/docs/get-started/core-concepts/mcp-plugins).
  </Card>
</CardGroup>

Either way, these are prompts your SEs can use on day one:

* *"Demo with Acme tomorrow — what's their tech stack, what integration angles should I lean on, and what objections has the AE flagged?"*
* *"Globex is on my calendar next Tuesday. Walk me through how to position our enterprise SSO story for their auth setup."*
* *"What's the right demo flow for a mid-market manufacturing prospect comparing us against Competitor X?"*

## Where to take it

Once it's running, deepen it in three moves:

<CardGroup>
  <Card title="Give it a playbook" icon="book">
    Shape it with a [prompt](/docs/build/agents/build-your-agent/prompt) and your demo playbook, battle cards, and objection docs in [Knowledge](/docs/build/knowledge/create-knowledge).
  </Card>

  <Card title="Automate it on signals" icon="bolt">
    Wrap it in a [workflow](/docs/build/workforces/create-a-workforce) that fires on a [trigger](/docs/build/agents/build-your-agent/triggers).
  </Card>

  <Card title="Let it improve" icon="arrows-rotate">
    Feed back which brief sections SEs use into the Agent's [evals](/docs/build/agents/build-your-agent/evals) so briefs sharpen over time.
  </Card>
</CardGroup>

## Common pitfalls

<AccordionGroup>
  <Accordion title="Briefs that are too long" icon="hourglass-half">
    A five-page brief doesn't get read in the five minutes before a call. Cap the length and force the prompt to prioritize.
  </Accordion>

  <Accordion title="Generic recommendations" icon="clone">
    Without segment-specific Knowledge, the brief recommends the same demo flow for every prospect. Branch on segment and use real exemplars.
  </Accordion>

  <Accordion title="Stale prospect data" icon="clock-rotate-left">
    The CRM hasn't been updated since the discovery call but the brief reads as if it were live. Have the Agent disclose when the freshest signal is and flag stale data.
  </Accordion>

  <Accordion title="Missing the AE's context" icon="user-check">
    The brief ignores what the AE already learned because notes live in another tool. Pull AE notes into the source set explicitly.
  </Accordion>
</AccordionGroup>


# Technical discovery summaries
Source: https://relevanceai.com/docs/guides/solution-engineering/discovery-summaries

Post-call writeups that turn an hour of conversation into a structured summary with action items.

The richest part of a deal is the technical discovery call — and the writeup is usually a CRM note half the team won't read. A discovery Agent turns the transcript into a structured summary: requirements, decisions, open questions, next steps. Everyone joining the next call walks in with the same picture.

## When this pays off

<CardGroup>
  <Card title="Writeups slip" icon="hourglass">
    The SE meant to write up the call. Three days later, it's vibes-in-a-CRM-note.
  </Card>

  <Card title="Stakeholder handoffs lose detail" icon="user-group">
    The next SE / AE on the deal can't pick up where the last call left off because the writeup is too sparse.
  </Card>

  <Card title="Action items get forgotten" icon="list-check">
    "We'll send you the SAML doc" — and then nobody does, because it wasn't tracked.
  </Card>

  <Card title="Demo prompts repeat themselves" icon="rotate">
    The next call repeats discovery because the last one wasn't captured well.
  </Card>
</CardGroup>

## The shape of this use case

A discovery Agent takes a call recording or transcript and returns a structured summary.

<CardGroup>
  <Card title="Inputs" icon="arrow-right-to-bracket">
    Call recording or transcript, opportunity context, prior call notes.
  </Card>

  <Card title="Sources" icon="globe">
    [Fireflies](/docs/integrations/popular-integrations/fireflies) or other transcript source, CRM history, your discovery framework / MEDDICC template.
  </Card>

  <Card title="Output" icon="file-lines">
    A structured summary — requirements, technical decisions, open questions, action items per party, links to relevant Knowledge.
  </Card>

  <Card title="Delivery" icon="paper-plane">
    Written back to the CRM as a structured note, posted in the deal [Slack](/docs/integrations/popular-integrations/slack) channel, emailed to attendees with a "did we get this right?" prompt.
  </Card>
</CardGroup>

## Where to start

Two ways in, depending on whether you want something running today or built to your exact spec.

<CardGroup>
  <Card title="Clone a pre-built Agent" icon="copy">
    Open **[Nate, Meeting Notetaker](https://marketplace.relevanceai.com/listing/693db710-3d82-4269-9bce-ab6c9199e122)**. More in the [Marketplace](/docs/get-started/marketplace/introduction).
  </Card>

  <Card title="Build your own" icon="hammer">
    Start from scratch in the [builder](/docs/build/introduction), or by describing it in Claude Code or Cursor with [MCP & Plugins](/docs/get-started/core-concepts/mcp-plugins).
  </Card>
</CardGroup>

Either way, these are prompts your SEs can use on day one:

* *"Summarize this discovery call with Acme — pull out their requirements, our open questions, and the action items per side."*
* *"What were the technical concerns raised on the Globex call last week, and which ones did we resolve?"*
* *"Walk me through the SSO discussion from yesterday's Initech call — what did they say their constraint was?"*

## Where to take it

Once it's running, deepen it in three moves:

<CardGroup>
  <Card title="Give it a playbook" icon="book">
    Shape it with a [prompt](/docs/build/agents/build-your-agent/prompt), your discovery framework in [Knowledge](/docs/build/knowledge/create-knowledge), and [Bulk Schedule](/docs/build/agents/give-your-agent-tasks/bulk-schedule).
  </Card>

  <Card title="Automate it on signals" icon="bolt">
    Wrap it in a [workflow](/docs/build/workforces/create-a-workforce) that fires on a [trigger](/docs/build/agents/build-your-agent/triggers).
  </Card>

  <Card title="Let it improve" icon="arrows-rotate">
    Feed back which sections SEs reference into the Agent's [evals](/docs/build/agents/build-your-agent/evals) so summaries track what matters.
  </Card>
</CardGroup>

## Common pitfalls

<AccordionGroup>
  <Accordion title="Summary that loses the actual technical content" icon="magnifying-glass">
    Generic LLM summarization smooths out the specifics. Force the prompt to preserve technical terms, exact constraints, and the customer's literal phrasing.
  </Accordion>

  <Accordion title="No action-item ownership" icon="list-check">
    Action items without an owner and a due date don't get done. Require the Agent to assign and date each.
  </Accordion>

  <Accordion title="Auto-emailing without SE review" icon="paper-plane">
    Sending an inaccurate summary to the customer hurts trust. Keep SE review on the customer-facing copy until you've watched it for several deals.
  </Accordion>

  <Accordion title="Missing the implied dealbreakers" icon="eye">
    Some constraints are stated as "wishes" but they're really hard requirements. Have the prompt flag any ambiguous language for SE review.
  </Accordion>
</AccordionGroup>


# Solution Engineering
Source: https://relevanceai.com/docs/guides/solution-engineering/getting-started

Put AI Agents to work across pre-sales technical work — RFPs, demos, discovery — so SEs spend time architecting solutions, not chasing documents.

Solution engineers spend half their time on document work — RFP responses, security questionnaires, demo prep — and half on the strategic conversations that actually win deals. AI Agents take the document work so SEs can focus on the architecture and the relationships.

## Start with a Marketplace template

The Marketplace ships pre-built Agents you can clone in one click. Adapt them to your products, your security posture, your standard demo flow. Or [build a custom Agent from scratch](/docs/build/introduction).

<CardGroup>
  <Card title="Pre-Meeting Prepper" icon="clipboard-check" href="https://marketplace.relevanceai.com/listing/e2bc9eb0-6afd-4e24-b2ba-16160259981a">
    Pre-call research for discovery and demo prep — pulls company, tech stack, and stakeholder context before you walk in.
  </Card>

  <Card title="AI RFP Response Generator" icon="file-contract" href="https://marketplace.relevanceai.com/listing/d1092947-1746-4df1-9596-255d13796a2c">
    Upload an RFP or security questionnaire plus your knowledge sources — get a filled draft back with citations and trust scores.
  </Card>

  <Card title="Browse the Marketplace" icon="store" href="/get-started/marketplace/introduction">
    The full catalog of Agents — RFP, security review, demo prep, discovery synthesis.
  </Card>
</CardGroup>

Click **Clone Agent** on any listing, connect the systems it needs (CRM, document storage, security reference docs), and run on a real opportunity.

## Pick your path

How hands-on do you want to be? Pick a tab — the rest of the page is written for you.

<div>
  <Tabs>
    <Tab title="No build">
      You don't need to touch the builder. Open the cloned Agent above, ask in plain English, and let it work across the systems it's connected to.

      ## Sharpen the answers with knowledge

      Upload your security questionnaire archive, prior RFP responses, architecture diagrams, and standard demo scripts. Every conversation references them — sharper output, no new setup.

      <Card title="Add knowledge" icon="book" href="/build/knowledge/create-knowledge">
        Upload context once. Every conversation references it.
      </Card>

      ## Make it run on its own

      When the same kinds of opportunities keep landing, set a trigger — every new SE-flagged deal gets a demo-prep brief, every inbound RFP triggers a first-pass response. One toggle, no code, no builder.

      <Card title="Triggers" icon="bolt" href="/build/agents/build-your-agent/triggers">
        Schedule runs, fire on CRM events, listen to webhooks.
      </Card>

      ## Ready to go further?

      When prompting and templates stop being enough — when you want the Agent to follow your exact security posture and orchestrate multi-doc deliverables — pick the **Build visually** tab above.
    </Tab>

    <Tab title="Build visually">
      You're comfortable in product UIs. You don't want to wire up an Agent from scratch, but you do want it to draft RFP answers in your exact voice and reference your real architecture decisions. Open the cloned Agent in the builder and shape every piece.

      ## Connect the tools it needs

      SE Agents work best when they can read from and write to the systems your team already uses.

      * **CRM** — [HubSpot](/docs/integrations/popular-integrations/hubspot) or [Salesforce](/docs/integrations/popular-integrations/salesforce) for opportunity context
      * **Document storage** — [Google Drive](/docs/integrations/popular-integrations/google-drive), [Notion](/docs/integrations/popular-integrations/notion), [Confluence](/docs/integrations/popular-integrations/confluence) for prior RFPs and architecture docs
      * **Security reference** — your past security questionnaires, SOC 2 evidence, policy docs
      * **Comms** — [Gmail](/docs/integrations/popular-integrations/gmail), [Slack](/docs/integrations/popular-integrations/slack) for delivery and internal review

      See more about [Integrations](/docs/integrations/introduction).

      ## Shape every surface

      Each piece below changes how the Agent behaves. Run it on a real opportunity between every change — small iterations beat big rewrites.

      <CardGroup>
        <Card title="Prompt" icon="message" href="/build/agents/build-your-agent/prompt">
          Edit the system prompt to match your team's tone, your security posture, your standard demo flow.
        </Card>

        <Card title="Tools" icon="wrench" href="/build/agents/build-your-agent/tools">
          Add or remove the actions the Agent can take — search past RFPs, query the CRM, draft response docs, run another Agent.
        </Card>

        <Card title="Knowledge" icon="book" href="/build/knowledge/create-knowledge">
          Upload prior RFP responses, security artifacts, architecture diagrams, demo scripts. The Agent draws on these whenever it works.
        </Card>

        <Card title="Triggers" icon="bolt" href="/build/agents/build-your-agent/triggers">
          Run the Agent automatically — on a schedule, when a new opportunity needs SE involvement, when an RFP lands.
        </Card>

        <Card title="Alerts" icon="bell" href="/build/agents/build-your-agent/alerts">
          Get notified when a security question can't be answered confidently or when a demo prep flags missing info.
        </Card>

        <Card title="Memory" icon="brain" href="/build/agents/build-your-agent/memory">
          Persistent context across deals — prior architecture discussions, previously raised security concerns, repeated objections.
        </Card>
      </CardGroup>

      ## Chain Agents into a workforce

      When a single Agent isn't enough — discovery summary, then demo prep, then technical proposal — wire them together on the Workforce canvas.

      <Card title="Build a workforce" icon="diagram-project" href="/build/workforces/create-a-workforce">
        Drag, connect, and run multi-Agent pipelines without code.
      </Card>

      ## Test before you trust

      Set up evals so a wrong security answer or hallucinated architecture detail gets caught before it lands in front of a prospect.

      <Card title="Evals" icon="vial" href="/build/agents/build-your-agent/evals">
        Define test cases in the UI. Run them on every change.
      </Card>

      ## Ready to go further?

      When clicking through the builder starts to feel slow — when you'd rather have your AI coding assistant build and edit Agents for you, drop into custom Python or JS code blocks, or call Agents from your own app via the SDK — pick the **Build with AI** tab above.
    </Tab>

    <Tab title="Build with AI">
      You already work in an AI coding environment. Clicking through a UI to edit RFP response logic feels slow. Have your AI assistant build, shape, and orchestrate Agents directly from your editor.

      ## MCP & Plugins

      Connect your AI coding environment to Relevance AI over MCP. Describe what you want in natural language; your AI assistant reads and writes your Relevance workspace — system prompts, Tools, knowledge, triggers, workforces, evals.

      <CardGroup>
        <Card title="Claude Code plugin" icon="terminal" href="/integrations/mcp/claude-code">
          The fastest setup. Install the Relevance AI plugin and build Agents from your terminal.
        </Card>

        <Card title="MCP server" icon="plug" href="/integrations/mcp/mcp-server">
          Connect from Claude Desktop, Cursor, VS Code, ChatGPT, OpenAI Codex, or any MCP client.
        </Card>

        <Card title="Agent skills" icon="graduation-cap" href="/integrations/mcp/agent-skills">
          Clone the skills repo so your coding assistant has built-in knowledge of Relevance patterns.
        </Card>

        <Card title="MCP & Plugins overview" icon="book" href="/get-started/core-concepts/mcp-plugins">
          The full reference for what you can do programmatically.
        </Card>
      </CardGroup>
    </Tab>
  </Tabs>
</div>

## The four levels in solution engineering

How the [autonomy ladder](/docs/guides) plays out across pre-sales:

<CardGroup>
  <Card title="L1 — Assisted" icon="hand">
    An SE asks the Agent to answer a specific security question or summarize prior architecture work for an account — the Agent returns a draft with sources.
  </Card>

  <Card title="L2 — Copilot" icon="map">
    An SE asks for an 80-question security questionnaire response — the Agent drafts all 80, the SE reviews and finalizes.
  </Card>

  <Card title="L3 — Autopilot" icon="bolt">
    A new opportunity hits Stage 2 → discovery summary, demo prep, and a first-draft technical proposal all generate, the SE reviews and refines.
  </Card>

  <Card title="L4 — Self-driving" icon="infinity">
    You set the objective — "improve technical win rate" — and the Workforce drafts RFP answers, discovery summaries, and demo prep, with its drafting criteria tuned through evals against win/loss outcomes.
  </Card>
</CardGroup>

## Zoom into a single use case

Each use case below is a working entry point — clone a template, then run it, shape it visually, or build with AI depending on which tab fit you.

<CardGroup>
  <Card title="RFP & security questionnaires" icon="file-shield" href="/guides/solution-engineering/rfp-responses">
    Pull from your archive of prior answers to draft consistent, accurate responses to questionnaires.
  </Card>

  <Card title="Demo prep & pre-call briefs" icon="presentation-screen" href="/guides/solution-engineering/demo-prep">
    Account-aware technical briefs the SE can read in five minutes before any call.
  </Card>

  <Card title="Technical discovery summaries" icon="clipboard-list" href="/guides/solution-engineering/discovery-summaries">
    Post-call writeups that turn an hour of conversation into a structured summary with action items.
  </Card>
</CardGroup>

## See it in production

<Frame>
  <iframe title="Canva x Relevance AI — Agents@Work" />
</Frame>

<Frame>
  <iframe title="KPMG x Relevance AI — Agents@Work" />
</Frame>


# RFP and security questionnaire responses
Source: https://relevanceai.com/docs/guides/solution-engineering/rfp-responses

Pull from your archive of prior answers to draft consistent, accurate responses to questionnaires.

Every SE team has the same archive — last year's RFPs, last quarter's security questionnaires, the architecture answers written six different ways across six different deals. An RFP Agent reads each new question and pulls the best prior answer, adapts it to the deal context, and flags anything that needs human judgment.

## When this pays off

<CardGroup>
  <Card title="Same answers, six versions" icon="copy">
    The data-residency answer exists in 12 prior RFPs, all slightly different. New responses pick whatever's easiest to find.
  </Card>

  <Card title="Questionnaires take days" icon="hourglass">
    An 80-question security questionnaire eats two SE days. The deal slows because the document is the bottleneck.
  </Card>

  <Card title="Inconsistent answers across deals" icon="scale-unbalanced">
    Security or compliance answers vary by who wrote them, not by what's true. Risky for trust and contracts.
  </Card>

  <Card title="Archive is unsearchable" icon="folder-magnifying-glass">
    "Have we answered this before?" requires a 20-minute hunt across Drive, Notion, and Slack.
  </Card>
</CardGroup>

## The shape of this use case

An RFP Agent takes a question + opportunity context and returns a drafted answer with sources.

<CardGroup>
  <Card title="Inputs" icon="arrow-right-to-bracket">
    Question text, opportunity context (segment, geography, deal size), prior-answer archive.
  </Card>

  <Card title="Sources" icon="globe">
    Past RFP and security questionnaire responses, policy docs, architecture artifacts, product changelog.
  </Card>

  <Card title="Output" icon="file-lines">
    A drafted answer with citations to source documents and a confidence indicator — high-confidence answers ready to send, low-confidence flagged for SE review.
  </Card>

  <Card title="Delivery" icon="paper-plane">
    Drafted into the response doc (Google Doc, [Notion](/docs/integrations/popular-integrations/notion), Loopio export), posted in [Slack](/docs/integrations/popular-integrations/slack) for SE review on flagged answers.
  </Card>
</CardGroup>

## Where to start

Two ways in, depending on whether you want something running today or built to your exact spec.

<CardGroup>
  <Card title="Clone a pre-built Agent" icon="copy">
    Open the **[AI RFP Response Generator](https://marketplace.relevanceai.com/listing/d1092947-1746-4df1-9596-255d13796a2c)**. More in the [Marketplace](/docs/get-started/marketplace/introduction).
  </Card>

  <Card title="Build your own" icon="hammer">
    Start from scratch in the [builder](/docs/build/introduction), or by describing it in Claude Code or Cursor with [MCP & Plugins](/docs/get-started/core-concepts/mcp-plugins).
  </Card>
</CardGroup>

Either way, these are prompts your SEs can use on day one:

* *"What's our standard answer to 'describe your data residency options'? Cite past responses."*
* *"This question is asking about SOC 2 controls around access. Pull the right answer and adapt it for an enterprise EU customer."*
* *"Walk me through these 30 questions from Acme's questionnaire — flag the ones that need real SE attention."*

## Where to take it

Once it's running, deepen it in three moves:

<CardGroup>
  <Card title="Give it a playbook" icon="book">
    Shape it with a [prompt](/docs/build/agents/build-your-agent/prompt), your prior-response archive in [Knowledge](/docs/build/knowledge/create-knowledge), and [Bulk Schedule](/docs/build/agents/give-your-agent-tasks/bulk-schedule).
  </Card>

  <Card title="Automate it on signals" icon="bolt">
    Wrap it in a [workflow](/docs/build/workforces/create-a-workforce) that fires on a [trigger](/docs/build/agents/build-your-agent/triggers).
  </Card>

  <Card title="Let it improve" icon="arrows-rotate">
    Feed back which answers customers accepted into the Agent's [evals](/docs/build/agents/build-your-agent/evals) so it promotes the strong ones.
  </Card>
</CardGroup>

## Common pitfalls

<AccordionGroup>
  <Accordion title="Hallucinated security claims" icon="user-shield">
    The Agent confidently states a control you don't have. Force citations to actual policy / past-answer documents and fail visibly when the question isn't covered.
  </Accordion>

  <Accordion title="Stale answers" icon="calendar-xmark">
    Last year's accurate answer is this year's misrepresentation. Tag prior responses with dates and have the Agent prefer recent over old when in conflict.
  </Accordion>

  <Accordion title="No human review on legal/security" icon="scale-unbalanced">
    Some questions need legal sign-off. Always route security and contractual answers through human review until you've watched it for several deals.
  </Accordion>

  <Accordion title="Auto-applying without deal context" icon="sitemap">
    A "yes" to a feature in mid-market might be a "yes, but" in enterprise. Have the Agent read the opportunity record and adapt — not just look up the question.
  </Accordion>
</AccordionGroup>


# Add Integrations
Source: https://relevanceai.com/docs/integrations/add-integrations

Integrations transform your AI agents into powerful workflow automation tools by connecting them with external services and platforms.

## Browsing and filtering integrations

The **Integrations & API Keys** browser displays all available integrations. Use the search bar to find a specific service by name, or filter by category tag — CRM, Communication, AI Models, Automation & Integration, and more. Multiple tags can be selected at once.

## Setting Up Integrations

To begin using integrations with your AI agents:

1. Navigate to the **Integrations & API Keys** page from the left-hand menu of your Relevance AI account.
2. Browse the available integrations and select the service you want to connect.
3. Follow the authentication steps to authorize the connection.
4. Once connected, the integration will appear in your list of available connections.

## Adding Triggers to Agents

Triggers allow your agents to automatically activate when specific events occur in connected platforms:

1. Go to the Agent Settings page for the agent you want to configure.
2. Select Agent Profile from the menu.
3. Locate the Triggers section.
4. Click "Add Trigger" and select from the available trigger options.
5. Configure the trigger settings according to your needs.
6. Save your changes.

## Popular Integrations

### Slack Integration

Connect your agents to Slack to enable them to:

* Monitor channels for specific messages or keywords.
* Send notifications or updates to designated channels.
* Respond to direct messages or mentions.
* Perform actions based on Slack events.

### Google Calendar Integration

Link your agents with Google Calendar to:

* Schedule meetings and appointments.
* Send calendar invites to participants.
* Check availability before scheduling.
* Provide reminders for upcoming events.

### HubSpot Integration

Connect to HubSpot to allow your agents to:

* Create and update contact records.
* Track customer interactions.
* Manage deals and opportunities.
* Access customer data for personalized responses.


# Email warmups
Source: https://relevanceai.com/docs/integrations/integration-examples/email-warmups

Use case: Warm up your email before starting your email marketing campaigns.

## Why do you need to warmup your email inbox?

Email warmups refer to the process of gradually increasing the volume of emails sent from a new or unused email account to build a positive sender reputation with internet service providers (ISPs).

A warmedup email inbox helps you to avoid spam filters and ensure better deliverability rates.

We recommend that you create an email account especially for your email inbox manager agents. So warming up their inbox before sending out emails from their account will help your email marketing/inbox management campaigns start successfully.

## How to warmup your emails?

The process usually involvesn sending a small number of emails initially and slowly increasing the quantity over time. Start by sending a few emails to active, engaged recipients like colleagues who are likely to open and interact with your agent's emails.

We recommend starting with 4-8 warm-up emails per day per account with a target reply rate of 30%. After two weeks, you can increase the warm-up volume to 20-30 emails per day per account, with a higher target reply rate of 70% when engaging in outreach.

The maximum recommended warm-up email volume is 50 emails per day, with a reply rate of a max of 60%.

## Add email warmup integration to your Relevance account

You can use an in-built Relevance integration to warm up your chosen email account, by activating the email warmup integration.

Here is a video walking you through the process of doing this:

<div>
  <div>
    <iframe />
  </div>

  <p>
    This is one of our flagship agent templates, BDR Bosh (enterprise only).
  </p>
</div>

Step-by-step-process:

1. On the "Integrations" page.
2. Click on the "Google (Gmail, calendar & API)" integration.
3. Click on the 3 dot menu next to the email you want to warm up, then click on "Email Warmup Settings".
   * Sender name: This is who all your email warmup emails are going to come from. This could be your name, or just be a name you use to be able to classify the email warmup emails in your inbox.
   * App password, get it from the documentation link. When you click on that, make sure you’re in the right account first. Setup 2fa. Create and manage app passwords. Create a new password, call it Email Warmup. Copy paste the app password.
4. Click on 3 dots next to your email account, and click on "Email warmup settings" again.

You will now see basic statistics on how your email warmup is going.

<Frame>
  <img />
</Frame>

You can start and stop your email warmups at any time.


# Integrations
Source: https://relevanceai.com/docs/integrations/introduction

Connect your AI agents with external tools and services to create powerful, automated workflows across your entire tech stack.

## Overview

Integrations serve as bridges between your AI agents and the external services you rely on daily. By establishing these connections, you can create seamless workflows that span across multiple platforms, allowing your agents to access data from and perform actions within your entire digital ecosystem.

For new users, it's important to understand that integrations enable four key capabilities:

1. **Connecting external services** to your Relevance AI account
2. **Setting up triggers** that automatically activate your agents based on external events
3. **Using pre-built actions** that allow your agents to interact with external platforms
4. **Creating custom API calls** for advanced integration scenarios

Let's explore each of these capabilities in detail.

## Native vs. Pipedream integrations

Integrations fall into two types:

* **Native integrations** are built directly by Relevance AI. They offer deeper support, including pre-built tool steps and trigger events designed to work with agents out of the box.
* **Pipedream integrations** extend the catalog to 2,000+ services via a Pipedream connection. They support authentication and API access for a wide range of platforms not covered by native integrations.

<Tip>
  Start with native integrations when available — they come with ready-made actions and triggers that require less configuration.
</Tip>

## 1. Connecting Integrations

The first step in leveraging integrations is connecting your external services to Relevance AI:

1. Navigate to the **Integrations & API Keys** page from the left-hand menu of your Relevance AI account
2. Browse the available integrations and select the service you want to connect
3. Click on the integration card to begin the connection process
4. Follow the authentication steps to authorize the connection (typically involves logging into the service or providing API credentials)
5. Once connected, the integration will appear as "Connected" in your list of available integrations

<div>
  <iframe title="Connecting integrations" />
</div>

Each integration may have specific configuration options depending on the service. The platform guides you through the necessary steps to establish a secure and functional connection.

### Connecting LLM Accounts

<div>
  <iframe title="Adding API keys" />
</div>

A particularly valuable integration option is connecting your own Large Language Model (LLM) accounts:

1. Go to the **Integrations & API Keys** page in the left-hand menu
2. Look for the LLM provider you want to connect (such as OpenAI, Anthropic, or Google)
3. Click on the provider and enter your API key
4. Once connected, you can use your own LLM account with your agents

This capability allows you to leverage your existing LLM subscriptions and manage usage under your own account, giving you greater control over your AI resources and potentially reducing costs.

## 2. Setting Up Triggers

Triggers are events from integrated platforms that automatically activate your agents. This transforms your agents from reactive tools that wait for instructions into proactive assistants that respond to events in your digital environment.

### How to Set Up Triggers:

1. Go to the **Agent Settings** page for the agent you want to configure
2. Select **Agent Profile** from the menu
3. Locate the **Triggers** section
4. Click **"Add Trigger"** and select from the available trigger options
5. Configure the trigger settings according to your needs
6. Save your changes

<div>
  <iframe title="Setting up triggers" />
</div>

### Common Trigger Examples:

* **Email Integrations (Gmail, Outlook)**: Trigger an agent when you receive an email matching specific criteria
* **CRM Integrations (HubSpot, Salesforce)**: Activate an agent when a new contact is created or a deal stage changes
* **Support Integrations (Freshdesk, Zendesk)**: Start an agent workflow when a new support ticket is created
* **Messaging Integrations (Slack, WhatsApp Business)**: Trigger an agent when a message is received in a specific channel

For example, with the HubSpot integration, you could set up a trigger that activates your sales assistant agent whenever a new lead is created in your CRM. The agent could then automatically research the lead, prepare a personalized outreach message, and schedule a follow-up task.

## 3. Tools & Tool Steps (Actions)

This is where integrations truly shine! Once connected, integrations provide a library of pre-built actions that your agents can use to interact with external platforms. These actions can be incorporated into your agent's workflows at any point, not just as triggers.

### How to Add Integration Actions to Your Agent:

1. Go to the **Agent Settings** page for your agent
2. Select **Tools** under "Connected Resources" in the sidebar
3. Click **"+ Add tool"** to open the tool library
4. Search for the integration name (e.g., "HubSpot" or "Slack")
5. Browse the available actions for that integration
6. Select the action you want to add and click **"+ Add"**
7. Configure any required parameters for the action
8. Save your changes

<div>
  <iframe title="Adding integration actions" />
</div>

### Examples of Powerful Integration Actions:

**Slack Integration Actions:**

* Send a message to a specific channel
* Create a new channel
* Search for messages
* Update a message
* Upload a file
* Create a poll
* And many more!

**Google Calendar Integration Actions:**

* Create a new event
* Check availability
* Update an existing event
* Send calendar invites
* Get upcoming events
* Delete events
* And many more!

**HubSpot Integration Actions:**

* Create or update contacts
* Add notes to contact records
* Create deals
* Update deal stages
* Create tasks
* Search for contacts
* Get contact properties
* And many more!

**Notion Integration Actions:**

* Create a new page
* Update page content
* Search for pages
* Create databases
* Add items to databases
* Comment on pages
* And many more!

These actions are the building blocks of powerful workflows. For example, your agent could use the HubSpot "Get Contact" action to retrieve customer information, then use the "Create Task" action to set up a follow-up, and finally use the Slack "Send Message" action to notify your sales team.

The beauty of these actions is that they can be used in any combination and at any point in your agent's workflow. You're not limited to a predefined sequence - you can create custom workflows that perfectly match your business processes.

## 4. Using the Integration's API Tool Step (Advanced)

For advanced users who need more flexibility than the pre-built actions provide, Relevance AI offers direct API access to your connected integrations:

1. Create a new tool in the **Tools** section
2. Scroll down to **Tool Steps**
3. Click **"+ Add Step"**
4. Search for and select the integration's API tool step (e.g., "HubSpot API")
5. Select your connected account from the dropdown
6. Configure the API endpoint, method, and parameters
7. Save your tool

This advanced capability allows you to access any functionality exposed by the integration's API, even if there isn't a pre-built action for it yet. It's perfect for custom workflows that require specialized interactions with your external services.

## Creating Powerful Cross-Platform Workflows

The true power of integrations comes from combining multiple actions across different platforms into cohesive workflows. Here's an example of how you might use integrations to automate a lead nurturing process:

1. A new lead is created in HubSpot (trigger)
2. Your agent uses the Google Search action to research the lead's company
3. The agent uses the HubSpot action to add research notes to the contact record
4. The agent uses the Gmail action to send a personalized introduction email
5. The agent uses the Google Calendar action to schedule a follow-up task
6. The agent uses the Slack action to notify the sales team about the new lead

All of this happens automatically, without any manual intervention required. These workflows can dramatically increase efficiency, reduce human error, and ensure consistent execution of your business processes.

## Best Practices for Using Integrations

To get the most out of your integrations:

1. **Start with clear goals**: Define what you want to achieve with each integration before setting it up
2. **Begin with simple workflows**: Start with basic trigger-action combinations before building complex multi-step processes
3. **Test thoroughly**: Always test your integration workflows with sample data before deploying them
4. **Monitor performance**: Regularly check that your integrations are functioning as expected
5. **Secure your credentials**: Follow security best practices when connecting integrations, especially when using API keys
6. **Document your workflows**: Keep track of how data flows between systems for easier troubleshooting and optimization

## Related Features

[Agent Triggers](https://relevanceai.com/docs/build/agents/customise-agent/triggers) - Learn more about setting up events that automatically activate your agents based on specific conditions or actions.

[Tools](https://relevanceai.com/docs/get-started/core-concepts/tools) - Discover how to enhance your agents with specialized tools that extend their capabilities and allow them to perform specific tasks.

[Agent Configuration](https://relevanceai.com/docs/build/agents/agent-configuration) - Explore the complete process of setting up and customizing your AI agents to work with integrations and other features.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="How many integrations can I connect to my Relevance AI account?">
    There is no hard limit on the number of integrations you can connect. You can connect as many services as you need to support your workflows.
  </Accordion>

  <Accordion title="Can I connect multiple accounts from the same service?">
    Yes, you can connect multiple accounts from the same service. This is useful if you need to work with different accounts for different purposes or clients.
  </Accordion>

  <Accordion title="What happens if an integration temporarily fails?">
    The system will attempt to reconnect and, if unsuccessful, will notify you of the issue. Your agents will continue to function with their other capabilities while the integration is being restored.
  </Accordion>

  <Accordion title="Can I create custom integrations if what I need isn't available?">
    While Relevance AI continually expands its integration library, you can also use webhook integrations or API tool steps to connect with services that don't have dedicated integrations yet.
  </Accordion>

  <Accordion title="I added an API key on the Integrations page, how does it get used?">
    When you add an API key to a supported integration (e.g. Firecrawl, Lusha, Airtop), it's automatically used by all tool steps that require it. No extra configuration needed.
  </Accordion>
</AccordionGroup>


# Anthropic LLM models
Source: https://relevanceai.com/docs/integrations/llm-integrations/anthropic

Learn more about Anthropic LLM models and how to connect your AWS Bedrock or Google Cloud Platform account to access Anthropic models

Anthropic's Claude LLM models are focused on safe, reliable, and ethical AI responses. They are often better at reasoning and 'thoughtful' tasks.

<AccordionGroup>
  <Accordion title="What are they good for?">
    They're good for detailed explanations, structured outputs, and sensitive industries.
  </Accordion>

  <Accordion title="How much do they cost?">
    Credits charged per 1,000 tokens processed if you do not connect your AWS Bedrock account.
  </Accordion>

  <Accordion title="What native files can they process?">
    Anthropic models can process each of the following files:

    * .jpg
    * .jpeg
    * .png
    * .webp
    * .gif
  </Accordion>
</AccordionGroup>

## Setting Up AWS Bedrock with Relevance

Relevance supports two secure methods for connecting your AWS Bedrock account to access Anthropic models. Choose the option that best fits your organization's security requirements.

### Key Benefits

**Cross-Region Inference**: Ensure that you pick a model that has cross region inferencing available.

**Usage Oversight**: Both setup methods provide access to all supported Anthropic Claude models available in your selected AWS regions. You will be able to have increased cost control and insight into model usage.

### Option 1: IAM Credentials (Quick Setup)

**Best for**: Testing, proof-of-concepts, and organizations with flexible security policies.

#### Prerequisites

* AWS account with Anthropic models enabled in Bedrock
* Administrative access to create IAM users and policies

### Setup Steps

<Steps>
  <Step title="1. Create IAM Policy">
    In your AWS console, create a new policy with these permissions:

    ```json theme={null}
    {
      "Version": "2012-10-17",
      "Statement": [
        {
          "Effect": "Allow",
          "Action": [
            "bedrock:InvokeModel*"
          ],
          "Resource": [
            "arn:aws:bedrock:*:*:inference-profile/*",
            "arn:aws:bedrock:*::foundation-model/anthropic.*"
          ]
        }
      ]
    }
    ```
  </Step>

  <Step title="2. Create IAM user">
    * Create a new IAM user in your AWS account
    * Attach the policy created in step 1
    * Generate access keys for this user
  </Step>

  <Step title="3. Configure Relevance">
    In your Relevance dashboard:

    1. Navigate to **Integrations & API Keys** → **AWS Bedrock**
    2. Enter your credentials:
       * **Access Key ID**
       * **Secret Access Key**
       * **Region** (your preferred AWS region)
    3. Select your desired Claude model
  </Step>
</Steps>

### Option 2: Role Assumption (Enterprise Security)

**Best for**: Enterprise organisations requiring enhanced security, short-lived credentials, and detailed usage tracking.

#### Prerequisites

* Enterprise Relevance plan
* AWS account with Anthropic models enabled in Bedrock
* Administrative access to create IAM roles and policies

<Steps>
  <Step title="1. Contact your account team">
    Reach out to your Relevance Account Executive to initiate the role assumption setup. They will coordinate with our implementation team to exchange the necessary account details.
  </Step>

  <Step title="2. Create AWS role">
    Create an IAM role in your AWS account with this trust policy:

    ```json theme={null}
      {
        "Version": "2012-10-17",
        "Statement": [
          {
            "Effect": "Allow",
            "Principal": {
              "AWS": "arn:aws:iam::<RELEVANCE_ACCOUNT_ID>:root"
            },
            "Action": "sts:AssumeRole",
            "Condition": {
              "StringEquals": {
                "sts:ExternalId": "relevance:<YOUR_PROJECT_ID>"
              }
            }
          }
        ]
      }
    ```
  </Step>

  <Step title="3. Create and attach permissions policy">
    Create a policy with Bedrock permissions and attach it to your role:

    ```json theme={null}
      {
        "Version": "2012-10-17",
        "Statement": [
          {
            "Effect": "Allow",
            "Action": [
              "bedrock:InvokeModel*"
            ],
            "Resource": [
              "arn:aws:bedrock:*:*:inference-profile/*",
              "arn:aws:bedrock:*::foundation-model/anthropic.*"
            ]
          }
        ]
      }
    ```
  </Step>

  <Step title="4. Configure Relevance">
    In your Relevance dashboard:

    1. Navigate to **Integrations** → **AWS Bedrock**
    2. Enter your configuration:
       * **Region** (your preferred AWS region)
       * **Role ARN** (the ARN of the role you created)
  </Step>
</Steps>

<Note>
  If you run into any issues with either of these options, please reach out to your Account Executive or [our support team](https://relevanceai.com/docs/get-started/support).
</Note>

***

## AWS Bedrock Guardrails (Optional)

You can optionally configure AWS Bedrock Guardrails to apply content filtering and safety controls to your model invocations. Guardrails allow you to implement safeguards for your generative AI applications.

### Prerequisites

Before configuring guardrails in Relevance AI, you must:

1. Create a guardrail in AWS Bedrock console
2. Note the Guardrail ID or ARN
3. Optionally note the specific version you want to use
4. Enable Cross-region inferencing on the guardrail

<Note>
  For instructions on creating guardrails, see the [AWS Bedrock Guardrails documentation](https://docs.aws.amazon.com/bedrock/latest/userguide/guardrails.html).
</Note>

***

### Configuration

In the Relevance AI dashboard under **Project Settings > API Keys**, add the following:

| Key Name                         | Value                      | Required                                          |
| -------------------------------- | -------------------------- | ------------------------------------------------- |
| AWS Bedrock Guardrail Identifier | Your guardrail ID or ARN   | Yes (if using guardrails)                         |
| AWS Bedrock Guardrail Version    | Version number (e.g., "1") | No (defaults to DRAFT if you don't have versions) |

***

### IAM Policy Update

If you're using guardrails, add the following permission to your IAM policy:

```json theme={null}
{
  "Effect": "Allow",
  "Action": [
    "bedrock:ApplyGuardrail"
  ],
  "Resource": [
    "arn:aws:bedrock:<REGION>:<ACCOUNT ID>:guardrail/<GUARD RAIL ID>",
    "arn:aws:bedrock:<REGION>:<ACCOUNT ID>:guardrail-profile/<PROFILE ID>"
  ]
}
```

<Warning>
  Remember to add this permission to both the IAM user policy (Option 1) or the IAM role policy (Option 2) depending on which setup method you're using.
</Warning>

***

### Important considerations

#### Multi-Region Support

For optimal performance and availability, ensure your AWS policies include permissions for multiple regions. Use `*` in the region field or explicitly list all regions where you want Bedrock access.

#### Model Availability

Different Claude models are available in different AWS regions. Verify model availability in your chosen region using the [AWS Bedrock documentation](https://docs.aws.amazon.com/bedrock/latest/userguide/models-supported.html).

#### Security Precedence

If both IAM credentials and role ARN are configured, Relevance will automatically use the more secure role assumption method.

### Additional Resources

* [AWS Bedrock Inference Profiles Prerequisites](https://docs.aws.amazon.com/bedrock/latest/userguide/inference-profiles-prereq.html)
* [Cross-Region Inference Documentation](https://docs.aws.amazon.com/bedrock/latest/userguide/cross-region-inference.html)
* [Supported Models by Region](https://docs.aws.amazon.com/bedrock/latest/userguide/models-supported.html)

***

## Accessing Claude models using Vertex AI on Google Cloud Platform

If you use Google Cloud Platform, you can connect your Vertex AI API key to Relevance AI, which will allow you to use your Google Cloud Platform credits as opposed to your Relevance AI credits for Claude models.

You can do this by following these steps:

<Steps>
  <Step title="Enable your Vertex AI API key">
    In Google Cloud Console, enable the Vertex AI API by following the instructions [here](https://cloud.google.com/vertex-ai/docs/featurestore/setup#configure_project).
  </Step>

  <Step title="Enable the Claude models you wish to access">
    Enable the Claude model you want to use via [https://cloud.google.com/vertex-ai/generative-ai/docs/partner-models/claude](https://cloud.google.com/vertex-ai/generative-ai/docs/partner-models/claude) (click on the model card, then Enable)
  </Step>

  <Step title="Create a service account">
    Create a service account in the Google Cloud Console following the instructions at [https://cloud.google.com/iam/docs/service-accounts-create](https://cloud.google.com/iam/docs/service-accounts-create). **Make note of the project ID used to create the account.** The service account can have any name and description desired.

    <img alt="Screenshot of creating a Gemini service account" />
  </Step>

  <Step title="Add IAM role">
    Give the service account the Vertex AI Service Agent IAM role. This means the service account is permitted to use Vertex AI and call LLMs in Google Cloud.
  </Step>

  <Step title="Create private key">
    Create a JSON private key for the service account, and download the key file.
  </Step>

  <Step title="Extract the client_email and private_key fields">
    In the key file, extract the client\_email and private\_key fields exactly as they are written, removing the quotation marks.

    The client\_email should be an email address that normally ends with iam.gserviceaccount.com.

    The private\_key should be formatted as follows:

    ```shell theme={null}
    -----BEGIN PRIVATE KEY-----\nLOTS OF LETTERS AND NUMBERS HERE\n-----END PRIVATE KEY-----\n


    ```
  </Step>

  <Step title="Add details to Relevance AI">
    Log into Relevance AI and head to 'Integrations & API Keys'. Search for Google Cloud Platform, and add the Client Email and Private Key, and the Project ID from earlier in step 3.
  </Step>
</Steps>

Once your Google Cloud Platform details are added correctly, you will not be charged Relevance AI credits for Anthropic Claude models under the `gcp-vertexai` group of models.


# Azure LLM models
Source: https://relevanceai.com/docs/integrations/llm-integrations/azure

Learn more about Azure LLM models

Azure's LLM models are good for custom deployments hosted on Azure with higher context limits — ideal for large and detailed inputs.

<AccordionGroup>
  <Accordion title="What are they good for?">
    Enterprises needing private deployment, security, or very large tasks.
  </Accordion>

  <Accordion title="How much do they cost?">
    Credits charged per 1,000 tokens sent or received.
  </Accordion>
</AccordionGroup>


# Google's Gemini LLM models
Source: https://relevanceai.com/docs/integrations/llm-integrations/gemini

Learn more about Google's Gemini LLM models and how to access them via Google AI Studio and Google Cloud Platform

Google's Gemini LLM models are known for strong coding ability and task complexity handling (especially Google Gemini 2.5 models).

<AccordionGroup>
  <Accordion title="What are they good for?">
    Software development agents, complex task execution.
  </Accordion>

  <Accordion title="How much do they cost?">
    Credits charged per 1,000 tokens processed.
  </Accordion>

  <Accordion title="What native files can they process?">
    Gemini models can process each of the following files:

    * .pdf
    * .jpg
    * .jpeg
    * .png
    * .webp
    * .gif
    * .wav
    * .mp3
    * .aac
    * .ogg
    * .aiff
    * .flac
    * .mp4
    * .mpeg
    * .mov
    * .avi
    * .flv
    * .mpg
    * .webm
    * .wmv
  </Accordion>
</AccordionGroup>

## Understanding pricing & context caching

Gemini models automatically cache repeated input tokens (like agent instructions and tool definitions) in multi-turn conversations, charging them at **90% less** than uncached tokens. This happens automatically with no code changes required.

### Pricing example: Gemini 1.5 Pro

| Token Type                | Cost per 1,000 tokens |
| ------------------------- | --------------------- |
| **Uncached input tokens** | 0.315 credits         |
| **Cached input tokens**   | 0.0315 credits        |
| **Output tokens**         | 1.26 credits          |

**Example:** An agent with 19,000 tokens of instructions makes two calls:

* **First call:** 19,000 input (uncached) + 2,000 output = 8.51 credits
* **Second call:** 6,000 input (uncached) + 13,000 input (cached) + 1,500 output = 4.19 credits
* **Without caching**, the second call would cost 7.88 credits
* **Savings: 47% reduction** on the second call

### Why UI numbers may look different

<Callout icon="info-circle">
  The task credit breakdown displays **uncached input tokens only**. Cached tokens are charged separately at the lower rate but not shown in the breakdown. The **total cost is accurate** and includes both.
</Callout>

If you see "6,000 input tokens" in the UI, there may be additional cached tokens charged at 90% less. Your actual costs are **lower than expected** because of automatic caching.

<Tip>
  **No markup policy:** Relevance AI passes through Google's exact pricing, including all caching discounts. We do not add any markup to Vendor Credits.
</Tip>

For technical details about Gemini's context caching implementation, see [Gemini's Context Caching Documentation](https://ai.google.dev/gemini-api/docs/caching).

<Note>
  If you connect your own Google AI Studio or Google Cloud Platform credentials, you'll see the same caching behavior and cost savings directly in your Google billing.
</Note>

## Accessing Gemini models using Google AI Studio

If you use Google AI Studio, you can add your API key to Relevance AI, which will allow you to use your Google AI Studio credits as opposed to your Relevance AI credits.

You can add your Google AI Studio API key to Relevance AI by following these steps:

1. Generate your API key at [Google AI Studio](https://aistudio.google.com/welcome)
2. Log into Relevance AI and head to 'Integrations & API Keys'
3. Search for 'Google AI Studio Gemini API Key' and add your API key

Once you've successfully added your API key, you will not be charged Relevance AI credits when you use Google Gemini models in your LLM Tool steps and Agents.

## Accessing Gemini models using Vertex AI on Google Cloud Platform

If you use Google Cloud Platform, you can connect your Vertex AI API key to Relevance AI, which will allow you to use your Google Cloud Platform credits as opposed to your Relevance AI credits.

You can do this by following these steps:

<Steps>
  <Step title="Enable your Vertex AI API key">
    In Google Cloud Console, enable the Vertex AI API by following the instructions [here](https://cloud.google.com/vertex-ai/docs/featurestore/setup#configure_project).
  </Step>

  <Step title="Create a service account">
    Create a service account in the Google Cloud Console following the instructions at [https://cloud.google.com/iam/docs/service-accounts-create](https://cloud.google.com/iam/docs/service-accounts-create). **Make note of the project ID used to create the account.** The service account can have any name and description desired.

    <img alt="Screenshot of creating a Gemini service account" />
  </Step>

  <Step title="Add IAM role">
    Give the service account the Vertex AI Service Agent IAM role. This means the service account is permitted to use Vertex AI and call LLMs in Google Cloud.
  </Step>

  <Step title="Create private key">
    Create a JSON private key for the service account, and download the key file.
  </Step>

  <Step title="Extract the client_email and private_key fields">
    In the key file, extract the client\_email and private\_key fields exactly as they are written, removing the quotation marks.

    The client\_email should be an email address that normally ends with iam.gserviceaccount.com.

    The private\_key should be formatted as follows:

    ```shell theme={null}
    -----BEGIN PRIVATE KEY-----\\nLOTS OF LETTERS AND NUMBERS HERE\\n-----END PRIVATE KEY-----\\n


    ```
  </Step>

  <Step title="Add details to Relevance AI">
    Log into Relevance AI and head to 'Integrations & API Keys'. Search for Google Cloud Platform, and add the Client Email and Private Key, and the Project ID from earlier in step 2.
  </Step>

  <Step title="(Optional) Enter the region">
    Optionally, if the Gemini models are deployed into a particular Google Cloud Region, enter the region. Otherwise, leave it blank. If you enter a region that isn't deployed properly, calling models will fail. See [https://cloud.google.com/vertex-ai/generative-ai/docs/learn/locations](https://cloud.google.com/vertex-ai/generative-ai/docs/learn/locations) for a list of regions and what the string should look like.
  </Step>
</Steps>

Once your Google Cloud Platform details are added correctly, you will not be charged Relevance AI credits for Google Gemini models.


# OpenAI LLM models
Source: https://relevanceai.com/docs/integrations/llm-integrations/openai

Learn more about OpenAI LLM models

OpenAI's LLM models are known for advanced conversational abilities, creative writing, and broad general knowledge.

<AccordionGroup>
  <Accordion title="What are they good for?">
    Versatile agents, customer support, brainstorming.
  </Accordion>

  <Accordion title="How much do they cost?">
    Credits charged per 1,000 tokens used for prompts + replies if you do not connect your own OpenAI API key / account.
  </Accordion>

  <Accordion title="What native files can they process?">
    OpenAI models can process each of the following files:

    * .pdf
    * .jpg
    * .jpeg
    * .png
    * .webp
    * .gif
  </Accordion>
</AccordionGroup>

## Understanding pricing & implicit caching

OpenAI models automatically cache repeated input tokens (like agent instructions and tool definitions) in multi-turn conversations, charging them at **90% less** than uncached tokens. This happens automatically with no code changes required.

### Pricing example: GPT-4o

| Token Type                | Cost per 1,000 tokens |
| ------------------------- | --------------------- |
| **Uncached input tokens** | 0.63 credits          |
| **Cached input tokens**   | 0.063 credits         |
| **Output tokens**         | 5 credits             |

**Example:** An agent with 19,000 tokens of instructions makes two calls:

* **First call:** 19,000 input (uncached) + 2,000 output = 21.97 credits
* **Second call:** 6,000 input (uncached) + 13,000 input (cached) + 1,500 output = 12.10 credits
* **Without caching**, the second call would cost 19.47 credits
* **Savings: 38% reduction** on the second call

### Why UI numbers may look different

<Callout icon="info-circle">
  The task credit breakdown displays **uncached input tokens only**. Cached tokens are charged separately at the lower rate but not shown in the breakdown. The **total cost is accurate** and includes both.
</Callout>

If you see "6,000 input tokens" in the UI, there may be additional cached tokens charged at 90% less. Your actual costs are **lower than expected** because of automatic caching.

<Tip>
  **No markup policy:** Relevance AI passes through OpenAI's exact pricing, including all caching discounts. We do not add any markup to Vendor Credits.
</Tip>

For technical details about OpenAI's prompt caching implementation, see [OpenAI's Prompt Caching Documentation](https://platform.openai.com/docs/guides/prompt-caching).

<Note>
  If you connect your own OpenAI API key, you'll see the same caching behavior and cost savings directly in your OpenAI billing.
</Note>


# OpenRouter LLM models
Source: https://relevanceai.com/docs/integrations/llm-integrations/openrouter

Learn more about OpenRouter LLM models

OpenRouter models are a unified access point to a wide variety of open models like Meta’s LLaMA family, Google Gemini models, and others.

<AccordionGroup>
  <Accordion title="What are they good for?">
    Exploring experimental or emerging models.
  </Accordion>

  <Accordion title="How much do they cost?">
    Credits charged per 1,000 tokens used, depending on selected model. [Create an OpenRouter account here](https://openrouter.ai/).
  </Accordion>
</AccordionGroup>


# Agent skills
Source: https://relevanceai.com/docs/integrations/mcp/agent-skills

Give your AI coding assistant built-in knowledge of Relevance AI by cloning the agent skills repository.

The [Relevance AI agent skills](https://github.com/RelevanceAI/agent-skills) repository is a local reference that teaches your AI coding assistant how to work with Relevance AI. Clone it once and your assistant gets detailed context on agents, tools, workforces, knowledge, analytics, and evals — without needing to figure things out from scratch.

<Info>Agent skills work alongside the [MCP server](/docs/integrations/mcp/mcp-server). The MCP server gives your assistant the **ability** to call Relevance AI tools. Agent skills give it the **knowledge** to use them well.</Info>

***

## Quick start

<Steps>
  <Step title="Clone the agent skills repository">
    Choose the path that matches your AI coding assistant:

    <CodeGroup>
      ```bash Claude Code theme={null}
      git clone https://github.com/RelevanceAI/agent-skills ~/.claude/skills/relevance-ai
      ```

      ```bash OpenAI Codex theme={null}
      git clone https://github.com/RelevanceAI/agent-skills ~/.agents/skills/relevance-ai
      ```

      ```bash Cursor / VS Code / Others theme={null}
      git clone https://github.com/RelevanceAI/agent-skills .agents/skills/relevance-ai
      ```
    </CodeGroup>
  </Step>

  <Step title="Connect the MCP server">
    The agent skills provide context, but your assistant still needs the MCP server to actually interact with your Relevance AI project. Follow the setup for your client on the [MCP Server](/docs/integrations/mcp/mcp-server) page.
  </Step>

  <Step title="Verify the connection">
    Ask your AI assistant to list your agents or tools:

    ```
    > List all my Relevance AI agents
    ```

    If the assistant returns your agents, you're all set.
  </Step>
</Steps>

***

## What's included

The agent skills repository contains structured reference documentation that your AI assistant reads on-demand:

<CardGroup>
  <Card title="SKILL.md" icon="book">
    The main skill definition — covers all 46 MCP tools, critical usage rules, and workflow patterns your assistant should follow.
  </Card>

  <Card title="Reference docs" icon="folder-open">
    Detailed guides for agents, tools, workforces, knowledge, analytics, and evals that the assistant reads when working on specific tasks.
  </Card>
</CardGroup>

### Topics covered

<CardGroup>
  <Card title="Agents" icon="robot">
    Creating agents, system prompts, actions, memory, triggers, and troubleshooting.
  </Card>

  <Card title="Tools" icon="wrench">
    Building tools, transformations, OAuth configuration, versioning, and patterns.
  </Card>

  <Card title="Workforces" icon="users">
    Multi-agent orchestration concepts, setup, and debugging.
  </Card>

  <Card title="Knowledge" icon="database">
    Knowledge table operations — creating tables, adding rows, querying data.
  </Card>

  <Card title="Analytics" icon="chart-line">
    Usage metrics and reporting capabilities.
  </Card>

  <Card title="Evals" icon="check-double">
    Testing agent behavior with evaluation cases.
  </Card>
</CardGroup>

***

## Keeping skills up to date

The agent skills repository is a standard Git repo. Pull the latest changes periodically:

```bash theme={null}
cd ~/.agents/skills/relevance-ai  # or ~/.claude/skills/relevance-ai
git pull
```

***

## How it works

When your AI assistant encounters a Relevance AI task, it reads the relevant files from the agent skills repository to understand:

* **Which MCP tools to call** — the skill definition maps tasks to specific tool names
* **What parameters to use** — reference docs include required fields and correct formats
* **What pitfalls to avoid** — critical rules like "always fetch full agent config before updating" prevent common mistakes

This means your assistant can handle complex workflows (like creating an agent with tools, triggers, and a system prompt) correctly on the first try, rather than through trial and error.

<Tip>Agent skills follow the [Agent Skills open standard](https://github.com/anthropics/agent-skills). Any AI coding assistant that supports this convention can use them.</Tip>


# Claude Code plugin
Source: https://relevanceai.com/docs/integrations/mcp/claude-code

Use the Relevance AI Claude Code plugin to build and manage your agents directly from your terminal.

The [Relevance AI Claude Code plugin](https://github.com/RelevanceAI/cc-plugin) is the fastest way to build Relevance AI agents from your terminal with [MCP & Plugins](/docs/get-started/core-concepts/mcp-plugins). It bundles the MCP server along with built-in skills that teach Claude Code how to work with agents, tools, workforces, knowledge, analytics, and evals — all from your terminal.

## Quick start

<Steps>
  <Step title="Install the plugin">
    Install from the Claude Code marketplace:

    ```bash theme={null}
    claude plugin marketplace add RelevanceAI/cc-plugin
    claude plugin install relevance-ai@cc-plugin
    ```

    <Info>Requires Claude Code version **1.0.33** or later.</Info>
  </Step>

  <Step title="Authenticate">
    Run `/mcp` from within Claude Code, select the **relevance-ai** server, and click **Authenticate** to log in via your browser.

    Authentication is required only once.
  </Step>

  <Step title="Start building">
    You're ready to go. Try one of the example prompts below or describe what you want to build.
  </Step>
</Steps>

***

## What the plugin includes

<CardGroup>
  <Card title="MCP server connection" icon="server">
    Full access to your Relevance AI project's agents, tools, and knowledge.
  </Card>

  <Card title="Built-in skills" icon="graduation-cap">
    Pre-configured prompts and workflows for agents, tools, workforces, triggers, and evaluations.
  </Card>

  <Card title="Context awareness" icon="brain">
    Claude Code understands Relevance AI's data model, so you can describe what you want in plain language.
  </Card>
</CardGroup>

***

## Example prompts

Once authenticated, you can start building immediately. Here are some things to try:

<Tabs>
  <Tab title="Agents">
    ```
    > Create a BDR agent that qualifies inbound leads from HubSpot
    ```

    ```
    > Set up a workforce where a triage agent routes support tickets to specialist agents
    ```
  </Tab>

  <Tab title="Tools">
    ```
    > Build a tool that enriches a company URL with a one-paragraph summary
    ```

    ```
    > Create a tool that generates personalized cold emails from LinkedIn profiles
    ```
  </Tab>

  <Tab title="Evaluate & debug">
    ```
    > Pull the last 20 conversations for my Support Agent and identify failures
    ```

    ```
    > My BDR agent is letting unqualified leads through — review its instructions and fix it
    ```
  </Tab>
</Tabs>

<Tip>Use `/plan` in Claude Code to have Claude plan the work before executing — recommended for complex builds.</Tip>

***

## Working with multiple projects

If you work across multiple Relevance AI projects, add separate MCP server entries for each:

```bash theme={null}
claude mcp add relevance-project-1 --transport http https://mcp.relevanceai.com/
claude mcp add relevance-project-2 --transport http https://mcp.relevanceai.com/
```

Each entry authenticates independently against its own project. This lets you access tools and agents across all your projects without needing to log out and back in.

***

## Troubleshooting

<AccordionGroup>
  <Accordion title="Plugin not installing">
    * Ensure you are running Claude Code version **1.0.33** or later
    * Try running `claude plugin marketplace add RelevanceAI/cc-plugin` again
    * Check your internet connection
  </Accordion>

  <Accordion title="Authentication issues">
    * Run `/mcp` and select the **relevance-ai** server to re-authenticate
    * Make sure you have an active Relevance AI account with access to the project
    * Try removing and re-adding the plugin
  </Accordion>

  <Accordion title="Tools not appearing">
    * Verify you have completed authentication via `/mcp`
    * Check that your Relevance AI project has tools and agents configured
    * Try restarting Claude Code
  </Accordion>
</AccordionGroup>


# OpenAI Codex
Source: https://relevanceai.com/docs/integrations/mcp/codex

Connect OpenAI Codex to Relevance AI using the MCP server and agent skills.

[OpenAI Codex](https://openai.com/index/codex/) is OpenAI's cloud-based coding agent. You can connect it to Relevance AI by adding the MCP server and cloning the agent skills repository — giving Codex full access to your agents, tools, workforces, and knowledge.

<div>
  <iframe />
</div>

***

## Quick start

<Steps>
  <Step title="Add the MCP server">
    Register the Relevance AI MCP server with Codex:

    ```bash theme={null}
    codex mcp add relevance-ai --url https://mcp.relevanceai.com/
    ```
  </Step>

  <Step title="Authenticate">
    Log in to your Relevance AI account:

    ```bash theme={null}
    codex mcp login relevance-ai
    ```

    This opens your browser to authenticate. You only need to do this once per project.
  </Step>

  <Step title="Clone the agent skills">
    Give Codex built-in knowledge of Relevance AI by cloning the [agent skills repository](https://github.com/RelevanceAI/agent-skills):

    ```bash theme={null}
    git clone https://github.com/RelevanceAI/agent-skills ~/.agents/skills/relevance-ai
    ```

    This provides Codex with detailed context on how to work with agents, tools, workforces, knowledge, and more — so it can handle complex tasks correctly on the first try.

    <Tip>Learn more about what's included in the agent skills on the [Agent Skills](/docs/integrations/mcp/agent-skills) page.</Tip>
  </Step>

  <Step title="Start building">
    You're ready to go. Try one of the example prompts below or describe what you want to build.
  </Step>
</Steps>

***

## Example prompts

Once connected, you can start building immediately:

<Tabs>
  <Tab title="Agents">
    ```
    > Create a BDR agent that qualifies inbound leads from HubSpot
    ```

    ```
    > Set up a workforce where a triage agent routes support tickets to specialist agents
    ```
  </Tab>

  <Tab title="Tools">
    ```
    > Build a tool that enriches a company URL with a one-paragraph summary
    ```

    ```
    > Create a tool that generates personalized cold emails from LinkedIn profiles
    ```
  </Tab>

  <Tab title="Evaluate & debug">
    ```
    > Pull the last 20 conversations for my Support Agent and identify failures
    ```

    ```
    > My BDR agent is letting unqualified leads through — review its instructions and fix it
    ```
  </Tab>
</Tabs>

***

## Working with multiple projects

If you work across multiple Relevance AI projects, add a separate MCP server entry for each:

```bash theme={null}
codex mcp add relevance-project-1 --url https://mcp.relevanceai.com/
codex mcp login relevance-project-1

codex mcp add relevance-project-2 --url https://mcp.relevanceai.com/
codex mcp login relevance-project-2
```

Each entry authenticates independently against its own project.

***

## Troubleshooting

<AccordionGroup>
  <Accordion title="MCP server not connecting">
    * Ensure you have run `codex mcp add relevance-ai --url https://mcp.relevanceai.com/`
    * Check that authentication completed successfully with `codex mcp login relevance-ai`
    * Verify your internet connection and that `https://mcp.relevanceai.com/` is accessible
  </Accordion>

  <Accordion title="Authentication issues">
    * Make sure you have an active Relevance AI account
    * Check that you have access to the project you are trying to connect to
    * Try re-authenticating with `codex mcp login relevance-ai`
  </Accordion>

  <Accordion title="Tools not appearing">
    * Verify you have completed authentication
    * Check that your Relevance AI project has tools and agents configured
    * Try removing and re-adding the MCP server
  </Accordion>

  <Accordion title="Codex not using agent skills">
    * Confirm the agent skills are cloned to the correct path: `~/.agents/skills/relevance-ai`
    * Run `ls ~/.agents/skills/relevance-ai/SKILL.md` to verify the files exist
    * Pull the latest version: `cd ~/.agents/skills/relevance-ai && git pull`
  </Accordion>
</AccordionGroup>


# MCP client
Source: https://relevanceai.com/docs/integrations/mcp/mcp-client

Connect external MCP servers to your Relevance AI Agents, giving them access to Tools and resources from other services.

The MCP client integration lets you connect remote MCP servers to your Relevance AI Agents — either through preset connections (Notion, Canva, Atlassian) or by manually connecting any custom server URL. This gives your Agents access to Tools and resources hosted on external platforms without rebuilding them in Relevance AI.

<Warning>
  This page is about connecting external MCP servers **to** Relevance AI Agents. If you want to use Relevance AI **from** an AI client like Claude or Cursor, see [MCP & Plugins](/docs/get-started/core-concepts/mcp-plugins).
</Warning>

## Setting up an MCP connection for Agents

Agents can connect to remote MCP servers to run Tools and read resources exposed by those servers. You can choose from preset connections for popular services or connect your own custom MCP server.

<div>
  <iframe title="How to set up an MCP Server connection for an Agent" />
</div>

<Steps>
  <Step title="Open the Agent">
    Open the Agent you want to connect.
  </Step>

  <Step title="Go to the Tools tab">
    Navigate to the **Tools** tab in the Agent builder.
  </Step>

  <Step title="Add an MCP connection">
    Click **Add MCP**.
  </Step>

  <Step title="Choose your connection type">
    * **Preset connections**: Select from Notion, Canva, or Atlassian — follow the OAuth flow to authorize access and Tools are configured automatically.
    * **Custom connection**: Click **Connect my own**, add the server URL, label the connection, and enter any required authentication details.
  </Step>
</Steps>

<Note>
  The connection label you provide will appear in Tool descriptions as `(MCP: connection-name)`. This helps Agents distinguish between Tools from different connections, especially when using multiple instances of the same MCP server.
</Note>

Once connected, the Agent automatically fetches and displays the available Tools from the MCP server. Tools appear in the conversation task view and can be referenced in the Agent's core instructions.

<Warning>Local MCP server support (e.g. JSON configuration) is not available.</Warning>

## Connecting multiple instances of the same MCP server

You can connect the same MCP server URL multiple times within a project. Each connection operates independently with its own authentication credentials, label, and configuration. Tools from each connection display `(MCP: connection-name)` in their descriptions so Agents can distinguish which connection a Tool belongs to.

This is useful when you need different authentication contexts (e.g. separate user credentials for different data access), environment separation (staging vs production), or multi-tenant setups where each connection uses different tenant credentials.

### Best practices

<AccordionGroup>
  <Accordion title="Use descriptive connection labels">
    Choose labels that clearly indicate the purpose or context of each connection, such as `github-prod`, `github-staging`, or `linear-team-a`, `linear-team-b`.
  </Accordion>

  <Accordion title="Reference connections in Agent instructions">
    When building Agents that use multiple instances of the same server, include guidance in the Agent's core instructions about when to use Tools from each connection.
  </Accordion>
</AccordionGroup>

## Troubleshooting

<AccordionGroup>
  <Accordion title="Tools not loading from external MCP server">
    * Verify the MCP server URL is correct and the server is online
    * Check that any required authentication credentials are entered correctly
    * Ensure the external MCP server supports the Streamable HTTP transport
  </Accordion>

  <Accordion title="Agent cannot run external Tools">
    * Confirm the Tools are listed in the Agent's Tools tab after connecting
    * Check that the external MCP server has not revoked access or changed credentials
    * Try disconnecting and reconnecting the MCP server in the Agent builder
  </Accordion>

  <Accordion title="Connection timeout or errors">
    * The external MCP server may be temporarily unavailable — try again later
    * Check that the server URL does not require VPN or network access that Relevance AI cannot reach
    * For preset connections, try disconnecting and reconnecting to refresh the OAuth token
    * Contact the MCP server provider if the issue persists
  </Accordion>
</AccordionGroup>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What types of MCP servers can I connect?">
    You can connect any remote MCP server that supports the Streamable HTTP transport. This includes servers from services like Linear, Notion, GitHub, and any custom MCP servers you host.
  </Accordion>

  <Accordion title="Can I connect multiple MCP servers to one Agent?">
    Yes. You can add multiple MCP server connections to a single Agent, giving it access to Tools and resources from several external services at once.

    You can also connect the same MCP server URL multiple times with different labels and credentials. This is useful for environment separation (staging vs production), different authentication contexts, or multi-tenant setups. See [connecting multiple instances](#connecting-multiple-instances-of-the-same-mcp-server) for details.
  </Accordion>

  <Accordion title="Can I use locally hosted MCP servers?">
    No. Relevance AI only supports remote MCP servers accessible via a public URL. Local MCP servers running on your machine (e.g. via JSON configuration) are not supported.
  </Accordion>
</AccordionGroup>


# MCP server
Source: https://relevanceai.com/docs/integrations/mcp/mcp-server

Connect to Relevance AI from Claude Desktop, Cursor, VS Code, ChatGPT, and other MCP-compatible AI clients.

The Relevance AI MCP server gives any MCP-compatible AI client direct access to your agents, tools, and knowledge. Connect from the AI tools you already use and start building.

The MCP server is available at:

```
https://mcp.relevanceai.com/
```

<Tip>For Claude Code, we recommend using the [Relevance AI plugin](/docs/integrations/mcp/claude-code) instead of a manual MCP connection — it includes built-in skills and context that make the experience significantly better.</Tip>

<Warning>
  This page is about using Relevance AI **from** external AI clients. If you want to connect an external MCP server **to** a Relevance AI agent, see [MCP Client](/docs/integrations/mcp/mcp-client).
</Warning>

***

## Supported clients

<AccordionGroup>
  <Accordion title="Claude Desktop" icon="message-bot">
    1. Open Claude Desktop
    2. Go to **Settings** → **Connectors**
    3. Click **Add connector**
    4. Enter the server URL: `https://mcp.relevanceai.com/`
    5. Follow the authentication prompts to connect your Relevance AI project
  </Accordion>

  <Accordion title="Claude Web" icon="globe">
    1. Navigate to the [Connectors page](https://claude.ai/settings/connectors) in Claude.ai
    2. Click **Add connector**
    3. Enter the server URL: `https://mcp.relevanceai.com/`
    4. Follow the authentication prompts
  </Accordion>

  <Accordion title="ChatGPT" icon="comment">
    ChatGPT supports MCP servers through Developer Mode, available on Pro, Team, Enterprise, and Edu plans.

    1. Open ChatGPT **Settings**
    2. Go to **Connectors** → **Advanced** → **Developer Mode**
    3. Click **Add connector**
    4. Enter the server URL: `https://mcp.relevanceai.com/`
    5. Set Authentication to **OAuth** and follow the login flow

    Once connected, the Relevance AI tools will be available in both Chat and Deep Research modes.
  </Accordion>

  <Accordion title="Cursor" icon="i-cursor">
    1. Open Cursor Settings
    2. Navigate to the **MCP** tab
    3. Click **Add new MCP server**
    4. Use the following configuration in your `mcp.json`:

    ```json theme={null}
    {
      "mcpServers": {
        "relevance-ai": {
          "command": "npx",
          "args": ["-y", "mcp-remote", "https://mcp.relevanceai.com/"]
        }
      }
    }
    ```
  </Accordion>

  <Accordion title="VS Code / Copilot" icon="code">
    Add the following to your VS Code settings (`.vscode/mcp.json`):

    ```json theme={null}
    {
      "mcpServers": {
        "relevance-ai": {
          "command": "npx",
          "args": ["-y", "mcp-remote", "https://mcp.relevanceai.com/"]
        }
      }
    }
    ```
  </Accordion>

  <Accordion title="Windsurf" icon="wind">
    Add the following to your Windsurf MCP configuration:

    ```json theme={null}
    {
      "mcpServers": {
        "relevance-ai": {
          "command": "npx",
          "args": ["-y", "mcp-remote", "https://mcp.relevanceai.com/"]
        }
      }
    }
    ```
  </Accordion>

  <Accordion title="Codex" icon="robot">
    Register the MCP server, then log in:

    ```bash theme={null}
    codex mcp add relevance-ai --url https://mcp.relevanceai.com/
    codex mcp login relevance-ai
    ```

    See the [OpenAI Codex](/docs/integrations/mcp/codex) page for the full setup, including cloning the agent skills.
  </Accordion>

  <Accordion title="Zed" icon="pen-nib">
    Add the following to your Zed settings (`settings.json`):

    ```json theme={null}
    {
      "language_models": {
        "mcp": {
          "servers": {
            "relevance-ai": {
              "command": "npx",
              "args": ["-y", "mcp-remote", "https://mcp.relevanceai.com/"]
            }
          }
        }
      }
    }
    ```
  </Accordion>

  <Accordion title="v0 by Vercel" icon="bolt">
    Add the following MCP configuration in your v0 project settings:

    ```json theme={null}
    {
      "mcpServers": {
        "relevance-ai": {
          "command": "npx",
          "args": ["-y", "mcp-remote", "https://mcp.relevanceai.com/"]
        }
      }
    }
    ```
  </Accordion>

  <Accordion title="Claude Code (manual)" icon="terminal">
    If you prefer to add the MCP server directly without the plugin:

    ```bash theme={null}
    claude mcp add relevance-prod --transport http https://mcp.relevanceai.com/
    ```

    Once added, run `/mcp` from within Claude Code. You will see the new MCP server in the list. Select it to connect and follow the authentication steps.
  </Accordion>

  <Accordion title="Other clients" icon="plug">
    For any MCP-compatible client, use the server URL:

    ```
    https://mcp.relevanceai.com/
    ```

    If your client requires an `npx` command, use:

    ```bash theme={null}
    npx -y mcp-remote https://mcp.relevanceai.com/
    ```
  </Accordion>
</AccordionGroup>

***

## Authentication

When you first connect, you will be prompted to authenticate with your Relevance AI account. Authentication is **per project** — you will be connected to a specific Relevance AI project after logging in.

### Working with multiple projects

If you work across multiple Relevance AI projects, add a separate MCP server entry for each:

<CodeGroup>
  ```json Cursor theme={null}
  {
    "mcpServers": {
      "relevance-project-1": {
        "command": "npx",
        "args": ["-y", "mcp-remote", "https://mcp.relevanceai.com/"]
      },
      "relevance-project-2": {
        "command": "npx",
        "args": ["-y", "mcp-remote", "https://mcp.relevanceai.com/"]
      }
    }
  }
  ```

  ```json VS Code theme={null}
  {
    "mcpServers": {
      "relevance-project-1": {
        "command": "npx",
        "args": ["-y", "mcp-remote", "https://mcp.relevanceai.com/"]
      },
      "relevance-project-2": {
        "command": "npx",
        "args": ["-y", "mcp-remote", "https://mcp.relevanceai.com/"]
      }
    }
  }
  ```

  ```json Windsurf theme={null}
  {
    "mcpServers": {
      "relevance-project-1": {
        "command": "npx",
        "args": ["-y", "mcp-remote", "https://mcp.relevanceai.com/"]
      },
      "relevance-project-2": {
        "command": "npx",
        "args": ["-y", "mcp-remote", "https://mcp.relevanceai.com/"]
      }
    }
  }
  ```
</CodeGroup>

Each entry authenticates independently against its own project, so you can access tools and agents across all your projects without logging out and back in.

<Note>
  Alternatively, you can use a single connection and log out / log back in to switch projects — but the multi-connection approach above is preferred for convenience.
</Note>

***

## Access control

When you authenticate via OAuth, the level of access your MCP session gets depends on your project role. Some roles have a fixed mode; others can opt in to a restriction.

### Run-Only mode

Run-Only mode limits the MCP session to viewing and running existing agents. Write and destructive tools are removed from the tool list server-side — they are not present in the session at all, not merely hidden on the client.

The OAuth consent screen shows a **Run-Only mode** toggle for Member, Editor, and Admin users. The toggle defaults to off (full access). Viewer and Chat users do not see the toggle — their access mode is fixed and cannot be changed.

| Project role | MCP access mode | Configurable?                            |
| :----------- | :-------------- | :--------------------------------------- |
| Admin        | Full access     | Yes — opt into run-only at OAuth consent |
| Editor       | Full access     | Yes — opt into run-only at OAuth consent |
| Member       | Full access     | Yes — opt into run-only at OAuth consent |
| Viewer       | Run-only        | No — forced, cannot disable              |
| Chat         | Chat mode       | No — forced, cannot disable              |

<Note>
  To guarantee that a user's MCP session cannot create, edit, or delete assets: assign them the Viewer or Chat project role. The restriction is enforced server-side regardless of which MCP client they use.
</Note>

<Warning>
  Chat role users cannot set up dynamic (user-level) OAuth connections via MCP. If an agent requires dynamic authentication, assign the user a Member role in a dedicated project instead.
</Warning>

MCP access is bounded by your project role — the API enforces a ceiling so a session never grants more than the role allows.

***

## Add agent skills

The MCP server gives your AI assistant the ability to call Relevance AI tools, but it doesn't know *how* to use them well. For better results, pair it with the [agent skills](/docs/integrations/mcp/agent-skills) repository — a local reference that teaches your assistant how to work with agents, tools, workforces, knowledge, and more.

***

## Handling long-running agent executions

When triggering agents via MCP, you have two execution modes available depending on how long your agent takes to complete.

### Execution modes

<Tabs>
  <Tab title="Synchronous">
    The `relevance_trigger_agent` tool waits for the agent to finish and returns the result directly. It has a 120-second timeout, so use it for agents that complete quickly — single-step agents with minimal tool usage and no workforce nodes.
  </Tab>

  <Tab title="Asynchronous">
    The `relevance_trigger_agent_async` and `relevance_poll_agent_result` tools work together with no timeout limit. The trigger returns immediately with a conversation ID, which you then poll to check status and retrieve results. Use this for agents with workforce nodes, multi-step chains, external API calls, or any execution expected to exceed 2 minutes.
  </Tab>
</Tabs>

<Warning>
  If you encounter timeout errors with `relevance_trigger_agent`, switch to the async pattern. Agents with workforce nodes should always use async execution.
</Warning>

### Async execution workflow

<Steps>
  <Step title="Trigger the agent">
    Call `relevance_trigger_agent_async` with your agent parameters. This returns immediately with a `conversation_id`.
  </Step>

  <Step title="Poll for results">
    Use `relevance_poll_agent_result` with the `conversation_id` to check the execution status. Poll every 3-5 seconds until the status is `complete` or `failed`.
  </Step>

  <Step title="Retrieve the results">
    The status will be `working` or `in progress` while the agent is executing, `complete` when results are available, or `failed` if an error occurred. Once complete, the poll response contains the agent's output.
  </Step>
</Steps>

***

## Troubleshooting

<AccordionGroup>
  <Accordion title="Authentication issues">
    * Make sure you have an active Relevance AI account
    * Check that you have access to the project you are trying to connect to
    * Try removing and re-adding the MCP server connection
  </Accordion>

  <Accordion title="Tools not appearing">
    * Verify that you have tools configured in your Relevance AI project
    * Check that you are authenticated to the correct project
    * Try disconnecting and reconnecting the MCP server
  </Accordion>

  <Accordion title="Connection errors">
    * Ensure you have a stable internet connection
    * Check that `https://mcp.relevanceai.com/` is accessible from your network
    * Try removing and re-adding the MCP server connection in your client
    * Try clearing the auth cache: `rm -rf ~/.mcp-auth`
  </Accordion>

  <Accordion title="Agent execution timeouts">
    Timeout errors from `relevance_trigger_agent` mean your agent exceeded the 120-second synchronous limit. Switch to `relevance_trigger_agent_async` and `relevance_poll_agent_result` instead. See [handling long-running agent executions](#handling-long-running-agent-executions) for the full workflow. Agents with workforce nodes, multi-step chains, or complex workflows should always use the async pattern.
  </Accordion>
</AccordionGroup>

***

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What is MCP?">
    The Model Context Protocol (MCP) is an open standard that allows AI clients to connect to external tools and data sources. It provides a standardized way for AI assistants to access your Relevance AI workspace.
  </Accordion>

  <Accordion title="Is the MCP server free to use?">
    The MCP server itself is free. You will be billed for any Relevance AI usage (agent runs, tool executions, etc.) according to your plan.
  </Accordion>

  <Accordion title="Can I use multiple AI clients at the same time?">
    Yes. You can connect to the Relevance AI MCP server from as many clients as you like simultaneously. Each client authenticates independently.
  </Accordion>

  <Accordion title="Does authentication expire?">
    Authentication tokens may expire after a period of inactivity. If you are prompted to re-authenticate, simply follow the login flow again.
  </Accordion>

  <Accordion title="Can I restrict which tools are available via MCP?">
    Run-Only mode is the primary way to restrict MCP access. Assign a user the Viewer or Chat project role to force them into a restricted mode automatically — Viewers get run-only access (no write or destructive operations), and Chat users are limited to an even narrower set of tools. Member, Editor, and Admin users can opt into run-only mode from the toggle on the OAuth consent screen. For finer-grained control — such as exposing only a specific subset of agents — organize your tools across separate projects and authenticate each connection to the appropriate project. See [Access control](#access-control) for the full role matrix.
  </Accordion>
</AccordionGroup>


# Airtop
Source: https://relevanceai.com/docs/integrations/popular-integrations/airtop

Connect your Airtop browser automation account with Relevance AI to automate web interactions, data extraction, and browser-based tasks.

Connect your Airtop browser automation platform with Relevance AI to enable web automation capabilities for your AI agents and tools. The Airtop integration allows you to control browser windows, interact with web pages, extract content, and capture screenshots through automated workflows.

## Overview

The Airtop integration enables your AI agents and tools to perform browser automation tasks, including:

* Creating and managing browser windows
* Navigating to web pages and loading URLs
* Interacting with page elements (clicking, typing, scrolling)
* Extracting data from web pages
* Capturing screenshots
* Monitoring pages for specific conditions

This connection enables your agents to interact with websites at scale and with precision.

## Connecting the Integration

To connect the Airtop integration with Relevance AI:

1. Navigate to your Relevance AI dashboard and select **"Integrations & API Keys"** from the main menu.
2. Find and select **"Airtop"** from the available integrations.
3. Click **"Connect"** to begin the authorization process.
4. Provide your Airtop API key:
   * Log in to your Airtop account at [Airtop](https://www.airtop.ai/)
   * Navigate to the **API** section in your account dashboard
   * Click **"Create API Key"** or **"Generate API Key"**
   * Provide a descriptive name for your API key (e.g., "Relevance AI Integration")
   * Copy the generated API key immediately
5. Paste your API key into the Relevance AI connection form.
6. Click **"Save"** to establish the connection.

<Warning>
  API keys are only shown once when created. Make sure to copy and securely store your API key before closing the dialog. If you lose it, you'll need to create a new one.
</Warning>

Once connected, you'll see a confirmation message indicating successful integration. Your Airtop account will be available for use with Airtop browser automation tool steps.

## Use the integration's Tool steps

You can build custom tools that perform browser automation activities using Airtop Tool steps:

1. Create a new tool in Relevance AI.
2. Scroll down to Tool steps.
3. Search for "Airtop" to see all available Airtop browser automation tool steps.
4. Add the desired tool step (e.g., "Airtop - Create Window", "Airtop - Click Element", etc.).
5. Configure the parameters according to your needs.
6. This gives you access to Airtop's browser automation capabilities, allowing you to automate web interactions and data extraction.

<Info>You can find detailed information about each Airtop tool step in the [Airtop Browser Automation Tool steps](/docs/build/tools/tool-steps/airtop-browser-automation/window-management) documentation.</Info>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Do I need a paid Airtop account?">
    Check Airtop's current pricing plans. Some features may require a paid subscription, while basic browser automation may be available on free tiers. Refer to Airtop's documentation for the most current information.
  </Accordion>

  <Accordion title="What happens if I revoke my API key in Airtop?">
    If you revoke or delete your API key in Airtop, any Relevance AI tools using that integration will stop working. You'll need to create a new API key and update it in Relevance AI to restore functionality.
  </Accordion>

  <Accordion title="Can I use Airtop browser automation in agents?">
    Yes, once set up, you can use Airtop browser automation tool steps in both Tools and Agents. Agents can call these steps to automate web interactions and data extraction.
  </Accordion>
</AccordionGroup>


# Apollo
Source: https://relevanceai.com/docs/integrations/popular-integrations/apollo

Apollo is a B2B sales intelligence and engagement platform for finding, enriching, and engaging prospects.

Apollo is a B2B sales intelligence and engagement platform with a large database of contacts and companies. With Relevance AI's Apollo integration, you can build agents that search for prospects, enrich contact and organization data, and manage deals — all using your connected Apollo account.

<Warning>
  The Apollo integration in Relevance AI was built by Relevance AI, and is therefore supported by our team, not Apollo. If you have a question or issue with using Apollo in Relevance AI, please [reach out to our support team](/docs/get-started/support). If you have a question or issue that is only about Apollo, you can reach out to Apollo support directly.
</Warning>

## Connect the integration

1. Go to the **Integrations** page in the sidebar of your Relevance AI dashboard.
2. Click on **Apollo** from the available integrations.
3. Click **Add Integration**.
4. Enter your Apollo API key when prompted.
5. Once authenticated, your Apollo account will appear as a connected integration.

<Tip>
  You can find your Apollo API key in your Apollo account under **Settings → Integrations → API**. You need at least a Basic plan to access the API.
</Tip>

## Tool steps for Apollo

The Apollo integration provides actions for prospecting, contact management, deal tracking, and data enrichment. These actions can be used as tool steps in your agents' workflows.

### Contact and account management

<CardGroup>
  <Card title="Search contacts" icon="magnifying-glass">
    Search Apollo's database for contacts matching criteria like job title, company, location, or industry.
  </Card>

  <Card title="Create contact" icon="user-plus">
    Create a new contact record in Apollo.
  </Card>

  <Card title="Update contact" icon="user-pen">
    Update an existing contact record with new information.
  </Card>

  <Card title="Search accounts" icon="building-magnifying-glass">
    Search for company accounts in Apollo by name, domain, industry, or other criteria.
  </Card>

  <Card title="Create account" icon="building-circle-arrow-right">
    Create a new company account record in Apollo.
  </Card>

  <Card title="Update account" icon="building-circle-check">
    Update an existing company account record.
  </Card>
</CardGroup>

### Deal management

<CardGroup>
  <Card title="List deals" icon="list">
    Retrieve deals from Apollo, with optional filters by stage, owner, or account.
  </Card>

  <Card title="Create deal" icon="circle-plus">
    Create a new deal record in Apollo and associate it with a contact or account.
  </Card>

  <Card title="Update deal" icon="pen-to-square">
    Update an existing deal's stage, value, or other attributes.
  </Card>
</CardGroup>

### Person and organization enrichment

<CardGroup>
  <Card title="Enrich person" icon="address-card">
    Enrich a contact record with data from Apollo's database, including phone numbers, email addresses, job title, and social profiles.
  </Card>

  <Card title="Enrich organization" icon="building">
    Enrich a company record with firmographic data including employee count, revenue, industry, and technology stack.
  </Card>
</CardGroup>

<Tip>
  Type "Apollo" in the tool step search bar to see all available Apollo actions when building your tools.
</Tip>

## Phone number retrieval

The **Enrich Person** tool step supports phone number retrieval via Apollo's `reveal_phone_number` and `run_waterfall_phone` options. When you enable these options, Relevance AI handles the underlying webhook flow automatically and returns the phone numbers directly in the enrichment response — no additional configuration required on your end.

This means you can use Apollo's phone number enrichment the same way as any other enrichment field, and the result will include phone numbers when Apollo can locate them.

## Use the integration's API tool step (Advanced)

In addition to the pre-built tool steps, you can make custom API calls to any Apollo endpoint using the Apollo API Call tool step.

### How to use the Apollo API Call tool step

<Steps>
  <Step title="Create or open a tool">
    Create a new tool in Relevance AI, or open an existing tool you want to add Apollo functionality to.
  </Step>

  <Step title="Add the Apollo API Call tool step">
    1. Scroll down to **Tool steps**.
    2. Search for **Apollo API Call** in the tool step search bar.
    3. Add it to your workflow.
  </Step>

  <Step title="Select your Apollo account">
    Select your connected Apollo account from the dropdown.
  </Step>

  <Step title="Configure the API endpoint">
    Set the HTTP method, endpoint path, and any request body parameters. Refer to [Apollo's API documentation](https://apolloio.github.io/apollo-api-docs/) for available endpoints and required fields.
  </Step>

  <Step title="Test your configuration">
    Run a test to confirm the call returns the expected data before deploying your agent.
  </Step>
</Steps>

## Example use cases

<AccordionGroup>
  <Accordion title="Automated prospect research" icon="magnifying-glass-chart">
    Build an agent that takes an inbound lead's email or LinkedIn URL, uses Apollo to enrich the contact with job title, phone number, company details, and funding data, then posts a research summary to Slack or updates your CRM before the sales rep's first call.
  </Accordion>

  <Accordion title="ICP-based list building" icon="list-check">
    Create an agent that searches Apollo's database for contacts matching your ideal customer profile — filtering by industry, company size, seniority, and location — then exports the list to a Google Sheet or pushes it into HubSpot as new leads.
  </Accordion>

  <Accordion title="Deal pipeline automation" icon="diagram-project">
    Deploy an agent that monitors your CRM for trigger events (e.g., a contact replies to an email) and automatically creates or updates the corresponding deal in Apollo, keeping both systems in sync without manual data entry.
  </Accordion>

  <Accordion title="Enrichment before outbound sequences" icon="envelope-open-text">
    Build an agent that enriches a batch of contacts with Apollo before they enter an outreach sequence, ensuring each contact has a verified phone number and current job title. Contacts missing key data can be flagged for manual review.
  </Accordion>
</AccordionGroup>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Do I need an Apollo account to use this integration?">
    Yes, you need an active Apollo account with API access. Apollo offers a free plan with limited credits, and paid plans with higher limits. You'll need to generate an API key from your Apollo account settings to connect it to Relevance AI.
  </Accordion>

  <Accordion title="How do I get my Apollo API key?">
    1. Log in to your Apollo account.
    2. Go to **Settings → Integrations → API**.
    3. Copy your existing API key or generate a new one.
    4. Use this key when connecting Apollo to Relevance AI.

    Keep your API key secure and do not share it publicly.
  </Accordion>

  <Accordion title="Does phone number enrichment require extra setup?">
    No. When you use the `reveal_phone_number` or `run_waterfall_phone` options in the Enrich Person tool step, Relevance AI manages the webhook flow automatically. Phone numbers are returned directly in the enrichment response when Apollo can locate them.
  </Accordion>

  <Accordion title="What happens if Apollo doesn't find a match for an enrichment request?">
    If Apollo cannot find data for a contact or organization, the enrichment response will be empty or will indicate no match. Design your agent to handle these cases — for example, by trying alternative identifiers, using a fallback data source, or flagging the record for manual review.
  </Accordion>

  <Accordion title="Does Apollo charge credits for enrichment?">
    Yes, Apollo's enrichment endpoints consume credits from your Apollo plan. Phone number reveal requests consume additional credits depending on your plan. Monitor your Apollo credit usage in your Apollo account dashboard.
  </Accordion>
</AccordionGroup>


# Canva
Source: https://relevanceai.com/docs/integrations/popular-integrations/canva

Elevate your visual content creation capabilities with the Canva integration for Relevance AI. 

## Connect the Integration

Setting up the Canva integration with Relevance AI is straightforward:

1. Navigate to the Integrations section in your Relevance AI dashboard
2. Find and select "Canva" from the available integrations
3. Click "Connect"
4. You'll be redirected to Canva to authorize the connection
5. Log in to your Canva account (or create one if needed)
6. Review and approve the permissions requested
7. Once authorized, you'll be redirected back to Relevance AI with the integration active

After connecting, your Relevance AI agents will be able to interact with your Canva account, accessing templates, creating designs, and managing your Canva projects.

## Setting up Triggers

The Canva integration can be configured as a trigger for your AI agents, allowing them to automatically respond to events in your Canva account. This creates powerful workflows where design activities can initiate AI-driven processes.

To set up a Canva trigger *(coming soon)*:

1. Navigate to your agent's configuration page (”Agent profile”)
2. Under "Triggers" choose "Canva" from the available trigger sources
3. Select the specific event you want to trigger your agent:
   * New design created
   * Design shared with you
   * Design comment added
   * Design published
4. Configure any additional trigger parameters
5. Save your trigger configuration

With triggers properly configured, your agent can automatically perform tasks like:

* Generating social media copy when a new social post design is created

* Notifying team members when designs are ready for review

* Creating documentation when marketing materials are finalized

* Scheduling content distribution when designs are published

## Tools & Tool Steps

The Canva integration provides a rich set of actions that your agents can use to interact with Canva. These actions can be incorporated into your agent's workflows to automate design-related tasks.

### Available Canva Actions

Your agents can leverage numerous Canva actions, including:

* **Create Design**: Generate new designs from templates or from scratch
* **Edit Design**: Modify existing designs with new text, images, or elements
* **Export Design**: Download designs in various formats (PNG, JPG, PDF)
* **List Templates**: Browse available Canva templates by category
* **Add Brand Kit**: Upload and manage brand assets like logos and colors
* **Schedule Posts**: Plan social media content publication
* **Manage Folders**: Organize designs into project folders
* **Share Design**: Collaborate with team members on designs
* **Add Comments**: Provide feedback on designs
* **Search Elements**: Find graphics, photos, and other design elements

These actions give your agents the ability to handle complex design workflows without manual intervention. For example, an agent could:

1. Create a new social media post from a template
2. Customize it with your brand colors and logo
3. Add text based on marketing copy it generated
4. Export the design in multiple formats
5. Share it with your marketing team for approval

The integration supports a wide range of design types including social media posts, presentations, flyers, business cards, infographics, and many more. Your agents can work with any design format supported by Canva.

## Use the Integration's API Tool Step (Advanced)

For advanced users who need more customized functionality, the Canva API tool step provides direct access to Canva's API capabilities:

1. Create a new tool in Relevance AI

2. Scroll down to Tool-steps

3. Add "Canva API" tool-step

4. Select your connected Canva account in the dropdown

5. Configure the API request with the appropriate endpoint, method, and parameters

6. Test and save your custom tool

Using the API directly allows for more sophisticated interactions with Canva, such as:

* Batch processing multiple designs
* Creating custom design workflows
* Implementing advanced design automation
* Integrating with other services in your workflow

## Related Features

[Knowledge Base Integration](https://relevanceai.com/docs/build/knowledge/create-knowledge) - Store design guidelines, brand assets, and marketing strategies that your agents can reference when creating Canva designs.

[Slack Integration](https://relevanceai.com/docs/integrations/slack) - Connect your design workflow to Slack for seamless team collaboration. Agents can notify team members when designs are ready for review or automatically share completed designs in relevant channels.

[Email Integration](https://relevanceai.com/docs/integrations/outlook) - Enable your agents to send designs directly to clients or team members via email, streamlining the approval and delivery process.

## Best Practices

* **Maintain Brand Consistency**: Store your brand guidelines in a knowledge base for your agents to reference when creating designs
* **Template Organization**: Create and categorize templates in Canva for different use cases to help your agents select the right starting point
* **Clear Instructions**: When requesting designs, provide specific details about the intended audience, purpose, and key messages
* **Approval Workflows**: Set up multi-step approval processes for important designs
* **Asset Management**: Keep your brand assets organized in Canva for easy access by your agents

By integrating Canva with Relevance AI, you can automate design workflows, maintain brand consistency, and scale your visual content creation—all while leveraging the intelligence of your AI agents to deliver compelling visual communications.


# Chrome Extension
Source: https://relevanceai.com/docs/integrations/popular-integrations/chrome-extension

Extract content from any web page and send it directly to an AI agent chat in Relevance AI.

The Relevance AI Chrome Extension lets you extract content from any web page and send it directly to an AI agent chat, without copy-pasting.

<Info>Install the extension from the [Chrome Web Store](https://chromewebstore.google.com/detail/relevance-ai/oeejdfhhhceikclacdlikdgfenhehgcp).</Info>

### Supported browsers

<CardGroup>
  <Card title="Chrome" icon="chrome" />

  <Card title="Edge" icon="edge" />

  <Card title="Brave" icon="brave" />

  <Card title="Arc" icon="globe" />
</CardGroup>

Any Chromium-based browser that supports Manifest V3 is compatible. Firefox is not supported.

## Installation and setup

### Installing the extension

1. Open the [Chrome Web Store listing](https://chromewebstore.google.com/detail/relevance-ai/oeejdfhhhceikclacdlikdgfenhehgcp) in a supported browser.
2. Click **Add to Chrome** and confirm the prompt.
3. The Relevance AI icon appears in your browser toolbar. Pin it for easy access.

### Signing in

<img alt="Chrome extension sign in screen" />

Click the extension icon in your toolbar, then click **Sign in to send to chat**. An OAuth popup opens through `app.relevanceai.com/auth` where you authorize the extension with your Relevance AI account. Tokens are stored locally and refreshed automatically, so you stay signed in across browser sessions.

### Selecting a project and region

After signing in, click your user avatar in the extension popup header to open a dropdown where you can switch projects and regions. Your selection persists across sessions.

## Using the extension

### Main workflow

<Steps>
  <Step title="Browse to any web page">
    Navigate to the page you want to extract content from.
  </Step>

  <Step title="Open the extension">
    Click the toolbar icon, or use the keyboard shortcut: **Mac:** `Cmd+Shift+S` / **Windows/Linux:** `Ctrl+Shift+S`
  </Step>

  <Step title="Review extracted content">
    The extension automatically extracts content from the page and displays it in an editable text area. Edit as needed.
  </Step>

  <Step title="Send to chat">
    Click **Start new chat** to open a new agent conversation with the content as the first message.
  </Step>

  <Step title="Continue in Relevance AI">
    Click **Open in Chat** to jump directly to the conversation in Relevance AI.
  </Step>
</Steps>

### Use the extension without opening the popup

<div>
  <iframe title="Using the Chrome Extension right-click context menu" />
</div>

You can also trigger the extension without opening the popup:

1. Right-click anywhere on a page, or right-click on selected text.
2. Choose **Send to Chat** from the context menu.
3. The extension popup opens with content pre-loaded.

<Tip>While content is pending, a badge notification ("1") appears on the extension icon as a reminder.</Tip>

## Content extraction

The extension uses a multi-stage extraction pipeline that selects the appropriate method based on the page type.

<AccordionGroup>
  <Accordion title="LinkedIn profiles" icon="linkedin">
    Extracts name, headline, company, location, about section, experience, education, skills, and profile URL. Handles LinkedIn's single-page app navigation, including iframe-based content.
  </Accordion>

  <Accordion title="Articles and blog posts" icon="newspaper">
    Uses Mozilla Readability (the same library as Firefox Reader View) to extract the title, author, published date, and body text, plus the canonical URL. Works on most news sites and blog platforms.
  </Accordion>

  <Accordion title="Generic web pages" icon="globe">
    Extracts the page title, meta description, body text (up to 4,000 characters), and structured data including JSON-LD and Open Graph tags.
  </Accordion>

  <Accordion title="Selected text" icon="text">
    If you have text selected on the page when you open the extension, that selection is included alongside the page extraction.
  </Accordion>
</AccordionGroup>

Extracted content is formatted as markdown and wrapped with metadata before being sent to the agent. When you click **Start new chat**, the extension calls the TriggerWorkforceChat API to create a new conversation, and the extracted content becomes the first message.

<Warning>The extension cannot extract content from browser-internal pages such as `chrome://`, `about:`, or other extension pages.</Warning>

## Permissions and network access

<CardGroup>
  <Card title="activeTab">
    Read the current tab's URL and extract page content. Only used when you actively open the extension or trigger it via the context menu.
  </Card>

  <Card title="contextMenus">
    Adds the **Send to Chat** option to your browser's right-click menu.
  </Card>

  <Card title="storage">
    Stores authentication tokens, your project selection, and other preferences in local browser storage. Nothing is transmitted to a third party.
  </Card>

  <Card title="tabs">
    Detects tab URL changes to handle LinkedIn's single-page app navigation and the OAuth callback after sign-in.
  </Card>
</CardGroup>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What if a page won't extract content properly?">
    Try clicking the refresh button in the content area to re-run extraction. For pages that extract poorly, you can manually edit the content in the text area before sending.
  </Accordion>

  <Accordion title="How do I sign out?">
    Click your user avatar in the extension popup header and select **Sign out** from the dropdown.
  </Accordion>

  <Accordion title="What happens to my extracted content?">
    Content is sent directly to your Relevance AI account as the first message in a new agent chat. It is not stored by the extension beyond what is needed to display it in the popup before you send.
  </Accordion>

  <Accordion title="Can I use this with any agent?">
    The extension creates a new chat session in your selected project. The agent that handles the chat depends on your project configuration in Relevance AI.
  </Accordion>
</AccordionGroup>


# Confluence Cloud
Source: https://relevanceai.com/docs/integrations/popular-integrations/confluence

Confluence is a powerful collaboration platform that helps teams organize, create, and share knowledge.

Confluence is a powerful collaboration platform that helps teams organize, create, and share knowledge. With Relevance AI's Confluence integration, you can seamlessly connect your Confluence workspace to your AI agents, enabling them to access, create, and update Confluence content, making your knowledge management workflows more efficient and intelligent.

<Warning>
  The Confluence integration in Relevance AI was built by Relevance AI, and is therefore supported by our team, not Atlassian (Confluence's parent company). If you have a question / issue with using Confluence in Relevance AI, please [reach out to our support team](/docs/get-started/support). If you have a question / issue that is only about Confluence, you can [reach out to Atlassian support](https://support.atlassian.com/contact/).
</Warning>

## Connect the integration

Connecting your Confluence account to Relevance AI is a straightforward process:

1. Go to the **Integrations & API Keys** page in the sidebar of your Relevance AI dashboard.
2. Click on "Confluence" from the available integrations.
3. Click on the "Add Integration" button.
4. In the pop-up window, sign into your Atlassian account.
5. Grant the necessary permissions for Relevance AI to access your Confluence workspace.
6. Once authenticated, your Confluence account will appear as a connected integration.

## Tool steps for Confluence

The Confluence integration provides a rich set of actions that your agents can use to interact with your Confluence workspace. These actions can be incorporated into your agent's workflows as tool steps, enabling sophisticated knowledge management capabilities.

<Snippet />

## Use the integration's API tool step (Advanced)

In addition to the pre-built actions available in the tool directory, you can build custom tools that perform Confluence-specific activities using the Confluence API Call tool step:

1. Create a new tool in Relevance AI.
2. Scroll down to Tool-steps.
3. Add the Confluence API Tool step.
4. Select your connected Confluence account in the dropdown.
5. Configure the API endpoint, method, and parameters according to your needs.

This advanced approach gives you full access to the Confluence REST API, allowing you to implement custom functionality beyond what's available in the standard actions.

<Warning>
  **Check API Documentation for Required Headers**

  The Confluence API Call Tool step does **not** automatically add headers to your requests. Many Confluence REST API v2 endpoints require the `accept: application/json` header to function properly.

  **Always check the [Confluence API documentation](https://developer.atlassian.com/cloud/confluence/rest/v2/intro/) for your specific endpoint** to see which headers are required, and add them manually in the Headers section of the API Call Tool step.
</Warning>

### Adding Required Headers

When using the Confluence API Call Tool step, you must manually add any headers required by the endpoint you're calling. The tool does not add these automatically.

<Steps>
  <Step title="Check the API documentation">
    Before configuring your API call, visit the [Confluence REST API documentation](https://developer.atlassian.com/cloud/confluence/rest/v2/intro/) and find your specific endpoint. Look for the "Request" section to see which headers are required.

    Many endpoints require the `accept: application/json` header, but requirements vary by endpoint.
  </Step>

  <Step title="Add the API Call Tool step">
    In your tool configuration, add the Confluence API Call Tool step and select your connected Confluence account.
  </Step>

  <Step title="Configure the endpoint">
    Set up your API endpoint and method. For example, to get a page by ID:

    * **Method**: GET
    * **Endpoint**: `/wiki/api/v2/pages/{id}?body-format=storage`
  </Step>

  <Step title="Add required headers manually">
    In the Headers section of the API Call Tool step, add any headers specified in the API documentation. For most Confluence REST API v2 endpoints, this includes:

    ```json theme={null}
    {
      "accept": "application/json"
    }
    ```

    This header tells the Confluence API to return responses in JSON format. Without it, many endpoints will fail with errors.
  </Step>
</Steps>

<Info>
  **Why is this header needed?**

  The `accept: application/json` header tells the Confluence API what format you want the response in. Most REST API v2 endpoints require this header to return data in JSON format, which the tool can then process. Without this header, the API may return an error or data in an unexpected format.
</Info>

### Example API Call Configurations

Here are examples of calling the Confluence API with both object and array request bodies.

<Tabs>
  <Tab title="Object Body Example">
    **Use case**: Get a page by ID

    This endpoint requires the `accept: application/json` header according to the [Atlassian documentation](https://developer.atlassian.com/cloud/confluence/rest/v2/api-group-page/#api-pages-id-get).

    **API Method**: GET

    **Endpoint**: `/wiki/api/v2/pages/{id}?body-format=storage`

    **Headers** (must be added manually):

    ```json theme={null}
    {
      "accept": "application/json"
    }
    ```

    **Path Parameters**:

    * `id`: The ID of the Confluence page you want to retrieve

    **Expected Response**:

    ```json theme={null}
    {
      "id": "123456",
      "status": "current",
      "title": "Page Title",
      "spaceId": "789012",
      "body": {
        "storage": {
          "value": "<p>Page content here</p>",
          "representation": "storage"
        }
      }
    }
    ```
  </Tab>

  <Tab title="Array Body Example">
    **Use case**: Bulk update page properties

    Some Confluence endpoints require array bodies for bulk operations. This example shows updating multiple page properties at once.

    **API Method**: PUT

    **Endpoint**: `/wiki/api/v2/pages/{id}/properties`

    **Headers** (must be added manually):

    ```json theme={null}
    {
      "accept": "application/json",
      "content-type": "application/json"
    }
    ```

    **Body** (array format):

    ```json theme={null}
    [
      {
        "key": "status",
        "value": "reviewed"
      },
      {
        "key": "priority",
        "value": "high"
      },
      {
        "key": "last-updated-by",
        "value": "automation-bot"
      }
    ]
    ```

    **Path Parameters**:

    * `id`: The ID of the Confluence page

    <Info>
      This endpoint demonstrates the array body support added in PR #12797. The API Call tool step now accepts both JSON objects `{}` and JSON arrays `[]` as top-level request bodies.
    </Info>
  </Tab>
</Tabs>

<Tip>
  **Best Practice**: Always consult the [Confluence API documentation](https://developer.atlassian.com/cloud/confluence/rest/v2/intro/) for your specific endpoint before configuring your API call. The documentation will specify all required headers, parameters, and request body formats (including whether an object or array is expected).
</Tip>

## Example use cases

Here are some ways you can leverage the Confluence integration with your agents:

1. **Documentation Assistant**: Create an agent that automatically updates technical documentation in Confluence when new features are released or code changes are made.
2. **Knowledge Summarizer**: Build an agent that monitors specific Confluence spaces and creates executive summaries of new content added during the week.
3. **Content Organizer**: Deploy an agent that automatically adds appropriate labels to new Confluence pages based on their content analysis.
4. **Meeting Notes Processor**: Create an agent that takes meeting notes, formats them according to your team's template, and posts them to the relevant Confluence space.
5. **Knowledge Gap Identifier**: Build an agent that analyzes your Confluence content and identifies areas where documentation is missing or outdated.

## Related Features

* Knowledge Integration - Use Confluence as a knowledge source for your agents, enabling them to access and reference your team's documentation when responding to queries.
* Slack Integration - Combine Confluence and Slack integrations to create powerful workflows where agents can update documentation and notify team members about changes.
* Document Processing - Leverage document processing tools alongside Confluence integration to extract, analyze, and organize information from various sources into your knowledge base.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What permissions does the Confluence integration require?">
    The integration requires permissions to read and write content, access spaces, and manage page properties in your Confluence workspace. You can review the specific permissions during the authentication process.
  </Accordion>

  <Accordion title="Can I restrict my Agent to only access certain spaces in Confluence?">
    Yes, you can configure your tools and triggers to only interact with specific spaces by setting the appropriate filters and parameters.
  </Accordion>

  <Accordion title="Why am I getting errors when using the Confluence API Call Tool?">
    The most common issue is missing required headers. The API Call Tool does **not** automatically add headers to your requests.

    To troubleshoot:

    1. **Check the API documentation**: Visit the [Confluence API documentation](https://developer.atlassian.com/cloud/confluence/rest/v2/intro/) for your specific endpoint and verify which headers are required. Most REST API v2 endpoints require `accept: application/json`.
    2. **Add headers manually**: In the Headers section of the API Call Tool step, add all required headers as specified in the documentation.
    3. **Verify your connection**: Ensure your Confluence account is properly connected in Relevance AI.
    4. **Check permissions**: Confirm you have the necessary permissions for the endpoint you're calling.
    5. **Validate the endpoint**: Make sure your endpoint URL and parameters are correctly formatted according to the API documentation.
    6. **Use the correct API version**: The REST API v2 is recommended for new integrations.
  </Accordion>

  <Accordion title="Do I need to add the accept header for all Confluence API endpoints?">
    Not necessarily. While many Confluence REST API v2 endpoints require the `accept: application/json` header, requirements vary by endpoint.

    **Always check the official [Confluence API documentation](https://developer.atlassian.com/cloud/confluence/rest/v2/intro/)** for your specific endpoint to see which headers are required. The documentation will clearly specify all required headers in the "Request" section of each endpoint.

    Remember: The API Call Tool does not add any headers automatically, so you must manually add all required headers specified in the documentation.
  </Accordion>

  <Accordion title="When should I use array bodies vs. object bodies in Confluence API calls?">
    The request body format depends on what the specific Confluence API endpoint expects:

    * **Object bodies** `{}` are used for most single-item operations (creating a page, updating a single property)
    * **Array bodies** `[]` are required for bulk operations or specific endpoints that expect multiple items (like updating multiple page properties at once)

    Always check the [Confluence API documentation](https://developer.atlassian.com/cloud/confluence/rest/v2/intro/) for your endpoint to see which format is required. The documentation will show example request bodies in the correct format.
  </Accordion>
</AccordionGroup>


# Content Snare
Source: https://relevanceai.com/docs/integrations/popular-integrations/content-snare

Automate client onboarding, information collection, and document gathering workflows with the Content Snare integration.

Content Snare is a client information gathering and content management platform used by digital agencies, accountants, lawyers, and consultants to streamline collecting documents and information from clients. It eliminates email back-and-forth and reduces information gathering time significantly through structured, trackable requests.

Key concepts in Content Snare:

* **Clients** — individual client profiles that requests are sent to
* **Requests** — customized information collection forms sent to clients
* **Templates** — reusable request structures for repeatable workflows

This is a native OAuth integration. It replaces the older Pipedream-based method and requires no manual API token configuration.

## Connect the integration

<Steps>
  <Step title="Navigate to the Integrations page">
    Go to the **Integrations** page in your Relevance AI dashboard.
  </Step>

  <Step title="Find Content Snare">
    Search for or scroll to **Content Snare** in the integrations list and select it.
  </Step>

  <Step title="Add the integration">
    Click **Add Integration**.
  </Step>

  <Step title="Complete OAuth authorization">
    You will be redirected to Content Snare to authorize access. Log in to your Content Snare account and grant the requested permissions.
  </Step>

  <Step title="Confirm the connection">
    Once authorized, you will be redirected back to Relevance AI. The integration will appear as connected and is ready to use.
  </Step>
</Steps>

## Tools & tool steps

The following pre-built tool steps are available once you connect Content Snare.

### Client management

<AccordionGroup>
  <Accordion title="Content Snare - Create Client">
    Creates a new client profile in your Content Snare account. Use this when onboarding a new client — typically triggered by a new contact being added to your CRM or a form submission.

    Required inputs include the client's name and email address. Optional fields allow you to set additional profile details.
  </Accordion>
</AccordionGroup>

### Request management

<AccordionGroup>
  <Accordion title="Content Snare - Create Request">
    Initiates an information collection request and sends it to a specific client. Use this to trigger a document or data gathering workflow — for example, sending a tax document request after a new engagement is created.

    You specify the client, the template to use, and any deadline or customization options.
  </Accordion>
</AccordionGroup>

### Advanced

<AccordionGroup>
  <Accordion title="Content Snare API Call">
    A generic tool step for making direct calls to the Content Snare API. Use this when you need to access an endpoint not covered by the dedicated tool steps — for example, listing existing requests, retrieving client details, or checking request status.

    Authentication is handled automatically via OAuth. You provide the HTTP method, endpoint path, and any request body or query parameters.
  </Accordion>
</AccordionGroup>

## Use the integration's API tool step (advanced)

The **Content Snare API Call** tool step gives you access to any endpoint in the Content Snare Partner API. This is useful when the dedicated tool steps don't cover your specific workflow.

### Example: create a client

```json theme={null}
POST /partner_api/v1/clients
{
  "name": "Acme Corp",
  "email": "contact@acmecorp.com"
}
```

### Example: create a request

```json theme={null}
POST /partner_api/v1/requests
{
  "client_id": "abc123",
  "template_id": "tmpl_456",
  "name": "2024 Tax Documents",
  "due_date": "2025-03-31"
}
```

<Note>
  OAuth handles authentication automatically — you do not need to include any authorization headers manually.
</Note>

<Warning>
  Content Snare enforces rate limits of 50 read requests and 20 write requests per 10 seconds. Build in error handling for 429 responses if your workflow sends many requests in quick succession.
</Warning>

For the full list of available endpoints and request/response schemas, see the [Content Snare Partner API documentation](https://api.contentsnare.com/partner_api/v1/documentation).

## Example use cases

<CardGroup>
  <Card title="Digital agency client onboarding" icon="building">
    Trigger a Content Snare request automatically when a new project is created. Collect brand assets, copy, logins, and briefing documents without chasing clients over email.
  </Card>

  <Card title="Accountant tax document collection" icon="calculator">
    Send a tax document request to each client at the start of the filing season. Use templates to standardize what's collected and track completion status across your client list.
  </Card>

  <Card title="Lawyer client intake" icon="scale-balanced">
    Automate intake workflows by sending a Content Snare request when a new matter is opened. Collect identification documents, signed forms, and background information before the first meeting.
  </Card>

  <Card title="Consultant discovery phase" icon="magnifying-glass">
    Kick off engagements by sending a structured discovery questionnaire. Combine with a CRM integration so requests are sent automatically when a deal moves to a new stage.
  </Card>
</CardGroup>

### CRM-triggered workflows

Combine Content Snare with CRM integrations such as Salesforce or HubSpot to create end-to-end onboarding automation. When a new contact is created or a deal reaches a certain stage, an agent can create the client in Content Snare and immediately send an information request — with no manual steps required.

## Frequently asked questions

<AccordionGroup>
  <Accordion title="What permissions are granted during OAuth authorization?">
    The OAuth flow grants Relevance AI access to create and manage clients and requests in your Content Snare account. The exact scopes are defined by Content Snare's Partner API and cover the operations needed for the available tool steps.
  </Accordion>

  <Accordion title="Can I connect multiple Content Snare accounts?">
    Yes. You can add multiple Content Snare connections from the Integrations page — for example, if you manage separate Content Snare accounts for different business units. Each connection is authorized independently and can be selected when configuring tool steps.
  </Accordion>

  <Accordion title="How is authentication handled?">
    Authentication uses OAuth 2.0 only. There is no option to configure manual API tokens for this integration. If you need to rotate or revoke access, disconnect and reconnect the integration from the Integrations page.
  </Accordion>

  <Accordion title="What happens if I exceed the rate limit?">
    Content Snare returns a 429 status code when rate limits are exceeded (50 read or 20 write requests per 10 seconds). Your tool or agent should handle this response and retry after a short delay. Avoid designing workflows that send large batches of write requests simultaneously.
  </Accordion>

  <Accordion title="How do I troubleshoot connection issues?">
    If the integration fails to connect or becomes disconnected, try removing the integration from the Integrations page and reconnecting. Make sure you are logging in to the correct Content Snare account during the OAuth flow. If the issue persists, [contact Relevance AI support](/docs/get-started/support).
  </Accordion>
</AccordionGroup>


# Fireflies.ai
Source: https://relevanceai.com/docs/integrations/popular-integrations/fireflies

Connect Fireflies.ai to Relevance AI to access meeting transcripts, retrieve meeting data, and upload audio files for transcription

Fireflies.ai is an AI-powered meeting assistant that automatically transcribes, summarizes, searches, and analyzes voice conversations. It works across major video conferencing platforms including Zoom, Google Meet, Microsoft Teams, Webex, and more, providing 90%+ accuracy in transcription.

With the Fireflies integration in Relevance AI, you can access meeting transcripts, retrieve meeting data, and upload audio files for transcription directly within your AI workflows.

## Connect the integration

Connecting your Fireflies.ai account to Relevance AI is a straightforward process:

1. Go to the "Integrations" page in the sidebar of your Relevance AI dashboard.
2. Click on "Fireflies" from the available integrations.
3. Click on the "Add Integration" button.
4. In the pop-up window, sign into your Fireflies.ai account.
5. Grant the necessary permissions for Relevance AI to access your Fireflies.ai data.
6. Once authenticated, your Fireflies.ai account will appear as a connected integration.

## Tool steps for Fireflies

The Fireflies.ai integration provides a set of actions that your agents can use to interact with your meeting transcripts and audio processing. These actions can be incorporated into your agent's workflows as tool steps.

### Meeting Retrieval

<CardGroup>
  <Card title="Find Meeting by ID" icon="magnifying-glass">
    Retrieve specific meeting details using a meeting ID
  </Card>

  <Card title="Find Recent Meeting" icon="clock">
    Access the most recent meeting transcripts
  </Card>
</CardGroup>

### Audio Processing

<CardGroup>
  <Card title="Upload Audio" icon="upload">
    Upload audio files for transcription
  </Card>
</CardGroup>

<Tip>
  Type "Fireflies" in the tool step search bar to see all available Fireflies.ai actions when building your tools.
</Tip>

## Use the integration's API tool step (Advanced)

In addition to the pre-built actions available in the tool directory, you can build custom tools that perform Fireflies-specific activities using the Fireflies API Call tool step.

### How to use the API Call tool step

<Steps>
  <Step title="Create a new tool">
    Create a new tool in Relevance AI or open an existing tool you want to add Fireflies.ai functionality to.
  </Step>

  <Step title="Add the Fireflies API Call tool step">
    1. Scroll down to Tool-steps
    2. Search for "Fireflies API Call" in the tool step search bar
    3. Add the Fireflies API Call tool step to your workflow
  </Step>

  <Step title="Select your Fireflies account">
    Select your connected Fireflies.ai account from the dropdown menu.
  </Step>

  <Step title="Configure the API endpoint">
    Configure the API endpoint, method, and parameters according to your needs. Fireflies.ai uses a GraphQL API, so you'll typically use POST requests with GraphQL queries.
  </Step>

  <Step title="Test your API call">
    Test your configuration to ensure it works correctly before deploying.
  </Step>
</Steps>

### Example: Retrieve Meeting Transcript

Here's a practical example of using the Fireflies API Call tool step to retrieve a meeting transcript:

**Configuration**:

```json theme={null}
{
  "endpoint": "/graphql",
  "method": "POST",
  "body": {
    "query": "query Transcript($transcriptId: String!) { transcript(id: $transcriptId) { title date sentences { text speaker_name } } }"
  }
}
```

This configuration:

* Uses the POST method for GraphQL queries
* Retrieves transcript details including title, date, and speaker-identified sentences
* Returns structured data that can be used in subsequent workflow steps

<Info>
  You can find Fireflies.ai's complete API documentation at [https://docs.fireflies.ai/](https://docs.fireflies.ai/).
</Info>

### Common API endpoints

<AccordionGroup>
  <Accordion title="GraphQL API" icon="code">
    The main endpoint for all Fireflies.ai API operations is `/graphql`. Common queries include:

    * **Get transcript**: Retrieve full transcript with speaker identification
    * **List transcripts**: Get a list of all available transcripts
    * **Get user info**: Retrieve current user information

    [View API Documentation](https://docs.fireflies.ai/)
  </Accordion>
</AccordionGroup>

## Example use cases

Here are some ways you can leverage the Fireflies.ai integration with your agents:

<AccordionGroup>
  <Accordion title="Automated Meeting Notes Distribution" icon="envelope">
    Automatically send meeting summaries and transcripts to team members after calls, ensuring everyone stays informed even if they couldn't attend.
  </Accordion>

  <Accordion title="CRM Updates from Sales Calls" icon="database">
    Extract key information from sales meetings and update your CRM automatically, capturing important details like customer pain points, budget discussions, and next steps.
  </Accordion>

  <Accordion title="Action Item Extraction" icon="list-check">
    Identify and create tasks from action items mentioned in meetings, automatically assigning them to the right team members with relevant context.
  </Accordion>

  <Accordion title="Meeting Insights Dashboard" icon="chart-line">
    Aggregate meeting data to track trends, topics, and team engagement over time, helping you understand communication patterns and improve meeting effectiveness.
  </Accordion>

  <Accordion title="Compliance and Record-Keeping" icon="folder-open">
    Maintain searchable archives of important meetings for compliance purposes, with easy retrieval of specific conversations when needed.
  </Accordion>

  <Accordion title="Training and Onboarding Analysis" icon="graduation-cap">
    Analyze training sessions and onboarding calls to improve processes, identify common questions, and create better training materials.
  </Accordion>

  <Accordion title="Customer Feedback Aggregation" icon="comments">
    Automatically extract and categorize customer feedback from support calls and customer meetings to inform product development.
  </Accordion>

  <Accordion title="Sales Call Coaching" icon="headset">
    Review sales call transcripts to identify successful patterns, coaching opportunities, and best practices to share across the team.
  </Accordion>
</AccordionGroup>

## Follow along on YouTube

Watch this tutorial to see how to turn meeting notes into content using Fireflies and Relevance AI:

<Frame>
  <iframe title="Turn Meeting Notes Into Content with AI Agents (Fireflies x Relevance AI Demo)" />
</Frame>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What permissions does the Fireflies.ai integration require?">
    Fireflies.ai requires access to your meeting recordings and transcripts. The specific permissions depend on your video conferencing platform and Fireflies.ai account settings. Typically, you'll need to grant Fireflies.ai permission to join meetings and access recordings.
  </Accordion>

  <Accordion title="Which meeting platforms are supported?">
    Fireflies.ai supports all major video conferencing platforms including:

    * Zoom
    * Google Meet
    * Microsoft Teams
    * Webex
    * GoToMeeting
    * RingCentral
    * And many others

    It can also transcribe uploaded audio files from any source.
  </Accordion>

  <Accordion title="How accurate are the transcriptions?">
    Fireflies.ai provides 90%+ accuracy in transcription, with support for multiple languages and accents. Accuracy may vary based on audio quality, speaker clarity, background noise, and technical terminology.
  </Accordion>

  <Accordion title="Can I access historical meetings?">
    Yes, you can access historical meetings stored in your Fireflies.ai account using the "Find Meeting by ID" tool step. All meetings that have been transcribed by Fireflies.ai are available through the integration.
  </Accordion>

  <Accordion title="How do I find a meeting ID?">
    Meeting IDs can be found in your Fireflies.ai dashboard. Each transcribed meeting has a unique ID that you can use with the "Find Meeting by ID" tool step.
  </Accordion>

  <Accordion title="Are there rate limits?">
    Rate limits depend on your Fireflies.ai plan. Refer to your Fireflies.ai account settings or contact Fireflies.ai support for specific rate limit information.
  </Accordion>

  <Accordion title="Can I use this integration with multiple Fireflies.ai accounts?">
    Yes, you can add multiple Fireflies.ai integrations to your Relevance AI workspace, each connected to a different Fireflies.ai account.
  </Accordion>
</AccordionGroup>


# Firmable
Source: https://relevanceai.com/docs/integrations/popular-integrations/firmable

Firmable is a B2B sales intelligence platform specializing in Australian and New Zealand markets, providing accurate company and contact data for ANZ businesses.

Firmable is a B2B sales intelligence platform specializing in Australian and New Zealand markets, providing accurate company and contact data for ANZ businesses. With Relevance AI's Firmable integration, you can seamlessly enrich your leads with verified contact and company information specific to the AU/NZ region, enabling your AI agents to automate prospect research, qualify leads, and build comprehensive sales pipelines with locally-verified data.

<Warning>
  The Firmable integration in Relevance AI was built by Relevance AI, and is therefore supported by our team, not Firmable. If you have a question or issue with using Firmable in Relevance AI, please [reach out to our support team](/docs/get-started/support). If you have a question or issue that is only about Firmable, you can [reach out to Firmable support](https://www.firmable.com/contact).
</Warning>

## Connect the integration

Connecting your Firmable account to Relevance AI is a straightforward process:

1. Go to the "Integrations" page in the sidebar of your Relevance AI dashboard.
2. Click on "Firmable" from the available integrations.
3. Click on the "Add Integration" button.
4. Enter your Firmable API key when prompted.
5. Once authenticated, your Firmable account will appear as a connected integration.

<Tip>
  You can find your Firmable API key in your Firmable account settings under API Access. If you don't have a Firmable account yet, you can [sign up here](https://www.firmable.com/).
</Tip>

## Tool steps for Firmable

The Firmable integration provides a comprehensive set of actions that your agents can use to enrich leads, find contacts, and gather company intelligence specific to Australian and New Zealand markets. These actions can be incorporated into your agent's workflows as tool steps, enabling sophisticated sales automation and prospect research capabilities for ANZ businesses.

### People Enrichment

<CardGroup>
  <Card title="Firmable: People Enrichment" icon="user-plus">
    Enrich existing contact records with verified email addresses, phone numbers, and professional details using LinkedIn URL, email, or LinkedIn slug
  </Card>

  <Card title="Firmable: People Search" icon="users">
    Search for contacts matching specific criteria such as company, position, seniority, or department across Firmable's ANZ database
  </Card>
</CardGroup>

### Company Intelligence

<CardGroup>
  <Card title="Firmable: Company Enrichment" icon="building">
    Enrich company records with firmographic data using LinkedIn, domain, ABN (Australian Business Number), or website
  </Card>
</CardGroup>

### Advanced API Access

<CardGroup>
  <Card title="Firmable API Call" icon="code">
    Make custom API calls to access any Firmable API endpoint for advanced use cases
  </Card>
</CardGroup>

<Tip>
  Type "Firmable" in the tool step search bar to see all available Firmable actions when building your tools.
</Tip>

## Use the integration's API tool step (Advanced)

In addition to the pre-built actions available in the tool directory, you can build custom tools that perform Firmable-specific activities using the Firmable API Call tool step.

### How to use the Firmable API Call tool step

<Steps>
  <Step title="Create a new tool">
    Create a new tool in Relevance AI or open an existing tool you want to add Firmable functionality to.
  </Step>

  <Step title="Add the Firmable API Call tool step">
    1. Scroll down to Tool-steps
    2. Search for "Firmable API Call" in the tool step search bar
    3. Add the Firmable API Call tool step to your workflow
  </Step>

  <Step title="Select your Firmable account">
    Select your connected Firmable account from the dropdown menu.
  </Step>

  <Step title="Configure the API endpoint">
    Configure the API endpoint, method, and parameters according to your needs:

    * **Method**: Select the HTTP method (GET, POST)
    * **Endpoint**: Enter the API endpoint path (e.g., `/enrich/person`)
    * **Body**: Add any required request body data with search criteria
  </Step>

  <Step title="Test your API call">
    Test your configuration to ensure it works correctly before deploying.
  </Step>
</Steps>

### Example: Enriching a Contact by LinkedIn URL

Here's a practical example of using the Firmable API Call tool step to enrich a contact using their LinkedIn profile:

**API Endpoint**: `POST /enrich/person`

**Configuration**:

```json theme={null}
{
  "method": "POST",
  "endpoint": "/enrich/person",
  "body": {
    "linkedin_url": "https://www.linkedin.com/in/john-doe-au"
  }
}
```

This configuration:

* Uses the POST method to request contact enrichment
* Specifies the `/enrich/person` endpoint for individual contact lookup
* Provides the LinkedIn URL as the search parameter
* Returns verified contact information including email addresses, phone numbers, job title, company details, and location data specific to Australian and New Zealand professionals

<Info>
  You can find Firmable's complete API documentation at [https://www.firmable.com/api-documentation](https://www.firmable.com/api-documentation).
</Info>

### Common Firmable API Endpoints

Here are some commonly used Firmable API endpoints you can use with the API Call tool step:

<AccordionGroup>
  <Accordion title="Person Enrichment" icon="user">
    * **Enrich by LinkedIn URL**: `POST /enrich/person` with `{"linkedin_url": "https://linkedin.com/in/username"}`
    * **Enrich by email**: `POST /enrich/person` with `{"email": "user@example.com.au"}`
    * **Enrich by LinkedIn slug**: `POST /enrich/person` with `{"linkedin_slug": "john-doe-au"}`

    Returns: Email addresses, phone numbers, job title, company, location, and professional details for ANZ contacts

    [View API Documentation](https://www.firmable.com/api-documentation)
  </Accordion>

  <Accordion title="Company Enrichment" icon="building">
    * **Enrich by domain**: `POST /enrich/company` with `{"domain": "example.com.au"}`
    * **Enrich by LinkedIn**: `POST /enrich/company` with `{"linkedin_url": "https://linkedin.com/company/example"}`
    * **Enrich by ABN**: `POST /enrich/company` with `{"abn": "12345678901"}`
    * **Enrich by website**: `POST /enrich/company` with `{"website": "https://example.com.au"}`

    Returns: Company size, industry, location, ABN, employee count, technologies used, and firmographic data for Australian and New Zealand businesses

    [View API Documentation](https://www.firmable.com/api-documentation)
  </Accordion>

  <Accordion title="People Search" icon="magnifying-glass">
    * **Search by company**: `POST /search/people` with `{"company": "Acme Corp"}`
    * **Search by position**: `POST /search/people` with `{"position": "Sales Manager"}`
    * **Search by seniority**: `POST /search/people` with `{"seniority": "Director"}`
    * **Search by department**: `POST /search/people` with `{"department": "Marketing"}`

    Useful for building prospect lists of decision-makers within target ANZ companies

    [View API Documentation](https://www.firmable.com/api-documentation)
  </Accordion>

  <Accordion title="Advanced Filtering" icon="filter">
    * **Combined search criteria**: Use multiple filters together for precise targeting
    * **Location filtering**: Target specific Australian states or New Zealand regions
    * **Company size filtering**: Focus on SMBs or enterprise accounts

    Allows sophisticated prospect discovery across the ANZ market

    [View API Documentation](https://www.firmable.com/api-documentation)
  </Accordion>
</AccordionGroup>

## Example use cases

Here are some ways you can leverage the Firmable integration with your agents for Australian and New Zealand markets:

<AccordionGroup>
  <Accordion title="ANZ Lead Enrichment" icon="user-plus">
    Create an agent that automatically enriches incoming leads from Australian and New Zealand markets with verified contact information. The agent can take a LinkedIn URL, email address, or company ABN, use Firmable to gather phone numbers, job titles, and company details, then update your CRM with the enriched ANZ-specific data.
  </Accordion>

  <Accordion title="Australian Business Research Agent" icon="magnifying-glass-chart">
    Build an agent that conducts deep research on Australian businesses before sales calls. Given a company ABN, domain, or LinkedIn profile, the agent uses Firmable to gather company information, key decision-makers, and organizational structure, then compiles a comprehensive research brief tailored to the ANZ market.
  </Accordion>

  <Accordion title="ABN-Based Lead Qualification" icon="filter">
    Deploy an agent that automatically qualifies Australian leads using ABN lookups. The agent uses Firmable to verify company legitimacy, gather company size and industry data, then scores and routes leads to the appropriate sales representatives based on your ANZ qualification criteria.
  </Accordion>

  <Accordion title="ANZ Outbound List Builder" icon="list-check">
    Create an agent that builds targeted prospect lists for Australian and New Zealand outbound campaigns. The agent searches Firmable's database for contacts matching your ideal customer profile (job title, company size, industry, location within AU/NZ), enriches the results, and exports them to your sales engagement platform.
  </Accordion>

  <Accordion title="Regional CRM Data Enhancement" icon="database">
    Build an agent that maintains data quality for your ANZ contacts in your CRM. The agent identifies incomplete or outdated contact records for Australian and New Zealand businesses, uses Firmable to find current information including ABN verification, and automatically updates your CRM with verified local data.
  </Accordion>

  <Accordion title="ANZ Account-Based Marketing Research" icon="bullseye">
    Deploy an agent that supports ABM campaigns targeting Australian and New Zealand accounts. The agent uses Firmable to identify key decision-makers within target ANZ companies, gather their contact information, and map out the organizational structure for strategic regional outreach.
  </Accordion>

  <Accordion title="Australian Recruitment Sourcing" icon="user-tie">
    Create an agent that helps recruiters find and contact qualified candidates in Australia and New Zealand. The agent searches for professionals with specific skills and experience in the ANZ market, uses Firmable to find their contact information, and prepares personalized outreach messages.
  </Accordion>

  <Accordion title="Trans-Tasman Partnership Development" icon="handshake">
    Build an agent that identifies and researches potential business partners across Australia and New Zealand. The agent finds companies in complementary industries within the ANZ region, uses Firmable to identify partnership decision-makers, and gathers contact information for cross-border outreach.
  </Accordion>
</AccordionGroup>

## Best practices

<AccordionGroup>
  <Accordion title="Data Privacy & ANZ Compliance" icon="shield-check">
    **Always ensure compliance with Australian and New Zealand data privacy regulations:**

    * Comply with the Australian Privacy Act and New Zealand Privacy Act
    * Only use Firmable data for legitimate business purposes
    * Include proper opt-out mechanisms in your outreach
    * Store enriched data securely and delete when no longer needed
    * Review Firmable's terms of service and acceptable use policy
    * Respect the Australian Spam Act 2003 and New Zealand's Unsolicited Electronic Messages Act

    Firmable provides ANZ-compliant data, but you're responsible for how you use it.
  </Accordion>

  <Accordion title="Credit Management" icon="coins">
    **Optimize your Firmable credit usage:**

    * Each enrichment request consumes Firmable credits
    * Cache enriched data to avoid duplicate lookups
    * Implement logic to check if data already exists before enriching
    * Monitor your credit usage through Firmable's dashboard
    * Set up alerts when credits are running low
    * Prioritize high-value prospects for enrichment
  </Accordion>

  <Accordion title="Data Quality for ANZ Markets" icon="star">
    **Maximize the quality of enriched ANZ data:**

    * Provide as much information as possible in your search queries (LinkedIn URL + email is better than email alone)
    * Validate enriched data before using it in outreach
    * Implement fallback logic when Firmable doesn't find a match
    * Regularly update enriched data as contact information changes
    * Verify ABN numbers for Australian companies to ensure legitimacy
    * Consider regional variations in job titles and company structures between AU and NZ
  </Accordion>

  <Accordion title="Regional Targeting" icon="location-dot">
    **Leverage Firmable's ANZ market specialization:**

    * Use location filters to target specific Australian states or New Zealand regions
    * Consider time zones when scheduling outreach (AEST, AEDT, NZST, NZDT)
    * Tailor messaging to local business culture and practices
    * Use ABN lookups for Australian companies to verify business registration
    * Understand the differences between Australian and New Zealand business environments
  </Accordion>

  <Accordion title="Rate Limiting" icon="gauge">
    **Handle API rate limits gracefully:**

    * Firmable enforces rate limits to ensure service quality
    * Implement retry logic with exponential backoff for rate limit errors
    * Distribute enrichment tasks over time for large datasets
    * Monitor API response codes and handle errors appropriately
    * Contact Firmable support if you need higher limits for enterprise use
  </Accordion>
</AccordionGroup>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What is Firmable and why should I use it?">
    Firmable is a B2B sales intelligence platform specializing in Australian and New Zealand markets. It provides verified contact and company information specifically for ANZ businesses. Using Firmable with Relevance AI allows you to automate lead enrichment, prospect research, and data quality maintenance at scale for the AU/NZ region, with unique features like ABN lookups for Australian companies.
  </Accordion>

  <Accordion title="Do I need a Firmable account to use this integration?">
    Yes, you need an active Firmable account with API access. You can sign up for Firmable at [https://www.firmable.com/](https://www.firmable.com/). You'll need to generate an API key from your Firmable account settings to connect it to Relevance AI.
  </Accordion>

  <Accordion title="How do I get my Firmable API key?">
    To get your Firmable API key:

    1. Log in to your Firmable account
    2. Navigate to Settings or Account Settings
    3. Look for "API Access" or "API Keys"
    4. Generate a new API key or copy your existing key
    5. Use this key when connecting Firmable to Relevance AI

    Keep your API key secure and never share it publicly.
  </Accordion>

  <Accordion title="What does 'beta' status mean for the Firmable integration?">
    The beta status indicates that while the Firmable integration is fully functional and available for use, we're still gathering user feedback and may make refinements to improve the experience. We recommend testing your workflows thoroughly before deploying them in production. If you encounter any issues or have suggestions, please contact our support team.
  </Accordion>

  <Accordion title="What is an ABN and how do I use it with Firmable?">
    An ABN (Australian Business Number) is a unique 11-digit identifier for businesses operating in Australia. Firmable's company enrichment tool can look up Australian companies using their ABN, which is particularly useful for:

    * Verifying business legitimacy
    * Finding company details for Australian businesses
    * Enriching leads with official business registration data

    Simply provide the 11-digit ABN in the company enrichment tool step to retrieve comprehensive company information.
  </Accordion>

  <Accordion title="Does Firmable cover both Australia and New Zealand?">
    Yes, Firmable specializes in both Australian and New Zealand markets. The platform provides comprehensive B2B data for companies and contacts across both countries, making it ideal for businesses targeting the ANZ region. You can filter searches by country or specific regions within AU/NZ.
  </Accordion>

  <Accordion title="How many Firmable credits does each enrichment use?">
    Credit usage depends on the type of enrichment and the data returned. Typically:

    * Person enrichment: 1-2 credits per successful lookup
    * Company enrichment: 1 credit per successful lookup
    * Search operations: Credits vary based on result volume

    Check your Firmable plan details for specific credit allocations and pricing. You can monitor your credit usage in your Firmable dashboard.
  </Accordion>

  <Accordion title="What happens if Firmable doesn't find a match?">
    If Firmable doesn't find information for a contact or company, the API will return an empty result or indicate no match was found. Your agent should include logic to handle these cases, such as:

    * Trying alternative search parameters (e.g., LinkedIn URL instead of email)
    * Using fallback data sources
    * Flagging the record for manual research
    * Skipping to the next record in bulk operations

    Not all ANZ contacts are in Firmable's database, so plan for partial match rates.
  </Accordion>

  <Accordion title="Can I use Firmable for cold outreach in Australia and New Zealand?">
    Yes, but you must comply with all applicable laws and regulations:

    * Follow the Australian Spam Act 2003 and New Zealand's Unsolicited Electronic Messages Act
    * Include proper identification and opt-out mechanisms
    * Only contact people with legitimate business interest
    * Respect do-not-contact lists and preferences
    * Review Firmable's acceptable use policy

    Firmable provides the data, but you're responsible for how you use it in compliance with ANZ regulations.
  </Accordion>

  <Accordion title="How accurate is Firmable's ANZ data?">
    Firmable maintains high data accuracy through continuous verification and updates specific to Australian and New Zealand markets. However, contact information can change over time. Best practices:

    * Use the most recent enrichment data available
    * Implement email verification before sending campaigns
    * Update your records regularly
    * Verify ABN numbers for Australian companies
    * Have a process for handling bounced emails or wrong numbers
    * Consider that ANZ business data may update less frequently than global databases
  </Accordion>

  <Accordion title="Can I enrich contacts from LinkedIn profiles?">
    Yes! Firmable can enrich LinkedIn profiles for Australian and New Zealand professionals. You can provide a LinkedIn URL or LinkedIn slug to the Firmable people enrichment tool step, and it will return verified contact information including email addresses and phone numbers. This is particularly useful for ANZ sales prospecting and recruitment workflows.
  </Accordion>

  <Accordion title="What's the difference between the enrichment and search tool steps?">
    * **Enrichment tool steps**: Used when you have specific identifying information (LinkedIn URL, email, ABN, domain) and want to retrieve that specific contact or company's details
    * **Search tool steps**: Used when you want to discover contacts matching certain criteria (company, position, seniority, department) within the ANZ market

    Use enrichment for data completion, search for prospecting and list building.
  </Accordion>

  <Accordion title="Can I integrate Firmable with my CRM?">
    Yes! You can build agents that connect Firmable with popular CRMs like Salesforce, HubSpot, and others available in Relevance AI. Your agent can:

    * Pull ANZ contacts from your CRM
    * Enrich them with Firmable
    * Update the CRM with enriched data including ABN numbers
    * All automatically on a schedule or trigger

    Check our [integrations page](/docs/integrations/introduction) for available CRM connections.
  </Accordion>

  <Accordion title="Are there rate limits for the Firmable API?">
    Yes, Firmable enforces rate limits to ensure service quality for all users. The specific limits depend on your Firmable plan. If you hit rate limits:

    * Your agent will receive a rate limit error
    * Implement retry logic with delays
    * Distribute enrichment tasks over time
    * Contact Firmable support if you need higher limits

    Design your workflows to handle rate limits gracefully.
  </Accordion>
</AccordionGroup>


# Freshdesk
Source: https://relevanceai.com/docs/integrations/popular-integrations/freshdesk

Interact with your Freshdesk with Relevance.

Connect your Freshdesk customer support platform with Relevance AI to supercharge your support operations with intelligent automation. The Freshdesk integration enables your AI agents to handle support tickets, respond to customer inquiries, and streamline your customer service workflow.

## Overview

The Freshdesk integration allows you to connect your Freshdesk account with Relevance AI, enabling your AI agents to interact with your customer support platform. This powerful connection transforms how you handle support tickets by automating responses, categorizing issues, and providing intelligent assistance to both customers and support teams.

With this integration, you can build AI agents that monitor your Freshdesk tickets, respond to common inquiries automatically, escalate complex issues to human agents, and maintain a seamless support experience across your organization.

## Connecting the Integration

To connect the Freshdesk integration with Relevance AI:

1. Navigate to your Relevance AI dashboard and select **Integrations & API Keys** from the main menu.
2. Find and select "Freshdesk" from the available integrations.
3. Click "Connect" to begin the authorization process.
4. Enter your Freshdesk domain (e.g. yourcompany.freshdesk.com).
5. Provide your Freshdesk API key:
   * Log in to your Freshdesk account.
   * Go to your profile settings.
   * Find your API key in the "API Settings" section.
   * Copy the API key.
6. Paste your API key into the Relevance AI connection form.
7. Click "Authorize" to establish the connection.
8. Once connected, you'll see a confirmation message indicating successful integration.

## Setting Up Triggers

You can also trigger your Agent using Freshdesk. To do this, head to your Agent and click 'Triggers', then select the 'Freshdesk' Trigger.

## Use the integration's API tool step (advanced)

You can build custom tools that perform Freshdesk-specific activities using the Freshdesk API Call tool step:

1. Create a new tool in Relevance AI.
2. Scroll down to Tool steps.
3. Add the Freshdesk API Tool step.
4. Select your connected Freshdesk account in the dropdown.
5. Configure the API endpoint, method, and parameters according to your needs.
6. This advanced approach gives you full access to the Freshdesk REST API, allowing you to implement custom functionality

<Info>You can find Freshdesk's API documentation [here](https://developers.freshdesk.com/api/#introduction).</Info>


# GitHub
Source: https://relevanceai.com/docs/integrations/popular-integrations/github

GitHub is the world's leading platform for version control and collaborative software development.

GitHub is the world's leading platform for version control and collaborative software development. With Relevance AI's GitHub integration, you can connect your GitHub repositories to your AI agents, enabling them to manage code, issues, pull requests, and workflows. Your agents can also respond automatically to GitHub events like new issues, pull requests, commits, and more through instant webhook triggers.

<Warning>
  The GitHub integration in Relevance AI was built by Relevance AI, and is therefore supported by our team, not GitHub. If you have a question or issue with using GitHub in Relevance AI, please [reach out to our support team](/docs/get-started/support). If you have a question or issue that is only about GitHub, you can [reach out to GitHub support](https://support.github.com/).
</Warning>

## Connect the integration

Connecting your GitHub account to Relevance AI is a straightforward process:

1. Go to the "Integrations" page in the sidebar of your Relevance AI dashboard.
2. Click on "GitHub" from the available integrations.
3. Click on the "Add Integration" button.
4. In the pop-up window, sign into your GitHub account.
5. Grant the necessary permissions for Relevance AI to access your GitHub repositories.
6. Once authenticated, your GitHub account will appear as a connected integration.

## Triggers for GitHub

The GitHub integration allows you to set up triggers that automatically activate your AI agents when specific events occur in your repositories. Most GitHub triggers are **instant** (webhook-based), meaning your agents respond in real-time as soon as events happen — whether it's a new issue being opened, a pull request being merged, or a workflow run completing. This enables event-driven automation that keeps your development workflows moving without manual intervention.

The following trigger options are available, organized by category:

<AccordionGroup>
  <Accordion title="Repository events" icon="folder">
    Monitor changes at the repository level.

    * **New repository (Instant)** — Triggers when a new repository is created in your account or organization. Useful for applying default settings, adding collaborators, or creating initial issues.

    * **New fork (Instant)** — Triggers when someone forks your repository. Use this to track adoption, send welcome messages, or log fork activity.

    * **New star (Instant)** — Triggers when a user stars your repository. Ideal for tracking growth milestones or sending thank-you notifications.

    * **New watcher (Instant)** — Triggers when a user starts watching your repository for notifications.

    * **New tag (Instant)** — Triggers when a new tag is pushed to a repository. Useful for initiating release workflows or changelog generation.
  </Accordion>

  <Accordion title="Issue & pull request events" icon="code-pull-request">
    Respond to activity on issues and pull requests.

    * **New issue (Instant)** — Triggers when a new issue is opened. Perfect for auto-labeling, assigning issues to the right team member, or sending acknowledgment replies.

    * **Updated issue (Instant)** — Triggers when an issue is modified (title, body, labels, assignees, or status). Use this to sync issue state with external project management tools.

    * **New pull request (Instant)** — Triggers when a pull request is opened or ready for review. Ideal for starting automated code review, running checks, or notifying reviewers.

    * **Updated pull request (Instant)** — Triggers when a pull request is modified or changes state (e.g., converted to draft, marked ready, or closed).

    * **New comment (Instant)** — Triggers when a comment is added to an issue or pull request. Useful for extracting action items, detecting questions, or routing feedback.

    * **New label on issue (Instant)** — Triggers when a label is applied to an issue. Use this to route issues into category-specific workflows.

    * **New review request (Instant)** — Triggers when a reviewer is requested on a pull request. Useful for notifying reviewers or starting pre-review automation.

    * **New pull request review (Instant)** — Triggers when a review is submitted on a pull request (approved, changes requested, or commented).
  </Accordion>

  <Accordion title="Commit & branch events" icon="code-branch">
    React to new code activity in your repositories.

    * **New commit (Instant)** — Triggers when a new commit is pushed to any branch. Use this to run linting, update changelogs, or notify downstream systems.

    * **New push (Instant)** — Triggers on any push to a repository, including multiple commits. Useful for triggering CI/CD steps or deployment checks.

    * **New branch (Instant)** — Triggers when a new branch is created. Ideal for setting up branch-specific configurations or notifying the team.
  </Accordion>

  <Accordion title="Workflow events" icon="diagram-project">
    Monitor GitHub Actions activity.

    * **Workflow run completed (Instant)** — Triggers when a GitHub Actions workflow run finishes, whether it succeeded, failed, or was cancelled. Use this to send alerts on failure or trigger post-deployment steps.

    * **Workflow job completed (Instant)** — Triggers when an individual job within a workflow run completes. Useful for fine-grained monitoring of CI pipeline steps.
  </Accordion>

  <Accordion title="Collaboration events" icon="users">
    Track team and contributor activity.

    * **New collaborator (Instant)** — Triggers when a collaborator is added to a repository. Useful for onboarding automation or access auditing.

    * **New mention (Instant)** — Triggers when your account is mentioned in an issue, pull request, or comment. Use this to route mentions to the right team or create follow-up tasks.
  </Accordion>

  <Accordion title="Project events" icon="table-columns">
    Automate workflows around GitHub Projects.

    * **Project item status changed (Instant)** — Triggers when the status of an item in a GitHub Project (v2) changes. Use this to sync project state with external tools or notify stakeholders.

    * **New card in column (Instant)** — Triggers when a card is added to a specific column in a classic GitHub Project board.
  </Accordion>

  <Accordion title="Other events" icon="bell">
    Additional triggers for releases, gists, and discussions.

    * **New release (Instant)** — Triggers when a new release is published. Ideal for triggering release announcement workflows or documentation updates.

    * **New gist (Instant)** — Triggers when a new gist is created in your account. Use this to catalog or share snippets automatically.

    * **New discussion (Instant)** — Triggers when a new discussion is opened in a repository. Useful for monitoring community activity or routing questions to support.

    * **New discussion comment (Instant)** — Triggers when a comment is added to a discussion. Use this to detect follow-up questions or escalate unresolved topics.
  </Accordion>
</AccordionGroup>

## Tool steps for GitHub

The GitHub integration provides a comprehensive set of actions that your agents can use to interact with your repositories and development workflows. These actions can be incorporated into your agent's workflows as tool steps, enabling sophisticated development automation capabilities.

### Repository management

<CardGroup>
  <Card title="Create Repository" icon="folder-plus">
    Create a new GitHub repository
  </Card>

  <Card title="Get Repository Info" icon="circle-info">
    Retrieve information about a repository
  </Card>

  <Card title="Get Repository Content" icon="file-code">
    Access files and folders in a repository
  </Card>

  <Card title="Star Repo" icon="star">
    Star a repository
  </Card>
</CardGroup>

### Branch & commit operations

<CardGroup>
  <Card title="Create Branch" icon="code-branch">
    Create a new branch in a repository
  </Card>

  <Card title="Get Commit" icon="code-commit">
    Retrieve details about a specific commit
  </Card>

  <Card title="List Commits" icon="list">
    List commits in a repository
  </Card>

  <Card title="Create or Update File Contents" icon="file-pen">
    Create new files or update existing ones
  </Card>
</CardGroup>

### Issue management

<CardGroup>
  <Card title="Create Issue" icon="circle-exclamation">
    Create a new issue in a repository
  </Card>

  <Card title="Create Issue Comment" icon="comment">
    Add a comment to an existing issue
  </Card>

  <Card title="Update Issue" icon="pen-to-square">
    Update an existing issue
  </Card>

  <Card title="Get Issue Assignees" icon="users">
    Retrieve assignees for an issue
  </Card>

  <Card title="Search Issues and Pull Requests" icon="magnifying-glass">
    Search for issues and PRs across repositories
  </Card>
</CardGroup>

### Pull request operations

<CardGroup>
  <Card title="Create Pull Request" icon="code-pull-request">
    Create a new pull request
  </Card>

  <Card title="Get Reviewers" icon="user-check">
    Get reviewers for a pull request
  </Card>

  <Card title="Update Project (V2) Item Status" icon="list-check">
    Update the status of a project item
  </Card>
</CardGroup>

### Workflow automation

<CardGroup>
  <Card title="Create Workflow Dispatch" icon="play">
    Trigger a workflow run
  </Card>

  <Card title="Enable Workflow" icon="toggle-on">
    Enable a GitHub Actions workflow
  </Card>

  <Card title="Disable Workflow" icon="toggle-off">
    Disable a GitHub Actions workflow
  </Card>

  <Card title="Get Workflow Run" icon="circle-info">
    Get details about a workflow run
  </Card>

  <Card title="List Workflow Runs" icon="list">
    List workflow runs for a repository
  </Card>
</CardGroup>

### Gist management

<CardGroup>
  <Card title="Create Gist" icon="note-sticky">
    Create a new gist
  </Card>

  <Card title="Update Gist" icon="pen">
    Update an existing gist
  </Card>

  <Card title="List Gists for a User" icon="list">
    List all gists for a user
  </Card>
</CardGroup>

### Release management

<CardGroup>
  <Card title="List Releases" icon="tags">
    List releases for a repository
  </Card>
</CardGroup>

<Tip>
  Type "GitHub" in the tool step search bar to see all available GitHub actions when building your tools.
</Tip>

## Use the integration's API tool step (advanced)

In addition to the pre-built actions available in the tool directory, you can build custom tools that perform GitHub-specific activities using the GitHub API Call tool step.

<Warning>
  **Required Header:** When using the GitHub API Call tool step, you must include the header `accept: application/vnd.github+json` or you may encounter errors like `'Please enter a value for 'response_body' field'`.
</Warning>

### How to use the GitHub API Call tool step

<Steps>
  <Step title="Create a new tool">
    Create a new tool in Relevance AI or open an existing tool you want to add GitHub functionality to.
  </Step>

  <Step title="Add the GitHub API Call tool step">
    1. Scroll down to Tool-steps
    2. Search for "GitHub API Call" in the tool step search bar
    3. Add the GitHub API Call tool step to your workflow
  </Step>

  <Step title="Select your GitHub account">
    Select your connected GitHub account from the dropdown menu.
  </Step>

  <Step title="Configure the API endpoint">
    Configure the API endpoint, method, and parameters according to your needs:

    * **Method**: Select the HTTP method (GET, POST, PUT, DELETE, PATCH)
    * **Endpoint**: Enter the API endpoint path (e.g., `/repos/{owner}/{repo}/collaborators/{username}`)
    * **Body**: Add any required request body data
  </Step>

  <Step title="Add the required header">
    **This is the critical step!** In the Headers section, add:

    ```json theme={null}
    {
      "accept": "application/vnd.github+json"
    }
    ```

    This header tells GitHub's API what format to return the response in. Without it, the API may not return a properly formatted response, causing errors in your tool.
  </Step>

  <Step title="Test your API call">
    Test your configuration to ensure it works correctly before deploying.
  </Step>
</Steps>

### Example: Managing repository collaborators

Here's a practical example of using the GitHub API Call tool step to add a collaborator to a repository:

**API Endpoint**: `PUT /repos/{owner}/{repo}/collaborators/{username}`

**Configuration**:

```json theme={null}
{
  "method": "PUT",
  "endpoint": "/repos/myorg/myrepo/collaborators/johndoe",
  "headers": {
    "accept": "application/vnd.github+json"
  },
  "body": {
    "permission": "push"
  }
}
```

This configuration:

* Uses the PUT method to add or update a collaborator
* Specifies the repository owner, name, and collaborator username
* **Includes the required `accept` header**
* Sets the permission level to "push" (can also be "pull", "triage", "maintain", or "admin")

<Info>
  You can find GitHub's complete API documentation at [https://docs.github.com/en/rest](https://docs.github.com/en/rest).
</Info>

### Common GitHub API endpoints

Here are some commonly used GitHub API endpoints you can use with the API Call tool step:

<AccordionGroup>
  <Accordion title="Repository collaborators" icon="users">
    * **Check permissions**: `GET /repos/{owner}/{repo}/collaborators/{username}/permission`
    * **Add collaborator**: `PUT /repos/{owner}/{repo}/collaborators/{username}`
    * **Remove collaborator**: `DELETE /repos/{owner}/{repo}/collaborators/{username}`

    [View API documentation](https://docs.github.com/en/rest/collaborators/collaborators)
  </Accordion>

  <Accordion title="Repository contents" icon="folder">
    * **Get contents**: `GET /repos/{owner}/{repo}/contents/{path}`
    * **Create/update file**: `PUT /repos/{owner}/{repo}/contents/{path}`
    * **Delete file**: `DELETE /repos/{owner}/{repo}/contents/{path}`

    [View API documentation](https://docs.github.com/en/rest/repos/contents)
  </Accordion>

  <Accordion title="Issues" icon="circle-exclamation">
    * **List issues**: `GET /repos/{owner}/{repo}/issues`
    * **Get issue**: `GET /repos/{owner}/{repo}/issues/{issue_number}`
    * **Create issue**: `POST /repos/{owner}/{repo}/issues`
    * **Update issue**: `PATCH /repos/{owner}/{repo}/issues/{issue_number}`

    [View API documentation](https://docs.github.com/en/rest/issues/issues)
  </Accordion>

  <Accordion title="Pull requests" icon="code-pull-request">
    * **List PRs**: `GET /repos/{owner}/{repo}/pulls`
    * **Get PR**: `GET /repos/{owner}/{repo}/pulls/{pull_number}`
    * **Create PR**: `POST /repos/{owner}/{repo}/pulls`
    * **Update PR**: `PATCH /repos/{owner}/{repo}/pulls/{pull_number}`
    * **Merge PR**: `PUT /repos/{owner}/{repo}/pulls/{pull_number}/merge`

    [View API documentation](https://docs.github.com/en/rest/pulls/pulls)
  </Accordion>

  <Accordion title="Branches" icon="code-branch">
    * **List branches**: `GET /repos/{owner}/{repo}/branches`
    * **Get branch**: `GET /repos/{owner}/{repo}/branches/{branch}`
    * **Create branch**: `POST /repos/{owner}/{repo}/git/refs`

    [View API documentation](https://docs.github.com/en/rest/branches/branches)
  </Accordion>
</AccordionGroup>

<Tip>
  **Remember**: Always include the `accept: application/vnd.github+json` header in your API calls!
</Tip>

## Example use cases

Here are some ways you can use the GitHub integration with your agents:

<AccordionGroup>
  <Accordion title="Auto-triage new issues with AI classification" icon="filter">
    When the New Issue trigger fires, an agent analyzes the issue content and automatically applies labels (bug, feature request, documentation), assigns it to the appropriate team member based on the affected area, and posts an initial response with relevant links or questions — all before a human has seen it.
  </Accordion>

  <Accordion title="Automated pull request review" icon="magnifying-glass-chart">
    When the New Pull Request trigger fires, an agent reviews the diff for common issues — security vulnerabilities, missing tests, style violations — and posts inline comments with specific suggestions. The agent can request changes or approve based on configurable thresholds.
  </Accordion>

  <Accordion title="Workflow failure notifications" icon="bell">
    When the Workflow Run Completed trigger fires with a failed status, an agent posts a summary to your team's chat channel, links to the failed job logs, and creates a GitHub issue with reproduction steps if the same workflow has failed multiple times in a row.
  </Accordion>

  <Accordion title="Automated code review assistant" icon="code">
    Create an agent that reviews pull requests, checks for common issues, adds comments with suggestions, and requests changes when necessary. The agent can analyze code quality, security vulnerabilities, and adherence to coding standards.
  </Accordion>

  <Accordion title="Release notes generator" icon="newspaper">
    Deploy an agent that automatically generates comprehensive release notes by analyzing commits, pull requests, and issues between releases, formatting them according to your team's standards.
  </Accordion>

  <Accordion title="Repository maintenance agent" icon="broom">
    Create an agent that performs routine maintenance tasks like closing stale issues, updating dependencies, archiving old branches, and ensuring repository settings comply with organizational policies.
  </Accordion>

  <Accordion title="Onboarding automation" icon="user-plus">
    Build an agent that automates developer onboarding by creating repositories from templates, setting up branch protection rules, adding team members with appropriate permissions, and creating initial issues for setup tasks.
  </Accordion>

  <Accordion title="Documentation sync agent" icon="book">
    Deploy an agent that keeps documentation in sync across multiple repositories, automatically updates README files when changes are detected, and ensures documentation standards are maintained.
  </Accordion>

  <Accordion title="Security compliance monitor" icon="shield-halved">
    Create an agent that monitors repositories for security compliance, checks for exposed secrets, ensures branch protection is enabled, and alerts teams when security policies are violated.
  </Accordion>

  <Accordion title="Workflow orchestrator" icon="diagram-project">
    Build an agent that orchestrates complex GitHub Actions workflows, triggers deployments based on specific conditions, monitors workflow runs, and handles failures with automatic retries or notifications.
  </Accordion>
</AccordionGroup>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What permissions does the GitHub integration require?">
    The integration requires permissions to read and write to repositories, manage issues and pull requests, and access workflow information. You can review the specific permissions during the authentication process. You can also configure fine-grained access tokens in GitHub to limit access to specific repositories.
  </Accordion>

  <Accordion title="Why am I getting 'Please enter a value for response_body field' error?">
    This error typically occurs when the required `accept: application/vnd.github+json` header is missing from your GitHub API Call tool step. Make sure to add this header in the Headers section of your API call configuration.

    **Solution**: Add the following header to your API call:

    ```json theme={null}
    {
      "accept": "application/vnd.github+json"
    }
    ```
  </Accordion>

  <Accordion title="Can I restrict my agent to only access certain repositories?">
    Yes, you can configure your tools and triggers to only interact with specific repositories by setting the appropriate repository parameters. Additionally, you can use GitHub's fine-grained personal access tokens to limit access at the GitHub level.
  </Accordion>

  <Accordion title="How do I authenticate with GitHub for API calls?">
    When you connect your GitHub account through the Integrations page, Relevance AI handles authentication automatically. When using the GitHub API Call tool step, simply select your connected GitHub account from the dropdown, and authentication will be handled for you.
  </Accordion>

  <Accordion title="What's the difference between the pre-built GitHub tool steps and the API Call tool step?">
    Pre-built tool steps (like "Create Issue" or "Create Pull Request") are designed for specific, common tasks and have simplified interfaces with guided inputs. The GitHub API Call tool step gives you full access to GitHub's REST API, allowing you to implement any functionality available in the API, including advanced or custom operations not covered by pre-built steps.
  </Accordion>

  <Accordion title="Can I use GitHub webhooks to trigger my agents?">
    Yes. The GitHub integration includes instant (webhook-based) triggers that activate your agents when specific events occur in your repositories — such as new pull requests, issues, or commits. Set up these triggers directly from the Triggers section in Relevance AI; no manual webhook configuration in GitHub is required.
  </Accordion>

  <Accordion title="Are there rate limits for GitHub API calls?">
    Yes, GitHub enforces rate limits on API calls. For authenticated requests, the limit is typically 5,000 requests per hour. Your agents should be designed to handle rate limiting gracefully. You can check your current rate limit status using the `/rate_limit` endpoint.
  </Accordion>

  <Accordion title="Can I work with GitHub Enterprise?">
    The GitHub integration is designed to work with GitHub.com. For GitHub Enterprise Server installations, you may need to use the generic API Tool step with custom authentication. Contact our support team for assistance with GitHub Enterprise configurations.
  </Accordion>
</AccordionGroup>


# Gmail
Source: https://relevanceai.com/docs/integrations/popular-integrations/gmail

Connect and create emails with Relevance AI's Gmail integration

## Overview

Connect your Gmail account to Relevance AI and supercharge your email workflow with AI agents that can read, respond to, and take action on your emails. The Gmail integration enables you to automate email-related tasks, create powerful workflows, and save valuable time.

## Connect the integration

Setting up the Gmail integration with Relevance AI is straightforward:

1. Go to the **Integrations & API Keys** page in the sidebar of your Relevance AI dashboard.
2. Click on "Gmail" from the available integrations.
3. Click the "Add Integration" button.
4. In the pop-up window, sign into your Gmail account and authorize Relevance AI to access your emails.
5. Follow any additional prompts to complete the connection process.
6. Once connected, you'll see your Gmail account listed under connected integrations.

## Setting up triggers

The Gmail integration allows you to set up triggers that automatically activate your AI agents when specific email events occur. This enables your agents to respond to emails in real-time without manual intervention.

To set up Gmail as a trigger:

1. Create or edit an agent in Relevance AI.
2. Navigate to the "Triggers" section in "Agent Profile".
3. Select "Gmail" as the trigger source.
4. Configure the trigger conditions based on your needs:
   * **New Email**: Trigger when any new email arrives in your inbox
   * **Emails with Specific Labels**: Trigger only for emails with certain Gmail labels
   * **Emails from Specific Senders**: Trigger for emails from particular email addresses
   * **Emails with Specific Subject Lines**: Trigger based on keywords in subject lines
   * **Unread Emails**: Trigger only for unread messages
5. Set the frequency of checking for new emails (e.g., immediately, every 5 minutes, hourly).
6. Save your trigger configuration.

Once set up, your agent will automatically start working whenever the specified email conditions are met, allowing for seamless email automation.

## Tools & Tool Steps

The Gmail integration provides a rich set of actions that your AI agents can use to interact with your email account. These actions can be incorporated into your agent's workflow to perform various email-related tasks.

Here are some of the key Gmail actions available:

### Read Emails

* **Get Email Content**: Retrieve the full content of specific emails
* **Search Emails**: Find emails matching specific criteria
* **Get Email Attachments**: Access files attached to emails
* **Get Email Metadata**: Extract information like sender, recipients, date, etc.
* **Get Email Thread**: Retrieve entire conversation threads

### Compose and Send Emails

* **Send Email**: Create and send new emails
* **Reply to Email**: Respond to existing email threads
* **Forward Email**: Forward messages to other recipients
* **Draft Email**: Create draft emails without sending
* **Schedule Email**: Compose emails to be sent at a later time

### Email Management

* **Apply Labels**: Organize emails with Gmail labels
* **Mark as Read/Unread**: Change the read status of emails
* **Archive Email**: Move emails out of the inbox
* **Move to Folder**: Organize emails into different folders
* **Delete Email**: Remove unwanted emails
* **Star/Flag Email**: Mark important emails for follow-up

### Advanced Actions

* **Extract Information**: Parse email content for specific data
* **Analyze Sentiment**: Determine the tone and sentiment of emails
* **Categorize Emails**: Automatically sort emails by type or priority
* **Set Reminders**: Create follow-up reminders for important emails
* **Create Tasks**: Generate tasks based on email content

These are just some examples of the many actions available. The Gmail integration is highly versatile, allowing your agents to perform virtually any task you would normally do manually in your Gmail account.

## Use the integration's API tool step (advanced)

For more advanced use cases, you can leverage the Gmail API directly using the Gmail API Call tool step:

1. Create a new tool in Relevance AI.
2. Scroll down to Tool-steps.
3. Add the Gmail API tool-step.
4. Select your connected Gmail account in the dropdown.

With the Gmail API tool step, you can:

1. **Access Advanced Gmail Features**: Utilize Gmail API endpoints not covered by standard actions
2. **Custom Email Processing**: Create specialized workflows for unique email handling requirements
3. **Batch Operations**: Perform actions on multiple emails simultaneously
4. **Complex Filtering**: Implement sophisticated email filtering logic
5. **Integration with Other Services**: Connect your email workflows with other tools and platforms

### Example: Custom Email Processing with Gmail API

Here's an example of using the Gmail API tool step to implement a custom email processing workflow:

```json theme={null}
// Example of using Gmail API to process emails with specific criteria
{
  "method": "GET",
  "endpoint": "users/me/messages",
  "params": {
    "q": "is:unread from:important-client@example.com",
    "maxResults": 10
  }
}
```

This API call retrieves up to 10 unread emails from a specific sender, which your agent can then process according to your business rules.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What permissions does Relevance AI need for the Gmail integration?">Relevance AI requires permission to read, send, and manage your emails to provide full functionality. You can review these permissions during the connection process.</Accordion>
  <Accordion title="Can I use multiple Gmail accounts with Relevance AI?">Yes, you can connect multiple Gmail accounts to Relevance AI and configure different agents to work with specific accounts.</Accordion>
  <Accordion title="Is my email data secure?">Yes, Relevance AI maintains strict security protocols. Your email data is encrypted and only accessed as needed for the functions you authorize.</Accordion>
  <Accordion title="Can I limit which emails my agent can access?">Yes, you can configure triggers and actions to work only with specific labels, senders, or other criteria to limit the scope of emails your agent processes.</Accordion>
  <Accordion title="Will my agent's actions appear in my sent folder?">Yes, emails sent by your agent will appear in your Gmail sent folder, maintaining a complete record of all communications.</Accordion>
</AccordionGroup>

## Gmail Assistant

The Gmail Assistant is a pre-built agent available from the [Marketplace](https://app.relevanceai.com/marketplace) that handles email tasks through plain English. Search, send, organize, and manage your inbox through simple conversation — no manual clicking through Gmail required.

<Card title="Clone Gmail Assistant" icon="envelope" href="https://app.relevanceai.com/marketplace">
  Search "Gmail Assistant" in the Marketplace to clone it and start managing your inbox through conversation
</Card>


# Gong
Source: https://relevanceai.com/docs/integrations/popular-integrations/gong

Connect Gong to Relevance AI to access call recordings, transcripts, and revenue intelligence data.

Gong is a revenue intelligence platform that records, transcribes, and analyzes sales calls and customer conversations. It surfaces insights from your team's interactions to help you understand deal health, coaching opportunities, and customer sentiment.

With the native Gong integration in Relevance AI, your agents can retrieve call data, pull transcripts, and interact with Gong's API directly — without any third-party connectors.

## Connect the integration

<div>
  <iframe title="How to connect the Gong integration" />
</div>

1. Go to the **Integrations & API Keys** page in the sidebar of your Relevance AI dashboard.
2. In the **Integrations** section, search for **Gong**.
3. Click **Gong** — the native integration with the blue tick.
4. In the pop-up window, sign into your Gong account and authorize the requested permissions.
5. Once authenticated, your Gong account appears as a connected integration.

<Note>
  You need a Gong account with API access enabled. Gong API access typically requires Technical Admin permissions — check with your Gong administrator if you cannot complete the authorization.
</Note>

## Tool steps for Gong

<img alt="Gong tool steps in the tool step search bar" />

The Gong integration provides a range of tool steps that your agents can use to interact with your call data and Engage flows. Add these to any tool in the tool builder by searching for "Gong" in the tool step search bar.

Select a category below, then expand any tool step to see what it does.

<Tabs>
  <Tab title="Calls">
    <AccordionGroup>
      <Accordion title="List calls" icon="list">
        Retrieve a list of calls from your Gong account, filterable by date range, participants, and other criteria.
      </Accordion>

      <Accordion title="Get extensive data" icon="chart-bar">
        Fetch detailed metadata and analytics for a specific call, including duration, participants, talk ratios, and engagement metrics.
      </Accordion>

      <Accordion title="Retrieve transcripts" icon="file-lines">
        Pull the full transcript for a call, with speaker-identified turns and timestamps.
      </Accordion>

      <Accordion title="Add new call" icon="plus">
        Add a call record to Gong, such as uploading recordings from other platforms or logging external conversations.
      </Accordion>
    </AccordionGroup>
  </Tab>

  <Tab title="Engage flows">
    <AccordionGroup>
      <Accordion title="List Engage flows" icon="diagram-project">
        Retrieve the Gong Engage flows in your account.
      </Accordion>

      <Accordion title="List Engage flow folders" icon="folder">
        Retrieve the folders that organize your Gong Engage flows.
      </Accordion>

      <Accordion title="Assign prospects to flow" icon="user-plus">
        Add prospects to a Gong Engage flow.
      </Accordion>

      <Accordion title="List assigned flow for prospects" icon="users">
        See which flows a set of prospects are currently assigned to.
      </Accordion>

      <Accordion title="Unassign prospects by CRM ID" icon="user-minus">
        Remove prospects from a flow using their CRM ID.
      </Accordion>

      <Accordion title="Unassign prospects by flow instance ID" icon="user-xmark">
        Remove prospects from a flow using the flow instance ID.
      </Accordion>
    </AccordionGroup>
  </Tab>

  <Tab title="API access">
    <AccordionGroup>
      <Accordion title="API call" icon="plug">
        Call a specific Gong API endpoint.
      </Accordion>

      <Accordion title="Generic API call" icon="code">
        Make any request to the Gong REST API. Use this when you need functionality beyond the pre-built tool steps.
      </Accordion>
    </AccordionGroup>
  </Tab>
</Tabs>

### How to add a tool step

<div>
  <iframe title="How to add a Gong tool step" />
</div>

<Tip>
  Type "Gong" in the tool step search bar to see all available Gong actions when building your tools.
</Tip>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What Gong permissions are required to connect?">
    You need a Gong account with API access. In most organizations, this requires Technical Admin permissions in Gong. Contact your Gong administrator if the OAuth authorization fails or you do not see an API access option in your Gong settings.
  </Accordion>

  <Accordion title="Can I connect multiple Gong accounts?">
    Yes. You can add multiple Gong integrations to your Relevance AI workspace, each authenticated with a different Gong account.
  </Accordion>

  <Accordion title="How do I migrate from the previous Pipedream-based Gong actions?">
    If you previously used Pipedream-based Gong actions, migrate to the native tools — they offer the same functionality with a simpler OAuth-based connection. Connect the native Gong integration via OAuth (see steps above), then replace any Pipedream-based Gong tool steps in your tools with the equivalent native tool steps. The native tools do not require a separate Pipedream account.
  </Accordion>

  <Accordion title="How can I tell which Gong tool steps are native?">
    Native tool steps have a blue tick beside them in the tool step search bar. Use these in preference to the Pipedream-based actions.
  </Accordion>

  <Accordion title="Can the integration access calls in real time?">
    The integration can only access calls that have already been processed and indexed in Gong. There is typically a short delay after a call ends before it is available via the API. Real-time call access is not supported.
  </Accordion>

  <Accordion title="Are there rate limits?">
    Gong enforces API rate limits based on your Gong plan and usage tier. If you encounter rate limit errors, reduce the frequency of API calls or contact Gong support to review your limits.
  </Accordion>
</AccordionGroup>


# Google Analytics (GA4)
Source: https://relevanceai.com/docs/integrations/popular-integrations/google-analytics

Connect Google Analytics 4 to Relevance AI to automate reporting, track conversions, and manage properties with AI agents

Google Analytics 4 (GA4) is Google's analytics platform for measuring traffic, engagement, and conversions across your websites and apps. With the Google Analytics integration in Relevance AI, you can connect your GA4 properties to AI agents that run custom reports, create key events for conversion tracking, manage properties programmatically, and query the GA4 APIs directly.

<Note>
  This integration is built by Relevance AI. For integration-specific support, contact [Relevance AI support](/docs/get-started/support). For Google Analytics platform issues, visit [Google Analytics Help](https://support.google.com/analytics).
</Note>

## Connect the integration

The Google Analytics integration uses OAuth to authenticate with your Google account. You only need to complete this process once per Google account.

<Steps>
  <Step title="Navigate to Integrations & API Keys">
    Select **Integrations & API Keys** from the left sidebar in your Relevance AI dashboard.
  </Step>

  <Step title="Find Google Analytics">
    Browse the integrations directory and click on **Google Analytics**.
  </Step>

  <Step title="Click Connect">
    Click the **Connect** button to start the OAuth flow.
  </Step>

  <Step title="Sign in to Google">
    You'll be redirected to Google's authentication page. Select the Google account that has access to the GA4 properties you want to use.
  </Step>

  <Step title="Review permissions">
    Review the permissions Relevance AI is requesting. These are needed to read analytics data and manage your GA4 properties on your behalf.
  </Step>

  <Step title="Approve access">
    Click **Allow** to grant access. You'll be redirected back to Relevance AI.
  </Step>

  <Step title="Confirm connection">
    Your Google Analytics account will now appear as connected in the Integrations dashboard and will be available for use in your agents and tools.
  </Step>
</Steps>

You can connect multiple Google accounts if you manage GA4 properties across different Google logins. Each connected account appears separately in the integrations list.

<Warning>
  Your Google account must have at least **Viewer** access to the GA4 properties you want to query. To create properties or key events, you need **Editor** or **Administrator** access on those properties.
</Warning>

## Tools & tool steps

The Google Analytics integration provides four tool steps that your agents can use. Search for "Google Analytics" in the tool step search bar when building your tools.

### Run Reports

<AccordionGroup>
  <Accordion title="Run Reports" icon="chart-line">
    Retrieve analytics data from a GA4 property by specifying metrics (quantitative measurements) and dimensions (data categories). Use this to pull traffic data, engagement stats, conversion counts, or any other GA4 report data into your agent workflows.

    **Key inputs:**

    * **Property ID**: The GA4 property to query (e.g., `properties/123456789`)
    * **Metrics**: What to measure — for example, `sessions`, `activeUsers`, `conversions`, `screenPageViews`
    * **Dimensions**: How to break down the data — for example, `date`, `country`, `sessionSource`, `pagePath`
    * **Date range**: The time period to report on
    * **Filters**: Optional dimension or metric filters to narrow results

    **What are metrics and dimensions?**
    In GA4, *metrics* are numeric values (how much, how many), and *dimensions* are attributes used to categorize or segment that data. For example, `sessions` is a metric and `country` is a dimension — together they give you sessions by country.
  </Accordion>
</AccordionGroup>

### Create Key Events

<AccordionGroup>
  <Accordion title="Create Key Events" icon="flag-checkered">
    Create a key event in a GA4 property. Key events mark the actions that matter most to your business — they were called "conversions" in Universal Analytics (UA) and earlier versions of GA4. Creating a key event tells GA4 to highlight that event in reports and use it as a conversion signal.

    **Key inputs:**

    * **Property ID**: The GA4 property to add the key event to
    * **Event name**: The name of the existing GA4 event to mark as a key event (e.g., `purchase`, `sign_up`, `generate_lead`)

    **Note:** The event must already be collected in your GA4 property before it can be designated as a key event. You cannot create an event and mark it as a key event in the same step.
  </Accordion>
</AccordionGroup>

### Create GA4 properties

<AccordionGroup>
  <Accordion title="Create GA4 properties" icon="building">
    Create a new Google Analytics 4 property programmatically. This is useful when you're provisioning GA4 tracking for new client websites, new app launches, or when you manage analytics at scale across many sites.

    **Key inputs:**

    * **Display name**: The name for the new property
    * **Time zone**: The reporting time zone for the property
    * **Currency code**: The currency used for revenue metrics (e.g., `USD`, `EUR`)
    * **Industry category**: The industry the property belongs to (optional, used for benchmarking)

    **Note:** You must have **Administrator** access in the Google Analytics account where the property will be created. Properties are created under your Google Analytics account, not under a specific website — you'll still need to set up a data stream separately to start collecting data.
  </Accordion>
</AccordionGroup>

### Google Analytics API Call

<AccordionGroup>
  <Accordion title="Google Analytics API Call" icon="code">
    Make a custom API call to any Google Analytics Data API or Admin API endpoint. Use this when the pre-built tool steps don't cover your specific use case.

    This tool step handles OAuth authentication automatically, so you don't need to manage access tokens. See the [advanced section below](#use-the-google-analytics-api-tool-step-advanced) for full usage details and examples.
  </Accordion>
</AccordionGroup>

## Use the Google Analytics API Tool Step (Advanced)

For use cases beyond the pre-built tool steps, the Google Analytics API Call tool step gives you direct access to both the [GA4 Data API](https://developers.google.com/analytics/devguides/reporting/data/v1) and the [GA4 Admin API](https://developers.google.com/analytics/devguides/config/admin/v1).

### How to use the Google Analytics API Call tool step

<Steps>
  <Step title="Create or open a tool">
    Create a new tool in Relevance AI or open an existing one you want to add Google Analytics functionality to.
  </Step>

  <Step title="Add the tool step">
    Scroll down to **Tool-steps**, search for "Google Analytics API", and add the **Google Analytics API Call** tool step to your workflow.
  </Step>

  <Step title="Select your connected account">
    Choose your connected Google Analytics account from the dropdown.
  </Step>

  <Step title="Configure the request">
    Set the HTTP method, endpoint path, and request body:

    * **Method**: The HTTP method (`GET`, `POST`, `DELETE`, etc.)
    * **Endpoint**: The API path, starting from the base URL (e.g., `v1beta/properties/123456789:runReport`)
    * **Body**: The JSON request body (for POST requests)
  </Step>

  <Step title="Test the request">
    Run a test to confirm the API call returns the expected data before using it in production.
  </Step>
</Steps>

### Example: Run a custom report

This example retrieves the top 10 pages by page views for the past 30 days.

**API**: `POST https://analyticsdata.googleapis.com/v1beta/properties/{propertyId}:runReport`

```json theme={null}
{
  "method": "POST",
  "endpoint": "v1beta/properties/123456789:runReport",
  "body": {
    "dateRanges": [
      { "startDate": "30daysAgo", "endDate": "today" }
    ],
    "dimensions": [
      { "name": "pagePath" }
    ],
    "metrics": [
      { "name": "screenPageViews" }
    ],
    "orderBys": [
      {
        "metric": { "metricName": "screenPageViews" },
        "desc": true
      }
    ],
    "limit": 10
  }
}
```

### Example: Create a key event

This example marks the `purchase` event as a key event on a GA4 property.

**API**: `POST https://analyticsadmin.googleapis.com/v1beta/properties/{propertyId}/keyEvents`

```json theme={null}
{
  "method": "POST",
  "endpoint": "v1beta/properties/123456789/keyEvents",
  "body": {
    "eventName": "purchase"
  }
}
```

### Example: Query real-time data

This example retrieves how many users are active on the site right now, broken down by country.

**API**: `POST https://analyticsdata.googleapis.com/v1beta/properties/{propertyId}:runRealtimeReport`

```json theme={null}
{
  "method": "POST",
  "endpoint": "v1beta/properties/123456789:runRealtimeReport",
  "body": {
    "dimensions": [
      { "name": "country" }
    ],
    "metrics": [
      { "name": "activeUsers" }
    ]
  }
}
```

<Info>
  Replace `123456789` with your actual GA4 property ID. You can find this in your GA4 account under **Admin** > **Property Settings**.
</Info>

## Common GA4 API endpoints

<AccordionGroup>
  <Accordion title="Data API — reporting" icon="chart-bar">
    The GA4 Data API base URL is `https://analyticsdata.googleapis.com/`.

    * **Run report**: `POST v1beta/properties/{propertyId}:runReport`
    * **Run real-time report**: `POST v1beta/properties/{propertyId}:runRealtimeReport`
    * **Run pivot report**: `POST v1beta/properties/{propertyId}:runPivotReport`
    * **Batch run reports**: `POST v1beta/properties/{propertyId}:batchRunReports`
    * **Get metadata** (available metrics and dimensions): `GET v1beta/properties/{propertyId}/metadata`

    [View Data API documentation](https://developers.google.com/analytics/devguides/reporting/data/v1/rest)
  </Accordion>

  <Accordion title="Admin API — properties" icon="building">
    The GA4 Admin API base URL is `https://analyticsadmin.googleapis.com/`.

    * **List properties**: `GET v1beta/properties?filter=parent:accounts/{accountId}`
    * **Get property**: `GET v1beta/properties/{propertyId}`
    * **Create property**: `POST v1beta/properties`
    * **Update property**: `PATCH v1beta/properties/{propertyId}`
    * **Delete property**: `DELETE v1beta/properties/{propertyId}`

    [View Admin API — properties documentation](https://developers.google.com/analytics/devguides/config/admin/v1/rest/v1beta/properties)
  </Accordion>

  <Accordion title="Admin API — key events" icon="flag-checkered">
    * **List key events**: `GET v1beta/properties/{propertyId}/keyEvents`
    * **Create key event**: `POST v1beta/properties/{propertyId}/keyEvents`
    * **Get key event**: `GET v1beta/properties/{propertyId}/keyEvents/{keyEventId}`
    * **Delete key event**: `DELETE v1beta/properties/{propertyId}/keyEvents/{keyEventId}`

    [View Admin API — key events documentation](https://developers.google.com/analytics/devguides/config/admin/v1/rest/v1beta/properties.keyEvents)
  </Accordion>

  <Accordion title="Admin API — data streams" icon="wifi">
    * **List data streams**: `GET v1beta/properties/{propertyId}/dataStreams`
    * **Get data stream**: `GET v1beta/properties/{propertyId}/dataStreams/{dataStreamId}`
    * **Create data stream**: `POST v1beta/properties/{propertyId}/dataStreams`
    * **Update data stream**: `PATCH v1beta/properties/{propertyId}/dataStreams/{dataStreamId}`

    [View Admin API — data streams documentation](https://developers.google.com/analytics/devguides/config/admin/v1/rest/v1beta/properties.dataStreams)
  </Accordion>
</AccordionGroup>

## Example use cases

<AccordionGroup>
  <Accordion title="Automated performance reporting" icon="file-chart-column">
    Build an agent that runs GA4 reports on a schedule, compiles key metrics (sessions, conversions, revenue), and delivers formatted summaries to Slack, email, or a Google Sheet. The agent can compare performance week-over-week or month-over-month and flag significant changes automatically.
  </Accordion>

  <Accordion title="Conversion tracking automation" icon="flag-checkered">
    Create an agent that reads from a CRM or project management tool and automatically creates GA4 key events when new business goals are defined. This removes the manual step of updating GA4 settings each time your team identifies a new conversion action to track.
  </Accordion>

  <Accordion title="Multi-property management" icon="building">
    Deploy an agent that provisions GA4 properties at scale — useful for agencies managing analytics for multiple clients or organizations rolling out tracking across many regional sites. The agent creates properties, applies consistent settings, and logs the property IDs for reference.
  </Accordion>

  <Accordion title="Custom dashboard creation" icon="table-columns">
    Connect GA4 data to internal dashboards by having an agent pull specific metrics and dimensions on demand. The agent can combine data from multiple GA4 properties, apply business logic, and write results to a database or BI tool for visualization.
  </Accordion>

  <Accordion title="Threshold-based alerts" icon="bell">
    Set up an agent that monitors GA4 metrics and sends alerts when values cross defined thresholds — for example, when bounce rate exceeds 80%, when daily sessions drop by more than 20% compared to the previous week, or when conversion rate falls below a target. The agent runs reports at regular intervals and routes alerts to the appropriate channel.
  </Accordion>

  <Accordion title="Cross-platform analytics" icon="arrows-rotate">
    Build an agent that combines GA4 data with data from your CRM, advertising platforms, or marketing automation tools. For example, pull GA4 conversion data alongside HubSpot lead data to get a full picture of how web traffic translates into pipeline, or combine GA4 session data with ad spend data to calculate cost per session by channel.
  </Accordion>

  <Accordion title="Marketing ROI calculation" icon="calculator">
    Create an agent that retrieves GA4 conversion data (key events, revenue) alongside campaign spend data from advertising platforms, then calculates return on ad spend (ROAS), cost per acquisition (CPA), and other ROI metrics. Results can be written to a spreadsheet or sent as a periodic report.
  </Accordion>

  <Accordion title="Audience behavior analysis" icon="users">
    Deploy an agent that runs detailed GA4 reports segmented by user attributes — device type, location, traffic source, landing page — and identifies patterns in how different audience segments engage with your site. The agent can surface insights about which segments convert at higher rates or where users drop off in key flows.
  </Accordion>
</AccordionGroup>

## Best practices

* **Request minimal permissions**: OAuth grants access to all GA4 properties in your Google account. Where possible, use a dedicated Google account with access only to the properties your agent needs, rather than a personal account with access to many unrelated properties.
* **Respect rate limits**: The GA4 Data API enforces quotas per property, including limits on requests per day, tokens per day, and concurrent requests. Design your agents to batch reports where possible and avoid making redundant API calls. See [GA4 API quotas](https://developers.google.com/analytics/devguides/reporting/data/v1/quotas) for current limits.
* **Account for data processing delays**: GA4 data is typically processed within 24–48 hours. Real-time reports show data from the last 30 minutes but exclude some processed signals. If your use case requires fully processed data (including attribution and session stitching), query for dates at least 48 hours in the past.
* **Use property IDs, not names**: Property names can change but property IDs don't. Store and reference properties by their numeric ID (visible in GA4 Admin > Property Settings) rather than by display name.
* **Handle empty responses**: GA4 reports return no rows when there's no matching data for the requested dimensions and date range. Make sure your agent handles empty result sets gracefully rather than treating them as errors.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What Google account permissions do I need?" icon="lock">
    To connect the integration, your Google account must have at least **Viewer** access on the GA4 properties you want to query. To create key events or manage property settings, you need **Editor** access. To create new properties, you need **Administrator** access on the Google Analytics account.

    You can check your access level in GA4 under **Admin** > **Account Access Management** or **Property Access Management**.
  </Accordion>

  <Accordion title="Can I connect multiple Google accounts?" icon="user-group">
    Yes. You can connect multiple Google accounts by completing the OAuth flow once for each account. Each connected account will appear separately in your Relevance AI integrations list, and you can select which account to use when configuring individual tool steps.
  </Accordion>

  <Accordion title="What are GA4 key events?" icon="flag-checkered">
    Key events are the actions you've identified as important to your business — for example, purchases, form submissions, or sign-ups. In GA4, you first collect an event (via your tracking code or Google Tag Manager), then designate it as a key event to highlight it in reports and use it for conversion measurement. Key events were called "conversions" in Universal Analytics and in earlier versions of GA4; Google renamed them to key events in 2024.
  </Accordion>

  <Accordion title="What are GA4 API rate limits?" icon="gauge-high">
    The GA4 Data API enforces per-property quotas including:

    * **Core reporting**: 200,000 tokens per day, 40,000 tokens per hour, 10 concurrent requests
    * **Real-time reporting**: 75,000 tokens per day, 15,000 tokens per hour

    Tokens are consumed based on the complexity of your queries. Simple queries use fewer tokens than complex ones with many dimensions or large date ranges. See the [GA4 quotas documentation](https://developers.google.com/analytics/devguides/reporting/data/v1/quotas) for full details.
  </Accordion>

  <Accordion title="Why is my GA4 data showing a delay?" icon="clock">
    GA4 processes most data within 24 hours, but full processing (including attribution, session stitching, and some conversions) can take up to 48 hours. Real-time data is available within minutes but represents a subset of the fully processed data. If you need accurate, fully processed metrics, query for dates at least 48 hours in the past.
  </Accordion>

  <Accordion title="Can I access Universal Analytics (UA) data?" icon="circle-info">
    No. Universal Analytics properties stopped processing new data in July 2023. The Google Analytics integration in Relevance AI connects to GA4 only. Historical UA data is no longer accessible via the API.
  </Accordion>

  <Accordion title="How do I find my GA4 property ID?" icon="magnifying-glass">
    In Google Analytics, go to **Admin** > **Property Settings**. Your property ID is the numeric ID shown at the top of the page (e.g., `123456789`). When using the API, you reference it as `properties/123456789`.
  </Accordion>

  <Accordion title="Why am I getting permission errors when creating properties?" icon="triangle-exclamation">
    Creating GA4 properties requires **Administrator** access on the Google Analytics account (not just the property). If you receive a permission error, check your access level under **Admin** > **Account Access Management** and ensure your connected Google account has Administrator rights on the account where you want to create the property.
  </Accordion>
</AccordionGroup>

## Additional resources

<CardGroup>
  <Card title="GA4 Data API reference" icon="chart-line" href="https://developers.google.com/analytics/devguides/reporting/data/v1">
    Complete reference for querying GA4 analytics data, including all available metrics, dimensions, and report types
  </Card>

  <Card title="GA4 Admin API reference" icon="gear" href="https://developers.google.com/analytics/devguides/config/admin/v1">
    Reference for managing GA4 properties, data streams, key events, and account configuration
  </Card>

  <Card title="GA4 key events documentation" icon="flag-checkered" href="https://support.google.com/analytics/answer/9267568">
    Learn how key events work in GA4 and how to use them for conversion measurement
  </Card>

  <Card title="GA4 API quotas" icon="gauge-high" href="https://developers.google.com/analytics/devguides/reporting/data/v1/quotas">
    Details on GA4 Data API rate limits, quota types, and how tokens are calculated
  </Card>
</CardGroup>

## Remove the integration

To disconnect your Google Analytics account from Relevance AI:

1. Navigate to **Integrations & API Keys** from the left sidebar.
2. Find your Google Analytics connection in the integrations list.
3. Click on the integration to expand its details.
4. Click the three-dot menu and select **Remove**.

Removing the integration revokes Relevance AI's access to your Google Analytics data. Any agents or tools using Google Analytics tool steps will stop working until you reconnect. Removing the integration from Relevance AI does not affect your Google Analytics account or data in any way.

***

<CardGroup>
  <Card title="Need help?" icon="life-ring" href="/get-started/support">
    Contact our support team for assistance with the Google Analytics integration
  </Card>

  <Card title="Explore more integrations" icon="puzzle-piece" href="/integrations/introduction">
    Discover other integrations to extend your agent workflows
  </Card>
</CardGroup>


# Google Calendar
Source: https://relevanceai.com/docs/integrations/popular-integrations/google-calendar

Connect your Google Calendar from Relevance AI

## Overview

The Google Calendar integration allows you to seamlessly connect your AI agents with your Google Calendar, enabling powerful automation for scheduling, event management, and calendar-based workflows. This integration bridges the gap between your AI agents and your calendar, allowing for intelligent scheduling assistance, automated event creation, and calendar-based decision making.

## Connecting the Google Calendar Integration

To connect the Google Calendar integration to Relevance AI:

1. Select **Integrations & API Keys** from the left sidebar
2. Find and click on "Google Calendar" in the integrations directory
3. Click "Connect"
4. You'll be redirected to Google's authentication page
5. Select the Google account you want to connect
6. Review and approve the requested permissions
7. Once authenticated, you'll be redirected back to Relevance AI with your Google Calendar now connected

Your Google Calendar integration will now be available for use with your AI agents. You can connect multiple Google Calendar accounts if needed, and manage all connections from the Integrations dashboard.

## Setting Up Triggers with Google Calendar

Google Calendar can serve as a powerful trigger for your AI agents, initiating workflows based on calendar events. Here's how to set up calendar-based triggers:

1. Navigate to your Agent Profile

2. In the trigger section, select "Google Calendar"

3. Configure trigger settings:
   * Select which calendar(s) to monitor

   * Set any filtering criteria (e.g., only events with specific keywords)

   * For time-based triggers, specify the advance notice period

4. Save your trigger configuration

Once set up, your AI agent will automatically activate when the specified calendar events occur, enabling workflows like:

* Sending meeting preparation materials before scheduled calls
* Following up after appointments with summary notes
* Blocking focus time when calendar shows you're busy
* Rescheduling conflicting appointments

## Tools & Tool Steps

The Google Calendar integration provides a rich set of actions that your AI agents can use throughout their workflows. These actions allow your agents to interact with your calendar in sophisticated ways:

### Event Management Actions

* **Create Event**: Schedule new events on your calendar with customizable details
* **Update Event**: Modify existing events (time, location, description, etc.)
* **Delete Event**: Remove events from your calendar
* **Get Event Details**: Retrieve comprehensive information about specific events

### Calendar Query Actions

* **List Events**: Retrieve events within a specified date range
* **Find Available Time**: Identify open slots in your calendar for scheduling
* **Check Availability**: Determine if a specific time period is free or busy
* **Search Events**: Find events matching specific criteria (keywords, attendees, etc.)

### Attendee Management

* **Add Attendees**: Invite participants to calendar events
* **Remove Attendees**: Update event guest list
* **Check Attendee Responses**: View who has accepted, declined, or not responded

### Calendar Management

* **List Calendars**: View all calendars in your Google account
* **Create Calendar**: Set up new calendars for specific purposes
* **Update Calendar Settings**: Modify calendar properties and sharing settings

These actions can be combined in powerful ways to create sophisticated calendar management workflows. For example, your agent could:

* Find available time slots, create a meeting, and send invitations
* Reschedule conflicting appointments based on priority rules
* Create follow-up tasks after calendar events conclude
* Generate meeting agendas based on event descriptions and attendees

## Using the Google Calendar API Tool Step (Advanced)

For advanced use cases requiring deeper integration with Google Calendar, you can leverage the Google Calendar API directly:

1. Create a new tool in your agent profile
2. Scroll down to Tool-steps
3. Add "Google Calendar API" tool-step
4. Select your connected Google Calendar account from the dropdown
5. Configure the API request:
   * Specify the API endpoint (e.g., `/calendars/{calendarId}/events`)
   * Set the HTTP method (GET, POST, PUT, DELETE)
   * Define request parameters and body as needed
   * Configure response handling

This approach gives you full access to the Google Calendar API capabilities, allowing for custom implementations beyond the standard actions. Some advanced use cases include:

* Creating recurring events with complex patterns
* Setting up conference solutions (Google Meet, Zoom) with custom parameters
* Managing calendar resources and room bookings
* Implementing custom notification schemes
* Working with secondary calendars and calendar groups

## Best Practices

* **Permission Management**: Only request the minimum calendar permissions needed for your agent's functionality
* **Error Handling**: Implement robust error handling for calendar conflicts and availability issues
* **User Confirmation**: For critical calendar changes, consider adding user confirmation steps
* **Time Zone Awareness**: Always account for time zone differences when scheduling across regions
* **Privacy Considerations**: Be mindful of calendar data sensitivity and implement appropriate safeguards


# Google Drive
Source: https://relevanceai.com/docs/integrations/popular-integrations/google-drive

Trigger your Agents automatically when files in Google Drive are created, modified, or deleted.

## Overview

Connect your Google Drive account to Relevance AI and configure your Agents to respond automatically when files change, or give them tool steps to read, create, and manage files directly in Drive.

<Info>The Google Drive trigger is currently rolling out in early access. If you don't see it in your account yet, it will become available soon.</Info>

## Connect the integration

1. Navigate to the "Integrations & API Keys" page in the sidebar of your Relevance AI dashboard.
2. Click on "Google Drive" from the available integrations.
3. Click the **+ Connect account** button.
4. In the pop-up window, sign into your Google account and authorize the connection.
5. Once authorized, your Google account will appear in the connected accounts list.

## Setting up triggers

The Google Drive integration lets you trigger Agents automatically when files in a Google Drive change. The trigger fires on every change within the drive (or folders) you watch — created, modified, trashed, or removed — and the event type is included in the message your Agent receives.

### What fires the trigger

A Google Drive trigger fires on each of the following events within the drive or folders you've selected:

* **File created** — a new file is added
* **File modified** — an existing file is updated
* **File deleted** — a file is moved to trash or permanently removed

<Note>Each Google Drive file gets its own ongoing conversation. When the same file changes again, your Agent picks up where it left off rather than starting a new conversation — so it has the full history for each file.</Note>

### How to set up a Google Drive trigger

The setup process differs slightly depending on whether you're configuring a trigger for an Agent or a Workforce.

#### For Agents

1. Open the Agent you want to configure in the Agent builder.
2. In the left sidebar, click the "Triggers" section.
3. Click **+ Add trigger** and select "Google Drive" from the trigger types.
4. Choose the connected Google account you want to use. If you haven't connected one yet, set it up first in the Integrations & API Keys page.
5. Select a **Drive** — defaults to "My Drive", or pick any shared drive you have access to.
6. Optionally, select one or more **Folders** to restrict the watch to only those folders. Leave empty to watch the entire drive.
7. Optionally, enable **Watch for file property changes** to also trigger on changes to custom file properties (not just content changes).
8. Save the trigger.

#### For Workforces

1. Open the Workforce you want to configure in the Workforce builder.
2. From the node tray at the bottom of the screen, drag a trigger node onto the canvas.
3. Click the trigger node to open its configuration panel on the left, then select "Google Drive" as the trigger type.
4. Choose the connected Google account you want to use. If you haven't connected one yet, set it up first in the Integrations & API Keys page.
5. Select a **Drive** — defaults to "My Drive", or pick any shared drive you have access to.
6. Optionally, select one or more **Folders** to restrict the watch.
7. Optionally, enable **Watch for file property changes**.
8. Drag a connection line from the trigger node to the Agent that should respond when the trigger fires.

## Tool steps

Beyond triggers, the Google Drive integration provides pre-built tool steps your Agents can use to read, create, and manage files in Drive. Add these in the Tool builder by searching "Google Drive" in the tool-step picker.

### File operations

* **Google Drive: Upload File** — upload a file to Google Drive
* **Google Drive: Download File** — download a file from Google Drive (specify a MIME type to export Google Workspace files like Docs, Sheets, or Slides)
* **Google Drive: Create File from Text** — create a new file from plain text
* **Google Drive: Create File from Template** — create a new Google Doc from a template, replacing placeholders
* **Google Drive: Update File** — update a file's metadata or content
* **Google Drive: Copy File** — duplicate a file
* **Google Drive: Move File** — move a file from one folder to another
* **Google Drive: Move to Trash** — send a file or folder to trash
* **Google Drive: Delete File** — permanently delete a file or folder

### Search and discovery

* **Google Drive: List Files** — list files from a specific folder
* **Google Drive: Find File** — search for a file by name
* **Google Drive: Find Folder** — search for a folder by name
* **Google Drive: Find Forms** — list or search Google Form documents
* **Google Drive: Find Spreadsheets** — search for Google Sheets by name
* **Google Drive: Get File by ID** — fetch file metadata by ID
* **Google Drive: Get Folder ID for Path** — resolve a folder path to its ID

### Folders

* **Google Drive: Create Folder** — create a new empty folder

### Shared drives

* **Google Drive: Create Shared Drive** — create a new shared drive
* **Google Drive: Get Shared Drive** — fetch metadata for a shared drive
* **Google Drive: Search Shared Drives** — search for shared drives
* **Google Drive: Update Shared Drive** — update an existing shared drive
* **Google Drive: Delete Shared Drive** — delete an empty shared drive

### Sharing and permissions

* **Google Drive: Share File or Folder** — add a sharing permission and return a sharing URL
* **Google Drive: List Access Proposals** — list pending requests for access to a file or folder
* **Google Drive: Resolve Access Proposal** — accept or deny a request for access

## Use the Google Drive API tool step (advanced)

For actions that aren't covered by a dedicated tool step, the **Google Drive API call** tool step lets your Agent make authorized requests directly to the Google Drive API.

1. Create a new Tool in Relevance AI.
2. Scroll to the Tool steps section.
3. Add the **Google Drive API call** tool step.
4. Select your connected Google account in the dropdown.
5. Configure the request — endpoint, method, parameters, and body — based on the [Google Drive API reference](https://developers.google.com/drive/api/reference/rest/v3).

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What's the difference between a Google Drive trigger and Google Drive as a Knowledge source?">
    Triggers fire your Agent automatically when files in Google Drive change — created, modified, or deleted. A Knowledge source gives your Agent read access to specific files so it can reference them during a conversation. You can use both together: a trigger to start the conversation, and a Knowledge source to give the Agent ongoing context.
  </Accordion>

  <Accordion title="Can Agents work with shared drives or only My Drive?">
    Both. The trigger config supports selecting either My Drive or any shared drive you have access to, and the tool steps include a full set of shared-drive actions for creating, searching, and updating shared drives.
  </Accordion>

  <Accordion title="Can I connect multiple Google accounts?">
    Yes. Connect each account in the Integrations & API Keys page, then choose which connected account each trigger or tool step uses.
  </Accordion>

  <Accordion title="Does it work with personal Gmail and Google Workspace accounts?">
    Both work. Sign in with whichever account has access to the files or folders you want your Agent to watch or act on.
  </Accordion>
</AccordionGroup>

## Google Drive as a Knowledge source

You can also use Google Drive as a Knowledge source, giving your Agents direct read access to files for retrieval and reference.

[Learn more about Google Drive as a Knowledge source](/docs/build/knowledge/integrated-knowledge-sources/google-drive).


# Google Sheets
Source: https://relevanceai.com/docs/integrations/popular-integrations/google-sheets

Connect to Google Sheets from Relevance AI.

## Overview

The Google Sheets integration allows you to connect your Relevance AI agents with Google Sheets, enabling powerful data workflows between your spreadsheets and AI agents. This integration empowers you to automate data extraction, analysis, and manipulation tasks directly from your Google Sheets documents, as well as use spreadsheet data to trigger agent workflows.

## Connecting the Google Sheets Integration

To connect the Google Sheets integration with Relevance AI:

1. Click on **Integrations & API Keys** in the left sidebar
2. Find and select "Google Sheets" from the available integrations
3. Click "Connect"
4. You'll be redirected to Google's authentication page
5. Grant Relevance AI permission to access your Google Sheets
6. Once authenticated, you'll be redirected back to Relevance AI with the integration now connected

## Setting Up Triggers with Google Sheets *(coming soon)*

The Google Sheets integration can be configured to trigger your AI agents based on specific spreadsheet events or conditions:

### Trigger Types

* **New Row Added**: Activate your agent when a new row is added to a specified sheet
* **Cell Value Changed**: Trigger your agent when values in specific cells or columns are modified
* **Threshold Reached**: Start your agent when a numeric value in your sheet crosses a defined threshold
* **Scheduled Updates**: Configure your agent to process sheet data at regular intervals

### Setting Up a Trigger

1. Navigate to your agent's profile
2. Select "Google Sheets" from the trigger options
3. Choose the specific Google Sheet you want to monitor
4. Select the trigger type (e.g., New Row Added)
5. Configure any additional parameters specific to your chosen trigger
6. Save your trigger configuration

Your agent will now automatically activate when the specified condition is met in your Google Sheets document.

## Tools & Tool Steps

The Google Sheets integration provides a variety of powerful actions that your agents can perform. These actions can be incorporated into your agent's workflow as tool steps, allowing for seamless interaction with your spreadsheet data.

### Available Actions

* **Read Sheet Data**: Extract data from specific sheets, ranges, or cells
* **Write to Sheet**: Add or update data in your spreadsheets
* **Create New Sheet**: Generate new spreadsheets with specified data
* **Format Cells**: Apply formatting to cells based on conditions or data analysis
* **Filter Data**: Extract specific information based on custom criteria
* **Sort Data**: Organize spreadsheet information in meaningful ways
* **Append Rows**: Add new rows of data to existing sheets
* **Delete Data**: Remove specific cells, rows, or columns

These actions represent just a small sample of what's possible with the Google Sheets integration. The tool library contains many more specialized actions to help you build powerful data workflows.

### Using Google Sheets Actions in Your Workflow

1. In your agent's workflow, add a new tool step
2. Search for "Google Sheets" in the tool library
3. Browse the available actions and select the one you need
4. Configure the action parameters (sheet ID, range, data to write, etc.)
5. Connect this action to other steps in your workflow

## Using the Google Sheets API Tool Step (Advanced)

For advanced users who need custom functionality beyond the pre-built actions, you can leverage the Google Sheets API directly:

1. Create a new tool in your agent's workflow
2. Scroll down to Tool-steps
3. Add "Google Sheets API" tool-step
4. Select your connected Google Sheets account from the dropdown
5. Specify the API endpoint and parameters for your custom request
6. Configure how the response data should be processed

This approach gives you full access to the Google Sheets API capabilities, allowing for highly customized interactions with your spreadsheet data.

## Use Cases

### Data Analysis and Reporting

* Extract data from multiple sheets for comprehensive analysis
* Generate automated reports based on spreadsheet data
* Create visualizations and summaries of key metrics

### Data Collection and Management

* Automatically organize and categorize incoming data
* Clean and standardize information across multiple sheets
* Maintain data integrity through automated validation

### Workflow Automation

* Trigger notifications when important spreadsheet values change
* Automatically update related systems when data is modified
* Create approval workflows based on spreadsheet entries

### Customer and Lead Management

* Track customer interactions in spreadsheets and trigger follow-ups
* Analyze lead data to prioritize high-value prospects
* Generate personalized communications based on spreadsheet data

## Best Practices

* **Use Descriptive Names**: When setting up Google Sheets actions, use clear names that describe what the action does
* **Handle Errors Gracefully**: Configure error handling for cases where sheets might be unavailable or data formats unexpected
* **Limit Data Access**: Only request access to the specific sheets and data ranges your agent needs
* **Test Thoroughly**: Before deploying in production, test your Google Sheets integration with sample data
* **Monitor Usage**: Keep track of API usage to stay within Google's rate limits

## Related Features

[Data Visualization Tools](https://relevanceai.com/docs/features/data-visualization) - Combine Google Sheets data with powerful visualization capabilities to create compelling visual representations of your insights.

[Workflow Automation](https://relevanceai.com/docs/features/workflow-automation) - Learn how to build complex workflows that incorporate Google Sheets data alongside other integrations.

[Data Processing](https://relevanceai.com/docs/features/data-processing) - Discover techniques for cleaning, transforming, and analyzing data from Google Sheets within your AI workflows.


# HubSpot
Source: https://relevanceai.com/docs/integrations/popular-integrations/hubspot

Interact with your HubSpot CRM with Relevance.

## Overview

HubSpot is designed to help your manage and grow your customer relationships through a wide range of tools for marketing, sales and customer service. With Relevance AI, your agents can interact with HubSpot much like your human team would.

Below, we'll show you how to get started with the HubSpot integration, including how to connect your HubSpot account, how to trigger agentic workflows based on HubSpot workflow triggers, and how to complete common HubSpot activities with your agents.

## Connecting the Hubspot integration

To add HubSpot as an integration, all you need to do is sign-into your HubSpot account via the integrations page:

1. Go to the **Integrations & API Keys** page in the sidebar.
2. Click on "HubSpot".
3. Click on the Add Integration button.
4. In the pop-up window, sign-into your HubSpot account.

## Setup HubSpot as an Agent Trigger

A trigger is an event that triggers your agent to start working. Once you have create a new agent, you can add HubSpot as a trigger:

<div>
  <iframe title="Create an agent from scratch on Relevance AI" />
</div>

### Trigger your agent to start working from a HubSpot workflow

Next, we need to create a workflow in HubSpot, which will send a message to your connected agent every time an event you care about happens.

There are three parts to this process:

1. Create a HubSpot workflow.
2. Create a HubSpot trigger.
3. Create an action which sends a message to your Relevance AI agent.

#### Step 1: Create a HubSpot workflow

In HubSpot, a workflow is made up of a series of automated actions that are perform every time a specific event happens. In this example, we'll create a Contact-based workflow which sends a message to your agent when a new contact added to your HubSpot CRM meets some criteria you define (HubSpot trigger).

Let's create a contact-based workflow:

<div>
  <iframe title="Create an agent from scratch on Relevance AI" />
</div>

#### Step 2: Create a HubSpot workflow trigger

A HubSpot trigger is an event or condition that starts the automated workflow. In the case of a contact-based workflow, a trigger is a condition related to a contact. It can be based on a contact filling out a form, clicking a link in an email, acheiving a specific lead score, or one of their properties being updated. It's entirely up to you.

<div>
  <iframe title="Create an agent from scratch on Relevance AI" />
</div>

#### Step 3: Add Relevance AI Trigger action

Once your workflow and trigger is setup, you can add an action that will send a message to your agent from HubSpot (triggering your agentic workflow):

<div>
  <iframe title="Create an agent from scratch on Relevance AI" />
</div>

## Get your agent to complete common HubSpot activities

Besides triggering your agent to start working from HubSpot, you can also use the HubSpot API call step in the Tool builder. This tool step allows your agent to complete HubSpot-specific activities, like managing Contacts, Notes, Tasks and more.

### Use HubSpot API tool-step

You can build custom tools that perform HubSpot activities, by using the HubSpot API Call tool-step:

1. Create a new tool.
2. Scroll down to Tool-steps.
3. Add HubSpot API tool-step.
4. Select your connected HubSpot account in the dropdown.

Below are examples of common HubSpot activities your agents can perform.

#### Get an existing Contact's details.

<img alt="" />

#### Update a contact's properties.

<img alt="" />

#### Create a note (e.g. lead qual research).

<img alt="" />

#### Create a task

<img alt="" />

#### Mark a task as complete

<img alt="" />

#### Retrieve HubSpot engagements (all activities a customer has performed)

<img alt="" />

<img alt="" />

## Remove HubSpot integration

You can remove HubSpot as a Trigger in your agent settings, by clicking the three dots next to your connected account, and then "Remove".

If you want to remove the HubSpot integration completely, you can:

1. Go to the **Integrations & API Keys** page from the sidebar.
2. Select Zoom from the list.
3. Click "..." on the account you want to remove.
4. Click Remove.

***

## HubSpot Assistant

The HubSpot Assistant is a pre-built agent available from the [Marketplace](https://app.relevanceai.com/marketplace) that manages your HubSpot CRM through plain English. Handle contacts, deals, companies, and more through simple conversation.

<Card title="Clone HubSpot Assistant" icon="chart-network" href="https://app.relevanceai.com/marketplace">
  Search "HubSpot Assistant" in the Marketplace to clone it and start managing your CRM through conversation
</Card>

***

## Follow along on YouTube

We also have a number of YouTube tutorials you can use to follow along with and learn how to build Agents.

<Tabs>
  <Tab title="Building a HubSpot-Triggered Agent">
    <Frame>
      <iframe title="YouTube video player" />
    </Frame>
  </Tab>

  <Tab title="HubSpot Assistant Demo">
    <Frame>
      <iframe title="We Built an AI Agent That Runs HubSpot for You" />
    </Frame>
  </Tab>
</Tabs>


# LinkedIn
Source: https://relevanceai.com/docs/integrations/popular-integrations/linkedin

Interact with your LinkedIn account with Relevance.

Connect your LinkedIn account to Relevance AI and supercharge your networking, prospecting, and engagement capabilities with AI-powered automation. The LinkedIn Integration enables your agents to interact with LinkedIn much like a human team member would, creating a seamless bridge between your AI workforce and one of the world's most valuable professional networks.

## Connect the integration

Getting started with the LinkedIn Integration is straightforward:

1. Navigate to the "Integrations & API Keys" page in the sidebar of your Relevance AI dashboard
2. Click on "LinkedIn" from the available integrations
3. Click the "Add Integration" button
4. In the pop-up window, sign into your LinkedIn account and authorize the connection
5. Once authorized, your LinkedIn account will appear in the connected accounts list

## Setting up triggers

The LinkedIn Integration offers powerful trigger capabilities that can initiate your AI agents based on specific LinkedIn events. This allows your agents to respond automatically to important network activities.

<Warning>The LinkedIn Trigger is a Premium trigger, available on the Pro plan and above.</Warning>

### Available LinkedIn triggers

Set up your agents to start working when:

* **All messages received** - Triggers when any message is received on LinkedIn
* **Outreach replies only** - Triggers only when replies to outreach messages are received

<Note>Connection accept notifications are only tracked if the connection request was originally sent through Relevance AI's LinkedIn integration. This is a technical limitation of the LinkedIn integration.</Note>

These triggers are available for both agents and workforces.

### How to set up a LinkedIn trigger

The setup process differs slightly depending on whether you're configuring a trigger for an agent or a workforce. Follow the instructions below based on your use case.

#### For Agents

1. Navigate to the agent builder by selecting the agent you want to configure with a LinkedIn trigger.

2. In the left sidebar of the agent builder, locate and click on the "Triggers" section.

3. Click the "Add trigger" button, then select "LinkedIn" from the available trigger types.

4. Choose the connected LinkedIn account you want to use for this trigger from the dropdown menu. If you haven't connected a LinkedIn account yet, you'll need to do so first in the Integrations & API Keys page.

5. Select the specific trigger event you want to use:
   * **All messages received** - Triggers when any message is received on LinkedIn
   * **Outreach replies only** - Triggers only when replies to outreach messages are received

6. If available on your plan, configure queue work hours to control when your agent should process LinkedIn triggers. This allows you to set specific times when your agent should be active.

#### For Workforces

1. Navigate to the workforce builder by selecting the workforce you want to configure with a LinkedIn trigger.

2. In the Build section of your workforce, locate the "Triggers" section in the node palette on the left side. Drag a trigger node onto your workspace.

3. Click on the trigger node you just added to open its configuration panel, then select "LinkedIn" as the trigger type.

4. Choose the connected LinkedIn account you want to use for this trigger from the dropdown menu. If you haven't connected a LinkedIn account yet, you'll need to do so first in the Integrations & API Keys page.

5. Select the specific trigger event you want to use:
   * **All messages received** - Triggers when any message is received on LinkedIn
   * **Outreach replies only** - Triggers only when replies to outreach messages are received

6. If available on your plan, configure queue work hours to control when your workforce should process LinkedIn triggers. This allows you to set specific times when your agents should be active.

7. Drag a connection line from the trigger node to the agent in your workforce that should respond when the trigger activates.

## Tool steps for LinkedIn

You can also add LinkedIn Tool steps to your Tools to interact with LinkedIn to get company and profile information, and perform actions in LinkedIn.

<Snippet />


# Lusha
Source: https://relevanceai.com/docs/integrations/popular-integrations/lusha

Lusha is a leading B2B sales intelligence and lead enrichment platform trusted by over 1.5 million users worldwide.

Lusha is a leading B2B sales intelligence and lead enrichment platform trusted by over 1.5 million users worldwide. With Relevance AI's Lusha integration, you can seamlessly enrich your leads with accurate contact and company information, enabling your AI agents to automate prospect research, qualify leads, and build comprehensive sales pipelines with verified data.

<Warning>
  The Lusha integration in Relevance AI was built by Relevance AI, and is therefore supported by our team, not Lusha. If you have a question or issue with using Lusha in Relevance AI, please [reach out to our support team](/docs/get-started/support). If you have a question or issue that is only about Lusha, you can [reach out to Lusha support](https://www.lusha.com/contact/).
</Warning>

## Connect the integration

Connecting your Lusha account to Relevance AI is a straightforward process:

1. Go to the "Integrations" page in the sidebar of your Relevance AI dashboard.
2. Click on "Lusha" from the available integrations.
3. Click on the "Add Integration" button.
4. Enter your Lusha API key when prompted.
5. Once authenticated, your Lusha account will appear as a connected integration.

<Tip>
  You can find your Lusha API key in your Lusha account settings under API & Integrations. If you don't have a Lusha account yet, you can [sign up here](https://www.lusha.com/).
</Tip>

## Tool steps for Lusha

The Lusha integration provides a comprehensive set of actions that your agents can use to enrich leads, find contacts, and gather company intelligence. These actions can be incorporated into your agent's workflows as tool steps, enabling sophisticated sales automation and prospect research capabilities.

### Contact Enrichment

<CardGroup>
  <Card title="Enrich Contacts" icon="user-plus">
    Enrich existing contact records with verified email addresses, phone numbers, and professional details
  </Card>

  <Card title="Find Contact" icon="magnifying-glass">
    Find specific contacts based on name, company, or other identifying information
  </Card>

  <Card title="Search Contacts" icon="users">
    Search for contacts matching specific criteria across Lusha's database
  </Card>
</CardGroup>

### Company Intelligence

<CardGroup>
  <Card title="Enrich Companies" icon="building">
    Enrich company records with firmographic data, employee counts, and industry information
  </Card>

  <Card title="Find Company" icon="magnifying-glass">
    Find specific companies based on name, domain, or other identifiers
  </Card>

  <Card title="Search Companies" icon="buildings">
    Search for companies matching your ideal customer profile criteria
  </Card>
</CardGroup>

### Advanced API Access

<CardGroup>
  <Card title="Lusha API Call" icon="code">
    Make custom API calls to access any Lusha API endpoint for advanced use cases
  </Card>
</CardGroup>

<Tip>
  Type "Lusha" in the tool step search bar to see all available Lusha actions when building your tools.
</Tip>

## Use the integration's API tool step (Advanced)

In addition to the pre-built actions available in the tool directory, you can build custom tools that perform Lusha-specific activities using the Lusha API Call tool step.

### How to use the Lusha API Call tool step

<Steps>
  <Step title="Create a new tool">
    Create a new tool in Relevance AI or open an existing tool you want to add Lusha functionality to.
  </Step>

  <Step title="Add the Lusha API Call tool step">
    1. Scroll down to Tool-steps
    2. Search for "Lusha API Call" in the tool step search bar
    3. Add the Lusha API Call tool step to your workflow
  </Step>

  <Step title="Select your Lusha account">
    Select your connected Lusha account from the dropdown menu.
  </Step>

  <Step title="Configure the API endpoint">
    Configure the API endpoint, method, and parameters according to your needs:

    * **Method**: Select the HTTP method (GET, POST)
    * **Endpoint**: Enter the API endpoint path (e.g., `/person`)
    * **Body**: Add any required request body data with search criteria
  </Step>

  <Step title="Test your API call">
    Test your configuration to ensure it works correctly before deploying.
  </Step>
</Steps>

### Example: Enriching a Contact by Email

Here's a practical example of using the Lusha API Call tool step to enrich a contact using their email address:

**API Endpoint**: `POST /person`

**Configuration**:

```json theme={null}
{
  "method": "POST",
  "endpoint": "/person",
  "body": {
    "email": "john.doe@example.com"
  }
}
```

This configuration:

* Uses the POST method to request contact enrichment
* Specifies the `/person` endpoint for individual contact lookup
* Provides the email address as the search parameter
* Returns verified contact information including phone numbers, job title, company details, and social profiles

<Info>
  You can find Lusha's complete API documentation at [https://www.lusha.com/api-documentation/](https://www.lusha.com/api-documentation/).
</Info>

### Common Lusha API Endpoints

Here are some commonly used Lusha API endpoints you can use with the API Call tool step:

<AccordionGroup>
  <Accordion title="Person Enrichment" icon="user">
    * **Enrich by email**: `POST /person` with `{"email": "user@example.com"}`
    * **Enrich by LinkedIn**: `POST /person` with `{"linkedInUrl": "https://linkedin.com/in/username"}`
    * **Enrich by name & company**: `POST /person` with `{"firstName": "John", "lastName": "Doe", "company": "Acme Corp"}`

    Returns: Email addresses, phone numbers, job title, company, location, social profiles

    [View API Documentation](https://www.lusha.com/api-documentation/)
  </Accordion>

  <Accordion title="Company Enrichment" icon="building">
    * **Enrich by domain**: `POST /company` with `{"domain": "example.com"}`
    * **Enrich by company name**: `POST /company` with `{"name": "Acme Corporation"}`

    Returns: Company size, industry, location, revenue, technologies used, social profiles

    [View API Documentation](https://www.lusha.com/api-documentation/)
  </Accordion>

  <Accordion title="Bulk Enrichment" icon="layer-group">
    * **Bulk person enrichment**: `POST /person/bulk` with array of person objects
    * **Bulk company enrichment**: `POST /company/bulk` with array of company objects

    Allows you to enrich multiple contacts or companies in a single API call for efficiency.

    [View API Documentation](https://www.lusha.com/api-documentation/)
  </Accordion>

  <Accordion title="Search & Discovery" icon="magnifying-glass">
    * **Search contacts**: Use search endpoints with filters for job title, company, location, industry
    * **Find decision makers**: Search for specific roles within target companies

    Useful for building prospect lists matching your ideal customer profile.

    [View API Documentation](https://www.lusha.com/api-documentation/)
  </Accordion>
</AccordionGroup>

## Example use cases

Here are some ways you can leverage the Lusha integration with your agents:

<AccordionGroup>
  <Accordion title="Automated Lead Enrichment" icon="user-plus">
    Create an agent that automatically enriches incoming leads from your CRM or forms with verified contact information. The agent can take an email address or LinkedIn profile, use Lusha to gather phone numbers, job titles, and company details, then update your CRM with the enriched data.
  </Accordion>

  <Accordion title="Prospect Research Agent" icon="magnifying-glass-chart">
    Build an agent that conducts deep research on prospects before sales calls. Given a LinkedIn URL or company name, the agent uses Lusha to gather contact details, company information, and decision-maker profiles, then compiles a comprehensive research brief for your sales team.
  </Accordion>

  <Accordion title="Lead Qualification Bot" icon="filter">
    Deploy an agent that automatically qualifies leads based on enriched data. The agent uses Lusha to gather company size, industry, and contact seniority, then scores and routes leads to the appropriate sales representatives based on your qualification criteria.
  </Accordion>

  <Accordion title="Outbound List Builder" icon="list-check">
    Create an agent that builds targeted prospect lists for outbound campaigns. The agent searches Lusha's database for contacts matching your ideal customer profile (job title, company size, industry, location), enriches the results, and exports them to your sales engagement platform.
  </Accordion>

  <Accordion title="CRM Data Enhancement" icon="database">
    Build an agent that continuously maintains data quality in your CRM. The agent identifies incomplete or outdated contact records, uses Lusha to find current information, and automatically updates your CRM with verified email addresses, phone numbers, and job titles.
  </Accordion>

  <Accordion title="Account-Based Marketing (ABM) Research" icon="bullseye">
    Deploy an agent that supports ABM campaigns by researching target accounts. The agent uses Lusha to identify key decision-makers within target companies, gather their contact information, and map out the organizational structure for strategic outreach.
  </Accordion>

  <Accordion title="Recruitment Sourcing Agent" icon="user-tie">
    Create an agent that helps recruiters find and contact qualified candidates. The agent searches for professionals with specific skills and experience, uses Lusha to find their contact information, and prepares personalized outreach messages.
  </Accordion>

  <Accordion title="Partnership Development Bot" icon="handshake">
    Build an agent that identifies and researches potential business partners. The agent finds companies in complementary industries, uses Lusha to identify partnership decision-makers, and gathers contact information for outreach.
  </Accordion>
</AccordionGroup>

## Best practices

<AccordionGroup>
  <Accordion title="Data Privacy & Compliance" icon="shield-check">
    **Always ensure compliance with data privacy regulations:**

    * Only use Lusha data for legitimate business purposes
    * Respect GDPR, CCPA, and other privacy regulations in your region
    * Include proper opt-out mechanisms in your outreach
    * Store enriched data securely and delete when no longer needed
    * Review Lusha's terms of service and acceptable use policy

    Lusha is GDPR and CCPA compliant, but you're responsible for how you use the data.
  </Accordion>

  <Accordion title="Credit Management" icon="coins">
    **Optimize your Lusha credit usage:**

    * Each enrichment request consumes Lusha credits
    * Use bulk enrichment endpoints when processing multiple records
    * Cache enriched data to avoid duplicate lookups
    * Implement logic to check if data already exists before enriching
    * Monitor your credit usage through Lusha's dashboard
    * Set up alerts when credits are running low
  </Accordion>

  <Accordion title="Data Quality" icon="star">
    **Maximize the quality of enriched data:**

    * Provide as much information as possible in your search queries (email + name + company is better than email alone)
    * Validate enriched data before using it in outreach
    * Implement fallback logic when Lusha doesn't find a match
    * Regularly update enriched data as contact information changes
    * Use Lusha's confidence scores to assess data reliability
  </Accordion>

  <Accordion title="Rate Limiting" icon="gauge">
    **Handle API rate limits gracefully:**

    * Lusha enforces rate limits to ensure service quality
    * Implement retry logic with exponential backoff for rate limit errors
    * Use bulk endpoints to reduce the number of API calls
    * Distribute enrichment tasks over time for large datasets
    * Monitor API response codes and handle errors appropriately
  </Accordion>
</AccordionGroup>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What is Lusha and why should I use it?">
    Lusha is a B2B sales intelligence platform that provides verified contact and company information. It helps sales and marketing teams find accurate email addresses, phone numbers, and company details for prospects. Using Lusha with Relevance AI allows you to automate lead enrichment, prospect research, and data quality maintenance at scale.
  </Accordion>

  <Accordion title="Do I need a Lusha account to use this integration?">
    Yes, you need an active Lusha account with API access. You can sign up for Lusha at [https://www.lusha.com/](https://www.lusha.com/). Lusha offers various plans including a free tier with limited credits. You'll need to generate an API key from your Lusha account settings to connect it to Relevance AI.
  </Accordion>

  <Accordion title="How do I get my Lusha API key?">
    To get your Lusha API key:

    1. Log in to your Lusha account
    2. Navigate to Settings or Account Settings
    3. Look for "API & Integrations" or "API Access"
    4. Generate a new API key or copy your existing key
    5. Use this key when connecting Lusha to Relevance AI

    Keep your API key secure and never share it publicly.
  </Accordion>

  <Accordion title="What does 'beta' status mean for the Lusha integration?">
    The beta status indicates that while the Lusha integration is fully functional and available for use, we're still gathering user feedback and may make refinements to improve the experience. We recommend testing your workflows thoroughly before deploying them in production. If you encounter any issues or have suggestions, please contact our support team.
  </Accordion>

  <Accordion title="How many Lusha credits does each enrichment use?">
    Credit usage depends on the type of enrichment and the data returned. Typically:

    * Contact enrichment: 1-2 credits per successful lookup
    * Company enrichment: 1 credit per successful lookup
    * Bulk operations: Credits per record in the batch

    Check your Lusha plan details for specific credit allocations and pricing. You can monitor your credit usage in your Lusha dashboard.
  </Accordion>

  <Accordion title="What happens if Lusha doesn't find a match?">
    If Lusha doesn't find information for a contact or company, the API will return an empty result or indicate no match was found. Your agent should include logic to handle these cases, such as:

    * Trying alternative search parameters
    * Using fallback data sources
    * Flagging the record for manual research
    * Skipping to the next record in bulk operations

    Not all contacts are in Lusha's database, so plan for partial match rates.
  </Accordion>

  <Accordion title="Can I use Lusha for cold outreach?">
    Yes, but you must comply with all applicable laws and regulations:

    * Follow CAN-SPAM, GDPR, CCPA, and other relevant regulations
    * Include proper identification and opt-out mechanisms
    * Only contact people with legitimate business interest
    * Respect do-not-contact lists and preferences
    * Review Lusha's acceptable use policy

    Lusha provides the data, but you're responsible for how you use it.
  </Accordion>

  <Accordion title="How accurate is Lusha's data?">
    Lusha maintains high data accuracy through continuous verification and updates. However, contact information can change over time. Best practices:

    * Use the most recent enrichment data available
    * Implement email verification before sending campaigns
    * Update your records regularly
    * Use Lusha's confidence scores when available
    * Have a process for handling bounced emails or wrong numbers
  </Accordion>

  <Accordion title="Can I enrich contacts from LinkedIn profiles?">
    Yes! Lusha specializes in enriching LinkedIn profiles. You can provide a LinkedIn URL to the Lusha enrichment tool steps, and it will return verified contact information including email addresses and phone numbers. This is particularly useful for sales prospecting and recruitment workflows.
  </Accordion>

  <Accordion title="What's the difference between 'Find' and 'Search' tool steps?">
    * **Find** tool steps: Used when you have specific identifying information (email, LinkedIn URL, company domain) and want to retrieve that specific contact or company's details
    * **Search** tool steps: Used when you want to discover contacts or companies matching certain criteria (job title, industry, location, company size)

    Use "Find" for enrichment, "Search" for prospecting and list building.
  </Accordion>

  <Accordion title="Can I integrate Lusha with my CRM?">
    Yes! You can build agents that connect Lusha with popular CRMs like Salesforce, HubSpot, and others available in Relevance AI. Your agent can:

    * Pull contacts from your CRM
    * Enrich them with Lusha
    * Update the CRM with enriched data
    * All automatically on a schedule or trigger

    Check our [integrations page](/docs/integrations/introduction) for available CRM connections.
  </Accordion>

  <Accordion title="Are there rate limits for the Lusha API?">
    Yes, Lusha enforces rate limits to ensure service quality for all users. The specific limits depend on your Lusha plan. If you hit rate limits:

    * Your agent will receive a rate limit error
    * Implement retry logic with delays
    * Use bulk endpoints when processing many records
    * Contact Lusha support if you need higher limits

    Design your workflows to handle rate limits gracefully.
  </Accordion>
</AccordionGroup>


# Marketo
Source: https://relevanceai.com/docs/integrations/popular-integrations/marketo

Marketo is a leading marketing automation platform for lead management, email marketing, and campaign analytics.

Marketo is a leading marketing automation platform that helps businesses manage leads, execute email campaigns, and analyze marketing performance. With Relevance AI's Marketo integration, you can seamlessly connect your Marketo instance to your AI agents, enabling them to retrieve campaign metrics, manage leads, access email statistics, and automate marketing workflows.

<Warning>
  The Marketo integration in Relevance AI was built by Relevance AI, and is therefore supported by our team, not Marketo. If you have a question or issue with using Marketo in Relevance AI, please [reach out to our support team](/docs/get-started/support). If you have a question or issue that is only about Marketo, you can [reach out to Adobe Marketo support](https://experienceleague.adobe.com/en/docs/marketo/using/home).
</Warning>

## Connect the integration

Connecting your Marketo account to Relevance AI uses API Tool Steps with 2-legged OAuth 2.0 authentication. Follow these steps to set up the connection:

<Steps>
  <Step title="Get your Marketo API credentials">
    You'll need three pieces of information from your Marketo instance:

    1. **Client ID**
    2. **Client Secret**
    3. **REST API Endpoint**

    To obtain these credentials:

    1. Log into your Marketo instance as an administrator
    2. Navigate to **Admin** > **Integration** > **LaunchPoint**
    3. Click **New** > **New Service**
    4. Configure the service:
       * **Display Name**: Enter a name (e.g., "Relevance AI Integration")
       * **Service**: Select "Custom"
       * **Description**: Optional description
       * **API Only User**: Select an API-only user (or create one in Admin > Users & Roles)
    5. Click **Create**
    6. Click **View Details** on your newly created service
    7. Copy the **Client ID** and **Client Secret**

    To get your REST API Endpoint:

    1. Navigate to **Admin** > **Integration** > **Web Services**
    2. Find the **REST API** section
    3. Copy the **Endpoint** URL (e.g., `https://123-ABC-456.mktorest.com/rest`)
  </Step>

  <Step title="Store credentials in Relevance AI">
    When configuring the Marketo API Call tool step in your tools, you'll need to provide these credentials for authentication. Keep them secure and accessible for when you build your Marketo-connected tools.
  </Step>

  <Step title="Test your connection">
    Once you've added the credentials to a tool step, test the connection by making a simple API call (e.g., retrieving leads or campaigns) to ensure everything is configured correctly.
  </Step>
</Steps>

<Info>
  **Authentication Method**: Marketo uses 2-legged OAuth 2.0. As of June 2025, authentication requires a Bearer token in the Authorization header. The Marketo API Call tool step handles token generation and refresh automatically.
</Info>

## Tool steps for Marketo

The Marketo integration provides access to Marketo's comprehensive REST API through the API Call tool step, enabling you to interact with all Marketo endpoints and functionality.

### Marketo API Call

<Card title="Marketo API Call" icon="plug">
  Make custom API calls to any Marketo REST API endpoint. This tool step provides full access to Marketo's API, allowing you to retrieve campaign metrics, manage leads, access email statistics, and more.
</Card>

<Tip>
  Type "Marketo" in the tool step search bar to find the Marketo API Call action when building your tools.
</Tip>

## Use the Marketo API Call tool step (Advanced)

The Marketo API Call tool step gives you complete access to Marketo's REST API, allowing you to implement any functionality available in the API, including lead management, campaign operations, email program analytics, and more.

<Warning>
  **Authentication Required**: As of June 2025, Marketo API calls require a Bearer token in the Authorization header. The Marketo API Call tool step handles this automatically when you provide your Client ID, Client Secret, and REST API Endpoint.
</Warning>

### How to use the Marketo API Call tool step

<Steps>
  <Step title="Create a new tool">
    Create a new tool in Relevance AI or open an existing tool you want to add Marketo functionality to.
  </Step>

  <Step title="Add the Marketo API Call tool step">
    1. Scroll down to Tool-steps
    2. Search for "Marketo API Call" in the tool step search bar
    3. Add the Marketo API Call tool step to your workflow
  </Step>

  <Step title="Configure authentication">
    Provide your Marketo credentials:

    * **Client ID**: Your Marketo API Client ID
    * **Client Secret**: Your Marketo API Client Secret
    * **REST API Endpoint**: Your Marketo REST API endpoint URL

    The tool step will automatically handle OAuth token generation and include the Bearer token in the Authorization header.
  </Step>

  <Step title="Configure the API endpoint">
    Configure the API endpoint, method, and parameters according to your needs:

    * **Method**: Select the HTTP method (GET, POST, DELETE)
    * **Endpoint**: Enter the API endpoint path (e.g., `/rest/v1/leads.json`)
    * **Query Parameters**: Add any required query parameters
    * **Body**: Add any required request body data (for POST requests)
  </Step>

  <Step title="Test your API call">
    Test your configuration to ensure it works correctly before deploying. Verify that you're receiving the expected data from Marketo.
  </Step>
</Steps>

### Example: Get Leads from Marketo

Here's a practical example of using the Marketo API Call tool step to retrieve leads:

**API Endpoint**: `GET /rest/v1/leads.json`

**Configuration**:

```json theme={null}
{
  "method": "GET",
  "endpoint": "/rest/v1/leads.json",
  "query_parameters": {
    "filterType": "email",
    "filterValues": "user@example.com"
  }
}
```

This configuration:

* Uses the GET method to retrieve leads
* Filters leads by email address
* Returns lead information including custom fields

### Example: Trigger a Campaign

**API Endpoint**: `POST /rest/v1/campaigns/{campaignId}/trigger.json`

**Configuration**:

```json theme={null}
{
  "method": "POST",
  "endpoint": "/rest/v1/campaigns/1234/trigger.json",
  "body": {
    "input": {
      "leads": [
        {
          "id": 5678
        }
      ]
    }
  }
}
```

This configuration:

* Uses the POST method to trigger a campaign
* Specifies the campaign ID in the endpoint
* Provides lead IDs to trigger the campaign for

<Info>
  You can find Marketo's complete API documentation at [https://developer.adobe.com/marketo-apis/](https://developer.adobe.com/marketo-apis/).
</Info>

## Common Marketo API endpoints

Here are some commonly used Marketo API endpoints you can use with the Marketo API Call tool step:

<AccordionGroup>
  <Accordion title="Leads" icon="user">
    * **Get leads**: `GET /rest/v1/leads.json?filterType={filterType}&filterValues={filterValues}`
    * **Create/update leads**: `POST /rest/v1/leads.json`
    * **Get lead by filter type**: `GET /rest/v1/leads.json?filterType=id&filterValues=1,2,3`
    * **Get lead activities**: `GET /rest/v1/activities/leadchanges.json`

    [View API Documentation](https://experienceleague.adobe.com/en/docs/marketo-developer/marketo/rest/lead-database/leads)
  </Accordion>

  <Accordion title="Campaigns" icon="bullhorn">
    * **Get campaigns**: `GET /rest/v1/campaigns.json`
    * **Trigger campaign**: `POST /rest/v1/campaigns/{id}/trigger.json`
    * **Schedule campaign**: `POST /rest/v1/campaigns/{id}/schedule.json`
    * **Get campaign by ID**: `GET /rest/v1/campaigns/{id}.json`

    [View API Documentation](https://experienceleague.adobe.com/en/docs/marketo-developer/marketo/rest/assets/smart-campaigns)
  </Accordion>

  <Accordion title="Email Programs" icon="envelope">
    * **Get email programs**: `GET /rest/asset/v1/programs.json?type=email`
    * **Get email content**: `GET /rest/asset/v1/email/{id}/content.json`
    * **Get email by name**: `GET /rest/asset/v1/email/byName.json?name={name}`
    * **Approve email**: `POST /rest/asset/v1/email/{id}/approveDraft.json`

    [View API Documentation](https://experienceleague.adobe.com/en/docs/marketo-developer/marketo/rest/assets/emails)
  </Accordion>

  <Accordion title="Statistics & Analytics" icon="chart-line">
    * **Get email program stats**: `GET /rest/asset/v1/program/{id}/stats.json`
    * **Get landing page stats**: `GET /rest/asset/v1/landingPage/{id}/stats.json`
    * **Get email performance**: `GET /rest/v1/activities/emailbounced.json`
    * **Get email opens**: `GET /rest/v1/activities/emailopened.json`

    [View API Documentation](https://experienceleague.adobe.com/en/docs/marketo-developer/marketo/rest/endpoint-reference)
  </Accordion>

  <Accordion title="Lists" icon="list">
    * **Get lists**: `GET /rest/v1/lists.json`
    * **Add leads to list**: `POST /rest/v1/lists/{listId}/leads.json`
    * **Remove leads from list**: `DELETE /rest/v1/lists/{listId}/leads.json`
    * **Check list membership**: `GET /rest/v1/lists/{listId}/leads.json`

    [View API Documentation](https://experienceleague.adobe.com/en/docs/marketo-developer/marketo/rest/lead-database/static-lists)
  </Accordion>
</AccordionGroup>

<Tip>
  For a complete list of available endpoints, visit the [Marketo REST API Endpoint Reference](https://experienceleague.adobe.com/en/docs/marketo-developer/marketo/rest/endpoint-reference).
</Tip>

## Example use cases

Here are some ways you can leverage the Marketo integration with your agents:

<AccordionGroup>
  <Accordion title="Campaign Performance Monitoring" icon="chart-bar">
    Create an agent that automatically retrieves campaign metrics, calculates average open rates, tracks click-through rates, and generates performance reports. The agent can monitor email program statistics and alert your team when campaigns exceed or fall below performance thresholds.
  </Accordion>

  <Accordion title="Lead Database Analytics" icon="database">
    Build an agent that retrieves the number of leads in your database, segments them by criteria, tracks lead growth over time, and provides insights into lead quality and distribution across different campaigns and programs.
  </Accordion>

  <Accordion title="Email Metrics Dashboard" icon="envelope-open-text">
    Deploy an agent that accesses campaign email metrics including open rates, click rates, bounce rates, and unsubscribe rates. The agent can compile this data into regular reports or dashboards for marketing team review.
  </Accordion>

  <Accordion title="Lead Scoring Automation" icon="star">
    Create an agent that automatically updates lead scores based on engagement activities, campaign interactions, and behavioral data. The agent can retrieve lead activities, calculate scores using custom logic, and update lead records in Marketo.
  </Accordion>

  <Accordion title="Campaign Trigger Automation" icon="play">
    Build an agent that triggers Marketo campaigns based on external events or conditions. For example, trigger a welcome campaign when a lead is created in your CRM, or launch a re-engagement campaign when leads meet specific criteria.
  </Accordion>

  <Accordion title="Lead Enrichment Workflow" icon="user-plus">
    Deploy an agent that enriches lead data by retrieving information from external sources, updating lead records in Marketo with additional fields, and ensuring data consistency across your marketing stack.
  </Accordion>

  <Accordion title="Marketing ROI Calculator" icon="calculator">
    Create an agent that retrieves campaign costs, performance metrics, and conversion data to calculate marketing ROI. The agent can track revenue attribution, cost per lead, and campaign effectiveness across different channels.
  </Accordion>

  <Accordion title="Automated Reporting Agent" icon="file-chart-column">
    Build an agent that generates weekly or monthly marketing reports by pulling data from multiple Marketo endpoints, combining statistics, and formatting them into comprehensive reports for stakeholders.
  </Accordion>
</AccordionGroup>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="How do I get Marketo API credentials?">
    To get your Marketo API credentials:

    1. Log into Marketo as an administrator
    2. Go to **Admin** > **Integration** > **LaunchPoint**
    3. Create a new service with type "Custom"
    4. Assign an API-only user to the service
    5. View the service details to get your Client ID and Client Secret
    6. Get your REST API Endpoint from **Admin** > **Integration** > **Web Services**

    You'll need all three pieces of information (Client ID, Client Secret, and REST API Endpoint) to authenticate with the Marketo API.
  </Accordion>

  <Accordion title="What are Marketo's API rate limits?">
    Marketo enforces rate limits on API calls:

    * **Standard limit**: 100 calls per 20 seconds
    * **Daily quota**: Varies by subscription (typically 10,000-50,000 calls per day)

    Your agents should be designed to handle rate limiting gracefully. The Marketo API will return a 606 error code when rate limits are exceeded. Consider implementing retry logic with exponential backoff for production workflows.
  </Accordion>

  <Accordion title="How does authentication work with the Marketo API?">
    Marketo uses 2-legged OAuth 2.0 authentication. When you provide your Client ID, Client Secret, and REST API Endpoint to the Marketo API Call tool step, it automatically:

    1. Requests an access token from Marketo's identity service
    2. Includes the Bearer token in the Authorization header of API requests
    3. Handles token refresh when tokens expire (typically after 3600 seconds)

    You don't need to manage tokens manually - the tool step handles this automatically.
  </Accordion>

  <Accordion title="Why am I getting authentication errors?">
    Common authentication issues include:

    * **Invalid credentials**: Double-check your Client ID, Client Secret, and REST API Endpoint
    * **API user permissions**: Ensure the API user assigned to your LaunchPoint service has appropriate permissions
    * **Expired tokens**: The tool step should handle token refresh automatically, but network issues can sometimes cause problems
    * **Incorrect endpoint URL**: Make sure you're using the correct REST API endpoint for your Marketo instance

    If problems persist, try creating a new LaunchPoint service with a fresh set of credentials.
  </Accordion>

  <Accordion title="Can I retrieve historical campaign data?">
    Yes! You can retrieve historical campaign data using various Marketo API endpoints:

    * Use the Activities API to get historical lead activities
    * Retrieve email program statistics for past campaigns
    * Access landing page performance data
    * Query lead changes and updates over time

    Note that data retention policies may vary based on your Marketo subscription level.
  </Accordion>

  <Accordion title="What's the difference between Smart Campaigns and Email Programs?">
    * **Smart Campaigns**: Automated workflows that can include multiple steps, triggers, and actions. Use the Campaigns API endpoints to trigger or schedule these.
    * **Email Programs**: Specifically designed for email marketing with built-in A/B testing and reporting. Use the Email Programs API endpoints to retrieve content and statistics.

    Both can be accessed through the Marketo API Call tool step using their respective endpoints.
  </Accordion>

  <Accordion title="How do I handle pagination in API responses?">
    Many Marketo API endpoints return paginated results. To handle pagination:

    1. Check the response for `nextPageToken` in the result
    2. Use the token in subsequent requests with the `nextPageToken` parameter
    3. Continue until no `nextPageToken` is returned

    Consider using looping in your tools to automatically retrieve all pages of results.
  </Accordion>

  <Accordion title="Can I create custom fields through the API?">
    Yes, you can create and manage custom fields using the Lead Database API. However, custom field creation requires specific API permissions. Ensure your API user has the necessary permissions in Marketo's role settings under **Admin** > **Users & Roles**.
  </Accordion>
</AccordionGroup>

## Additional resources

<CardGroup>
  <Card title="Marketo Developer Portal" icon="code" href="https://developer.adobe.com/marketo-apis/">
    Access the complete Marketo API documentation and developer resources
  </Card>

  <Card title="REST API Documentation" icon="book" href="https://experienceleague.adobe.com/en/docs/marketo-developer/marketo/rest/rest-api">
    Detailed REST API reference and guides
  </Card>

  <Card title="Authentication Guide" icon="key" href="https://experienceleague.adobe.com/en/docs/marketo-developer/marketo/rest/authentication">
    Learn about Marketo's OAuth 2.0 authentication
  </Card>

  <Card title="Endpoint Reference" icon="list" href="https://experienceleague.adobe.com/en/docs/marketo-developer/marketo/rest/endpoint-reference">
    Complete list of available API endpoints
  </Card>
</CardGroup>


# Microsoft Teams
Source: https://relevanceai.com/docs/integrations/popular-integrations/microsoft-teams

Microsoft Teams is a collaboration platform that combines workplace chat, video meetings, file storage, and application integration.

With Relevance AI's Microsoft Teams integration, you can connect your Teams workspace to your AI agents, enabling them to monitor channels, respond to messages, and automate communication workflows directly within Teams.

Relevance AI uses a unified Microsoft authentication that works across **Teams**, **Outlook**, **SharePoint**, and **OneDrive** — connecting one account gives you access to all services. See the [Outlook integration](/docs/integrations/popular-integrations/outlook) and [SharePoint knowledge source](/docs/build/knowledge/integrated-knowledge-sources/sharepoint) pages for details on those services.

## Prerequisites

Before setting up the Microsoft Teams integration, ensure you have:

<CardGroup>
  <Card title="Relevance AI account" icon="user">
    An active project in Relevance AI
  </Card>

  <Card title="Microsoft Teams access" icon="microsoft">
    Desktop or web app within your organization
  </Card>

  <Card title="Sufficient permissions" icon="lock">
    Enterprise organizations may require admin consent from a Global Administrator or Teams Administrator (see [Admin consent & permissions](#admin-consent-permissions))
  </Card>

  <Card title="A Teams channel or chat" icon="hashtag">
    At least one channel or chat where you want your agent to operate
  </Card>
</CardGroup>

## Setup overview

Setting up Microsoft Teams requires two separate steps. Both are required — completing only one is not enough.

<CardGroup>
  <Card title="1. Connect your Microsoft account" icon="plug">
    Authenticate via OAuth in Relevance AI to grant access to Microsoft services.
  </Card>

  <Card title="2. Install the Relevance AI app in Teams" icon="microsoft">
    Add the app from the Teams app store to the channels and chats where your agent should operate.
  </Card>
</CardGroup>

## Step 1 - Connect the integration

<div>
  <iframe title="Connect Microsoft Teams integration" />
</div>

<Steps>
  <Step title="Open the Integrations page">
    In the Relevance AI dashboard, click **Integrations & API Keys** in the sidebar.
  </Step>

  <Step title="Find the Microsoft integration">
    Locate **Microsoft (Teams, Outlook, SharePoint, OneDrive)** in the list of available integrations.
  </Step>

  <Step title="Add the integration">
    Click **Add Integration**. A pop-up window will open for Microsoft sign-in.

    <Tip>If a pop-up window doesn't appear, check your browser's pop-up blocker settings and allow pop-ups from Relevance AI.</Tip>
  </Step>

  <Step title="Sign in to your Microsoft account">
    Enter your Microsoft credentials in the pop-up window and sign in.

    <Warning>The Microsoft account you sign in with determines which Teams, channels, and chats are visible to the integration. Use the account that has access to the channels your agent needs to operate in.</Warning>
  </Step>

  <Step title="Grant permissions">
    Review the requested permissions and click **Accept**. These permissions allow Relevance AI to read and send messages, access channel and team information, and maintain a persistent connection.

    <Warning>If you see a "Need admin approval" message, your organization's Azure AD / Entra ID tenant restricts third-party app consent. See [Admin consent & permissions](#admin-consent-permissions) for next steps.</Warning>

    For a full list of requested scopes, see the [Permissions reference](#permissions-reference).
  </Step>

  <Step title="Confirm the connection">
    Once complete, you'll see a green **Connected** status indicator next to the Microsoft integration in Relevance AI.
  </Step>
</Steps>

### Linking your account via magic link

<Note>Both the magic link flow and the standard OAuth flow achieve the same outcome — your Microsoft Teams account is linked to your Relevance AI account.</Note>

You may receive a magic link from Microsoft Teams or via email inviting you to connect your accounts. This is an alternative to the standard OAuth flow above and achieves the same result.

1. Click the magic link in Microsoft Teams or your email.
2. If you're not logged in to Relevance AI, you'll be redirected to sign in first. After signing in, you'll be returned to the consent page automatically.
3. On the consent page, review your Relevance AI account email and your Microsoft Teams account details shown.
4. Click **Confirm** to link the accounts, or **Cancel** to abort the connection.

## Step 2 - Install the Relevance AI app in Microsoft Teams

After connecting your Microsoft account, you need to install the Relevance AI app in Teams so your agents can interact with channels and chats.

<Warning>The Relevance AI app must be added to every channel and group chat where you want your agent to respond. Triggers and tool steps will not work in channels or chats where the app has not been added.</Warning>

<Steps>
  <Step title="Open the Teams app store">
    In Microsoft Teams, click on **Apps** in the left sidebar.
  </Step>

  <Step title="Search for the Relevance AI app">
    Type **Relevance AI** in the search bar and locate the app in the results.

    <Note>If you don't see the Relevance AI app, your organization may restrict third-party app installations. See [Admin consent & permissions](#admin-consent-permissions) for how to request approval.</Note>
  </Step>

  <Step title="Install the app">
    Click **Add** or **Install** to add the Relevance AI app to your Teams workspace.
  </Step>

  <Step title="Add the app to specific channels and chats">
    After installing, add the app to each channel or group chat where you want your agent to operate.

    <Tabs>
      <Tab title="Channels">
        1. Navigate to the channel where you want your agent to work
        2. Click the **+** icon at the top of the channel to add a tab or app
        3. Search for **Relevance AI** in the app picker
        4. Select the app and click **Save**
      </Tab>

      <Tab title="Group chats">
        1. Open the group chat where you want your agent to work
        2. Click the **+** icon or **Add apps** button
        3. Search for and select **Relevance AI**
        4. Add the app to the chat
      </Tab>
    </Tabs>
  </Step>

  <Step title="Verify the app is active">
    Confirm the Relevance AI app appears in the channel's app list and that the channel shows up in the trigger setup dropdown in Relevance AI.
  </Step>
</Steps>

## Triggers

<Frame>
  <img alt="Trigger conversation in Teams" />
</Frame>

Triggers let your agents respond to Teams messages automatically, so your team can get answers, run workflows, and escalate issues without leaving their existing conversations. Instead of switching to a separate tool, users interact with agents right where they already communicate.

The Relevance AI app is available on the Microsoft Teams marketplace. Users can @mention or message the app in channels, group chats, or direct messages to trigger agents and have two-way conversations directly in Teams. The same trigger setup also works for [workforces](/docs/build/workforces/create-a-workforce).

<Warning>Teams triggers activate only on new messages. They do not trigger on new chat creation, group creation, or webhooks.</Warning>

### Setting up a trigger

<Note>Before setting up a trigger, confirm that you have completed both <a href="#step-1-connect-the-integration">Step 1: Connect the integration</a> and <a href="#step-2-install-the-relevance-ai-app-in-microsoft-teams">Step 2: Install the Relevance AI app</a>.</Note>

<div>
  <iframe title="Set up Microsoft Teams trigger" />
</div>

<Steps>
  <Step title="Open your agent's settings">
    Navigate to your agent in Relevance AI and go to the **Triggers** section.
  </Step>

  <Step title="Add a Microsoft Teams trigger">
    Click **Add Trigger** and select **Microsoft Teams** from the list.
  </Step>

  <Step title="Select the Team and channel">
    Choose the Microsoft account you connected, then select the **Team** and specific **channel** or **chat** you want to monitor.

    <Note>Only channels where the authenticated user is a member and the Relevance AI app has been added will appear in the dropdown.</Note>
  </Step>

  <Step title="Configure trigger conditions (optional)">
    Set up **keyword matching** to filter which messages activate your agent. Leave the keyword field empty to trigger on all messages, or enter specific keywords separated by commas. The trigger checks whether any of the keywords appear anywhere in the message text (case-insensitive).
  </Step>

  <Step title="Write agent instructions">
    In the **Core Instructions** section, write a prompt that guides how your agent should respond — including its role, tone, and when to respond.
  </Step>
</Steps>

<Tip>Set tool permissions to "approval mode" initially so your agent asks before sending messages. Switch to autopilot once you're confident.</Tip>

### Advanced trigger settings

#### Thread reply method

<Frame>
  <img alt="Thread Reply Method setting in Microsoft Teams trigger advanced settings" />
</Frame>

The Thread Reply Method controls whether your agent posts a response back to the Teams thread after processing a message. The setting is per-trigger and opt-in, so the default keeps existing triggers behaving exactly as before.

<CardGroup>
  <Card title="Agent auto-reply" icon="reply">
    Default. The agent processes the message and posts its response back into the Teams thread.
  </Card>

  <Card title="No reply" icon="comment-slash">
    The agent processes the message silently. Use this when the output goes somewhere other than Teams — a database write, an email, or a downstream workflow.
  </Card>
</CardGroup>

<Steps>
  <Step title="Open your trigger">
    Go to your agent's **Microsoft Teams** trigger in Relevance AI.
  </Step>

  <Step title="Expand Advanced Settings">
    Click **Advanced Settings** to reveal the additional trigger options.
  </Step>

  <Step title="Choose a reply method">
    Select **Agent auto-reply** or **No reply** under **Thread Reply Method**.
  </Step>
</Steps>

<Warning>
  When "No reply" is selected, you won't see confirmation in Teams that your agent processed the message. Monitor the agent's task history in Relevance AI to verify it's working correctly.
</Warning>

#### Outreach replies only

Enable this toggle to restrict the agent to responding only to replies to messages it previously sent in the channel, rather than all new messages.

This is useful for outreach workflows where the agent sends messages to users and needs to follow up only with those who respond. Rather than monitoring all channel activity, the agent stays focused on conversations it initiated.

<Steps>
  <Step title="Open your trigger settings">
    Go to your agent's **Microsoft Teams** trigger in Relevance AI.
  </Step>

  <Step title="Enable the toggle">
    Locate the **Outreach replies only** toggle, turn it on, and save your configuration.
  </Step>
</Steps>

When enabled, the trigger subtitle changes to **"On replies to outreach messages"**. The agent triggers only when a user replies to a message the agent itself sent — new messages posted by other users will not activate it.

### How triggered conversations work

Once a trigger fires, the agent responds directly inside Teams with the following capabilities:

<CardGroup>
  <Card title="Threaded conversations" icon="comments">
    Agent replies are threaded in the channel. Conversation threads stay active for 30 minutes.
  </Card>

  <Card title="Adaptive Cards" icon="table-layout">
    Responses are delivered as Microsoft Adaptive Cards — rich, interactive messages inside Teams.
  </Card>

  <Card title="Typing indicator" icon="ellipsis">
    Teams displays a typing indicator while the agent is processing a response.
  </Card>

  <Card title="File and image attachments" icon="paperclip">
    Agents can receive and process inline images, pasted screenshots, and uploaded files (PDFs, documents, images) from Teams conversations.
  </Card>

  <Card title="Proactive messages" icon="paper-plane">
    Agents can proactively initiate conversations in Teams DMs without waiting for user input first. Context is maintained across the full conversation, enabling back-and-forth exchanges after the initial outreach.
  </Card>

  <Card title="Workforce triggers" icon="users-gear">
    Teams messages can trigger an entire workforce, not just an individual agent.
  </Card>
</CardGroup>

<Note>Messages that contain only attachments with no text will show a "\[N attachments]" indicator to the agent.</Note>

## Tool steps for Microsoft Teams

<div>
  <iframe title="Microsoft Teams tool steps" />
</div>

The Microsoft Teams integration provides actions your agents can use as tool steps in their workflows.

<CardGroup>
  <Card title="Send Channel Message" icon="message">
    Post a message to a specific Teams channel
  </Card>

  <Card title="Send Chat Message" icon="comment">
    Send a direct message in Teams
  </Card>

  <Card title="Create Channel" icon="plus">
    Create a new channel in a Team
  </Card>

  <Card title="List Teams" icon="users">
    Get all teams the authenticated user belongs to
  </Card>

  <Card title="List Channels" icon="list">
    Get all channels in a Team
  </Card>

  <Card title="Retrieve Chat Messages" icon="inbox">
    Fetch messages from a specific chat conversation
  </Card>

  <Card title="Search Messages" icon="magnifying-glass">
    Search for messages across Teams channels and chats
  </Card>

  <Card title="List Shifts" icon="calendar-days">
    Retrieve shift information from Teams
  </Card>

  <Card title="Microsoft API Call" icon="code">
    Make custom calls to Microsoft Graph API
  </Card>
</CardGroup>

<Note>
  Most Microsoft Teams tool steps are currently in beta. Please report any issues to our support team.
</Note>

## Agent alerts

<Frame>
  <img alt="Agent notification configuration" />
</Frame>

You can receive notifications in Microsoft Teams when your agents enter specific statuses or encounter tool errors. This helps you monitor agent activity and respond quickly when issues occur.

<Note>Agent notifications are currently in beta and may not be available for all accounts.</Note>

### Setting up Teams notifications

<Steps>
  <Step title="Open the escalations panel">
    Navigate to your agent and click the **Build** tab, then click **Escalations** in the left sidebar.
  </Step>

  <Step title="Add an agent notification">
    Under "Agent Notifications", click **Add agent notification**.
  </Step>

  <Step title="Configure notification settings">
    Select **Microsoft Teams** as the platform, then choose the notification trigger: **Agent enters status** (select specific task statuses such as "Running", "Completed", or "Failed") or **Tool errors (any tool)** to get notified when any tool encounters an error. Select a Microsoft Teams account and the channel where you want to receive notifications. If no accounts appear, complete the [connection steps](#step-1-connect-the-integration) first.
  </Step>

  <Step title="Publish changes">
    Click **Publish changes** to save your notification configuration.
  </Step>
</Steps>

<Tip>Test the notification by running your agent — you should receive a notification in your selected Teams channel based on your configured triggers.</Tip>

### What you'll receive

When a notification is triggered, you'll receive a message in your Teams channel containing:

<CardGroup>
  <Card title="Agent status and context" icon="circle-info">
    Agent name, status, and task details
  </Card>

  <Card title="Direct link" icon="arrow-up-right-from-square">
    Link to view the full task in Relevance AI
  </Card>

  <Card title="Error details" icon="triangle-exclamation">
    For tool errors: which tool failed and why
  </Card>
</CardGroup>

## Admin consent & permissions

Enterprise organizations often require administrator approval before users can connect third-party applications or install apps in Microsoft Teams.

### Understanding admin consent

Microsoft Azure AD (Entra ID) allows tenant administrators to restrict whether users can grant consent to third-party apps on their own. When this restriction is active, individual users will see a "Need admin approval" screen during the OAuth flow instead of the standard consent prompt. This is a tenant-level policy — it is not specific to Relevance AI.

If you see "Need admin approval" or "This app requires admin approval", follow the steps below.

<Tabs>
  <Tab title="OAuth connection approval">
    If you see an admin approval message when connecting your Microsoft account:

    <Steps>
      <Step title="Copy the consent URL">
        Copy the **consent URL** that appears in the Microsoft login window. This URL contains the specific permissions Relevance AI is requesting.
      </Step>

      <Step title="Forward the URL to your administrator">
        Send the URL to your **Microsoft 365 Global Administrator or Application Administrator** with a request to grant consent. Include a brief explanation of what Relevance AI is and how it will be used.
      </Step>

      <Step title="Wait for admin approval">
        Your administrator will open the URL, review the requested permissions, and click **Accept** to grant consent on behalf of the organization.
      </Step>

      <Step title="Retry the connection">
        Return to Relevance AI and **retry the Microsoft integration connection**. The consent prompt should no longer appear.
      </Step>
    </Steps>
  </Tab>

  <Tab title="Teams app installation approval">
    If your organization requires admin approval for app installations:

    <Steps>
      <Step title="Contact your IT administrator">
        Request approval to install the Relevance AI app in Microsoft Teams.
      </Step>

      <Step title="Wait for admin approval">
        Your administrator will go to the [Microsoft Teams Admin Center](https://admin.teams.microsoft.com), navigate to **Teams apps** > **Manage apps**, search for **Relevance AI**, and change the app status to **Allowed**.
      </Step>

      <Step title="Install the app">
        Once approved, follow the steps in [Step 2: Install the Relevance AI app](#step-2-install-the-relevance-ai-app-in-microsoft-teams).
      </Step>
    </Steps>
  </Tab>
</Tabs>

## Permissions reference

The following OAuth scopes are requested when you connect your Microsoft account. These permissions allow Relevance AI to read and send messages, access team and channel metadata, and maintain persistent access.

| Scope                     | Description                                                    |
| ------------------------- | -------------------------------------------------------------- |
| `offline_access`          | Maintains access without requiring frequent re-authentication  |
| `Chat.Create`             | Create new chats                                               |
| `Chat.ReadWrite`          | Read and write to chats the user has access to                 |
| `ChatMessage.Read`        | Read chat messages                                             |
| `ChatMessage.Send`        | Send chat messages                                             |
| `ChannelMessage.Read.All` | Read messages in channels the user has access to               |
| `ChannelMessage.Send`     | Send messages to channels                                      |
| `Channel.ReadBasic.All`   | Read basic channel information (name, description, ID)         |
| `Team.ReadBasic.All`      | Read basic team information (name, description, ID)            |
| `User.Read`               | Read the signed-in user's profile                              |
| `User.Read.All`           | List users in the organization (used for trigger user filters) |

<Note>These permissions may evolve as new features are added to the integration. Any changes will be reflected in the OAuth consent prompt.</Note>

## Troubleshooting

<AccordionGroup>
  <Accordion title="'The application is not authorized' error">
    This error means your Azure AD / Entra ID tenant has not granted consent for Relevance AI. Ask your Microsoft 365 Global Administrator to grant admin consent using the steps in [Admin consent & permissions](#admin-consent-permissions). Once consent is granted, retry the connection from the Integrations page.
  </Accordion>

  <Accordion title="'Consent was denied' error">
    This occurs when you or your administrator explicitly declined the permissions request during the OAuth flow. To resolve this, retry the integration connection from the Integrations page and click **Accept** when prompted. If the consent was denied by an admin policy, your administrator will need to grant consent on your behalf.
  </Accordion>

  <Accordion title="Relevance AI app not visible in the Teams app store">
    If the Relevance AI app does not appear in the Teams app store, your organization likely restricts third-party app installations. Ask your IT administrator to allow the app in the [Microsoft Teams Admin Center](https://admin.teams.microsoft.com) under **Teams apps** > **Manage apps**. See [Admin consent & permissions](#admin-consent-permissions) for detailed steps.
  </Accordion>

  <Accordion title="Teams trigger option not visible in agent settings">
    Some Teams features are controlled by feature flags and may not be enabled for all accounts. If you don't see Microsoft Teams as a trigger option in your agent or workforce settings, [contact our support team](/docs/get-started/support) to check whether the feature needs to be enabled for your account.
  </Accordion>

  <Accordion title="Trigger is configured but the agent isn't responding">
    Verify that the Relevance AI app has been added to the specific channel or chat being monitored (not just installed in Teams), and that the Microsoft account used for the trigger is a member of the channel. Confirm the trigger is enabled and published in your agent's settings. If keyword matching is configured, check that messages contain the expected keywords. Review the agent's task logs in Relevance AI for errors — a `401 Unauthorized` error indicates the OAuth token has expired or been revoked, and you should reconnect the Microsoft integration.
  </Accordion>

  <Accordion title="Magic link expired or not working">
    Magic links are time-limited and may expire before you click them. If your magic link no longer works, connect your Microsoft Teams account directly through the standard OAuth flow: go to **Integrations & API Keys** in the sidebar, find the Microsoft integration, and click **Add Integration**.
  </Accordion>

  <Accordion title="Consent page shows wrong email address">
    The consent page shows two email addresses: your Relevance AI account email and your Microsoft Teams account email. If either address looks incorrect, click **Cancel** and verify that you're logged in to the correct Relevance AI account before trying again. To switch accounts, sign out of Relevance AI and sign in with the correct account, then click the magic link again.
  </Accordion>
</AccordionGroup>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Who supports the Microsoft Teams integration?">
    The Microsoft Teams integration is built and maintained by Relevance AI. For integration-specific questions, [contact our support team](/docs/get-started/support). For Microsoft Teams platform issues, contact [Microsoft support](https://support.microsoft.com/).
  </Accordion>

  <Accordion title="Why don't I see my Teams channel in the dropdown when setting up a trigger?">
    The most common cause is that the Relevance AI app has not been installed in Teams or added to the specific channel. The Microsoft account you connected must also be a member of that channel. If your organization requires admin consent, the app may not be visible until an administrator approves it. Verify that you've completed both setup steps and added the app to the channel.
  </Accordion>

  <Accordion title="Can I connect multiple Microsoft accounts?">
    Yes, you can connect multiple Microsoft accounts through the Integrations & API Keys page. Each account can be used for different triggers and tool steps.
  </Accordion>

  <Accordion title="Do triggers work for private messages?">
    Yes, Teams triggers can monitor both channel messages and private chats. To monitor direct messages, select the "Direct chats with Relevance AI" option when configuring your trigger instead of a specific channel.
  </Accordion>

  <Accordion title="Can my agent process Teams messages without replying?">
    Yes. Set the Thread Reply Method to "No reply" in your MS Teams trigger's Advanced Settings. This allows your agent to process messages without posting a response back to Teams — useful when the agent's output goes somewhere else, such as a database, email, or another workflow. The default is "Agent auto-reply", which maintains existing behavior. To verify the agent is running correctly, check the task history in Relevance AI.
  </Accordion>

  <Accordion title="Can my agent access files shared in Teams?">
    Yes, files shared in Teams channels are stored in SharePoint. You can access them through the Microsoft API Call tool step via the Microsoft Graph API.
  </Accordion>

  <Accordion title="What's the difference between the pre-built tool steps and the API Call tool step?">
    Pre-built tool steps (like "Send Channel Message") are designed for common tasks with simplified interfaces. The Microsoft API Call tool step gives you full access to Microsoft Graph API for advanced operations not covered by pre-built steps.
  </Accordion>

  <Accordion title="Are there rate limits for Microsoft Teams API calls?">
    Yes, Microsoft Graph API enforces rate limits depending on your Microsoft 365 subscription. Microsoft returns a `429 Too Many Requests` status code when limits are exceeded.
  </Accordion>

  <Accordion title="How do I remove the Microsoft Teams integration?">
    Go to the **Integrations & API Keys** page from the sidebar, find **Microsoft (Teams, Outlook, SharePoint, OneDrive)**, click "..." on the account you want to remove, and click "Remove" and confirm. This will disable all triggers and tool steps using this account across Teams, Outlook, SharePoint, and OneDrive. To also remove the app from Microsoft Teams, go to **Apps** in Teams, find Relevance AI, and click **Uninstall**.
  </Accordion>

  <Accordion title="What is the magic link I received from Microsoft Teams?">
    A magic link is an alternative way to connect your Microsoft Teams account to Relevance AI. When you click the link, you're taken to a consent page in Relevance AI that shows your Relevance AI account email and your Microsoft Teams account details. Click **Confirm** to complete the connection or **Cancel** to abort it. The magic link method achieves the same result as connecting via the standard OAuth flow on the Integrations page.
  </Accordion>

  <Accordion title="Do I need to use the magic link or can I connect directly from Relevance AI?">
    Both methods work and achieve the same result. You can connect your Microsoft Teams account through the standard OAuth flow on the **Integrations & API Keys** page, or by clicking a magic link you received from Microsoft Teams or via email. Use whichever is more convenient.
  </Accordion>

  <Accordion title="What happens if I'm not logged in when I click the magic link?">
    If you're not logged in to Relevance AI when you click the magic link, you'll be redirected to the sign-in page. After signing in, you'll be automatically returned to the consent page to complete the account linking.
  </Accordion>
</AccordionGroup>


# Notion
Source: https://relevanceai.com/docs/integrations/popular-integrations/notion

Seamlessly connect your AI agents with Notion workspaces to automate document creation, information retrieval, and workspace management.

## Connecting the Integration

To connect the Notion integration with Relevance AI:

1. Navigate to the Integrations section in your Relevance AI dashboard
2. Select "Notion" from the available integrations
3. Click "Connect"
4. You'll be redirected to Notion to authorize the connection
5. Select the Notion workspace you want to connect
6. Review the permissions requested and click "Allow access"
7. You'll be redirected back to Relevance AI with the integration now connected

Once connected, your Notion workspace will be available for your agents to access through various tools and actions.

## Notion API Authentication

To use Notion tools with your agent, you'll need to:

1. Create a Notion integration at [https://www.notion.so/my-integrations](https://www.notion.so/my-integrations)
2. Generate an API key for your integration
3. Share specific pages or databases with your integration
4. Provide the API key when configuring Notion tools in Relevance AI

### Creating a Notion Integration

1. Go to [https://www.notion.so/my-integrations](https://www.notion.so/my-integrations)

2. Click "New integration"

3. Enter a name for your integration (e.g., "Relevance AI Assistant")

4. Select the workspace where you want to use the integration

5. Upload an icon (optional)

6. Configure the capabilities your integration needs:
   * Read content: Required for retrieving information from Notion
   * Update content: Required for modifying existing pages
   * Insert content: Required for creating new pages or database entries

7. Click "Submit" to create your integration

### Generating and Managing API Keys

After creating your integration:

1. Navigate to the "Secrets" tab in your integration settings
2. You'll see your "Internal Integration Token" (API key)
3. Copy this token to use with Relevance AI tools
4. For security, you can regenerate this token at any time if needed

### Sharing Pages with Your Integration

For your integration to access specific Notion content:

1. Open the Notion page or database you want to share
2. Click the "..." menu in the top-right corner
3. Select "Add connections"
4. Find and select your integration from the list
5. Click "Confirm" to grant access

Remember that you must explicitly share each page or database you want your integration to access. Parent pages do not automatically grant access to child pages.

## Setting Up Triggers

The Notion integration can be configured to trigger your AI agents when specific events occur in your Notion workspace. This allows your agents to respond automatically to changes in your documentation.

To set up a Notion trigger *(coming soon)*:

1. Navigate to your "agent's profile" page
2. Under "Triggers", select "Notion"
3. Choose the trigger type:
   * New Page Created
   * Page Updated
   * Database Entry Added
   * Comment Added
   * Mention in Page
4. Configure the specific conditions for the trigger (e.g., which database to monitor)
5. Save your trigger configuration

With triggers configured, your agent will automatically activate when the specified events occur in your Notion workspace, allowing for real-time responses to documentation changes.

## Tools & Tool Steps

The Notion integration provides a rich set of tools and actions that your agents can use to interact with your Notion workspace. These actions can be incorporated into your agent's workflows to automate various documentation tasks.

### Available Notion Actions

Your agents can leverage these powerful Notion actions:

* **Create Notion Page**: Generate new pages with formatted content, including headings, lists, tables, and more
* **Update Notion Page**: Modify existing page content while preserving structure
* **Get Notion Page Content**: Retrieve and analyze the content of specific pages
* **Create Database Entry**: Add new items to Notion databases with structured properties
* **Query Database**: Search and filter database entries based on specific criteria
* **Get Database Structure**: Retrieve the schema of a database to understand its properties
* **Add Comments**: Post comments on pages for collaborative feedback
* **Create To-Do Items**: Generate actionable tasks within Notion pages
* **Get User Details**: Retrieve information about Notion workspace members
* **Search Notion**: Find relevant content across your entire workspace

And many more! These actions can be combined in powerful ways to create sophisticated documentation workflows.

### Example Use Cases

Here are some examples of how your agents can use Notion actions:

* **Meeting Documentation**: Automatically create meeting notes in Notion after calendar events
* **Knowledge Base Management**: Update documentation when new information becomes available
* **Project Tracking**: Create and update project status pages based on progress
* **Content Creation**: Generate draft documents for review based on specific requirements
* **Research Compilation**: Gather information from various sources and organize it in Notion pages

## Using the Notion API Tool Step (Advanced)

For advanced users who need more customized functionality, the Notion API tool step provides direct access to the Notion API. This allows you to build custom tools that perform specialized Notion operations beyond the standard actions.

To use the Notion API tool step:

1. Create a new tool in Relevance AI
2. Scroll down to Tool-steps
3. Add "Notion API" tool-step
4. Select your connected Notion account in the dropdown
5. Configure the API endpoint, method, and parameters
6. Test your custom API call
7. Save your tool configuration

The Notion API tool step gives you the flexibility to create highly customized Notion interactions tailored to your specific needs.

## Best Practices

To get the most out of the Notion integration:

* **Structure Your Workspace**: Organize your Notion workspace with clear hierarchies to make it easier for agents to navigate
* **Use Templates**: Create templates for common document types to ensure consistency
* **Set Clear Permissions**: Configure appropriate access levels for your agents in Notion
* **Monitor Activity**: Regularly review the actions your agents are taking in your Notion workspace
* **Start Simple**: Begin with basic documentation tasks before implementing more complex workflows
* **Secure Your API Keys**: Store your Notion API keys securely and rotate them periodically
* **Test Thoroughly**: Always test your integration in a controlled environment before deploying to production

## Troubleshooting

**Connection Issues**

* Ensure your Notion authorization hasn't expired
* Verify you've granted all necessary permissions
* Check that your Notion workspace is accessible
* Confirm your API key is valid and has not been revoked

**Content Creation Problems**

* Review the formatting in your agent's instructions
* Ensure the agent has access to the parent pages where content is being created
* Check for any API rate limits that might be affecting performance

**Trigger Reliability**

* Verify that your trigger conditions are correctly configured
* Ensure the monitored pages or databases are accessible to the integration
* Check that webhook notifications are enabled in your Notion workspace settings

**API Authentication Errors**

* Verify that your API key is correctly entered in the tool configuration
* Ensure the pages you're trying to access have been shared with your integration
* Check that your integration has the necessary capabilities enabled

For additional assistance, contact Relevance AI support or consult the detailed API documentation.

## Related Features

[Knowledge Integration](https://relevanceai.com/docs/build/knowledge/create-knowledge) - Connect your Notion workspace as a knowledge source for your agents, allowing them to reference your documentation when responding to queries.

[Document Processing Tools](https://relevanceai.com/docs/build/tools/templates) - Combine Notion integration with document processing capabilities to extract, analyze, and organize information from various sources.

[Workflow Automation](https://relevanceai.com/docs/build/agents/customise-agent/flowbuilder) - Create sophisticated workflows that incorporate Notion actions alongside other integrations for end-to-end process automation.

## Notion Assistant

The Notion Assistant is a pre-built agent available from the [Marketplace](https://app.relevanceai.com/marketplace) that builds databases, creates pages, and organizes your workspace through plain English. It validates requests before making changes and handles complex multi-step workflows automatically.

<Card title="Clone Notion Assistant" icon="note" href="https://app.relevanceai.com/marketplace">
  Search "Notion Assistant" in the Marketplace to clone it and start managing your workspace through conversation
</Card>


# Outlook
Source: https://relevanceai.com/docs/integrations/popular-integrations/outlook

Microsoft Outlook is a comprehensive email and calendar management platform used by millions of professionals worldwide.

Microsoft Outlook is a comprehensive email and calendar management platform used by millions of professionals worldwide. With Relevance AI's Outlook integration, you can seamlessly connect your Outlook account to your AI agents, enabling them to manage emails, calendar events, contacts, and more, making your communication workflows more efficient and intelligent.

<Warning>
  The Outlook integration in Relevance AI was built by Relevance AI, and is therefore supported by our team, not Microsoft. If you have a question or issue with using Outlook in Relevance AI, please [reach out to our support team](/docs/get-started/support). If you have a question or issue that is only about Outlook, you can [reach out to Microsoft support](https://support.microsoft.com/).
</Warning>

## Connect the integration

Connecting your Outlook account to Relevance AI is a straightforward process:

1. Go to the "Integrations & API Keys" page in the sidebar of your Relevance AI dashboard.
2. Click on "Outlook" from the available integrations.
3. Click on the "Add Integration" button.
4. In the pop-up window, sign into your Microsoft account.
5. Grant the necessary permissions for Relevance AI to access your Outlook data.
6. Once authenticated, your Outlook account will appear as a connected integration.

<Note>
  After connecting, you can use Outlook as a trigger for your agents to automatically respond to incoming emails, or use Outlook tool steps in your custom tools to manage emails, calendar events, and contacts.
</Note>

## Tool steps for Outlook

The Outlook integration provides a comprehensive set of actions that your agents can use to interact with your emails, calendar, and contacts. These actions can be incorporated into your agent's workflows as tool steps, enabling sophisticated communication automation capabilities.

### Email Management

<CardGroup>
  <Card title="Find Email" icon="magnifying-glass">
    Search for specific emails in your mailbox
  </Card>

  <Card title="List Folders" icon="folder">
    Retrieve all email folders in your account
  </Card>

  <Card title="List Labels" icon="tags">
    Get all email labels/categories
  </Card>

  <Card title="Add Label" icon="tag">
    Add a label or category to an email
  </Card>

  <Card title="Remove Label" icon="xmark">
    Remove a label or category from an email
  </Card>

  <Card title="Move Email to Folder" icon="folder-arrow-up">
    Move an email to a different folder
  </Card>

  <Card title="Send Email" icon="paper-plane">
    Send an email from your Outlook account
  </Card>
</CardGroup>

### Calendar Management

<CardGroup>
  <Card title="Check Availability" icon="calendar-check">
    Check if a time slot is available
  </Card>

  <Card title="Create Event" icon="calendar-plus">
    Create a new calendar event
  </Card>

  <Card title="Delete Event" icon="calendar-xmark">
    Delete an existing calendar event
  </Card>

  <Card title="List Events" icon="calendar-days">
    List calendar events within a date range
  </Card>

  <Card title="Update Event" icon="calendar-pen">
    Update an existing calendar event
  </Card>

  <Card title="Get Free/Busy Schedule" icon="clock">
    Retrieve free/busy schedule information
  </Card>
</CardGroup>

### Contact Management

<CardGroup>
  <Card title="Create Contact" icon="user-plus">
    Create a new contact in your address book
  </Card>

  <Card title="Find Contact" icon="address-book">
    Search for a specific contact
  </Card>

  <Card title="List Contacts" icon="users">
    List all contacts in your address book
  </Card>

  <Card title="Update Contact" icon="user-pen">
    Update an existing contact's information
  </Card>
</CardGroup>

### Advanced Operations

<CardGroup>
  <Card title="Microsoft Outlook API Call" icon="code">
    Make custom API calls to Microsoft Graph API
  </Card>
</CardGroup>

<Tip>
  Type "Outlook" in the tool step search bar to see all available Outlook actions when building your tools.
</Tip>

## Use the integration's API tool step (Advanced)

In addition to the pre-built actions available in the tool directory, you can build custom tools that perform Outlook-specific activities using the Microsoft Outlook API Call tool step. This gives you access to the full Microsoft Graph API for advanced email, calendar, and contact operations.

### How to use the Microsoft Outlook API Call tool step

<Steps>
  <Step title="Create a new tool">
    Create a new tool in Relevance AI or open an existing tool you want to add Outlook functionality to.
  </Step>

  <Step title="Add the Microsoft Outlook API Call tool step">
    1. Scroll down to Tool-steps
    2. Search for "Microsoft Outlook API Call" in the tool step search bar
    3. Add the Microsoft Outlook API Call tool step to your workflow
  </Step>

  <Step title="Select your Outlook account">
    Select your connected Outlook account from the dropdown menu.
  </Step>

  <Step title="Configure the API endpoint">
    Configure the API endpoint, method, and parameters according to your needs:

    * **Method**: Select the HTTP method (GET, POST, PUT, DELETE, PATCH)
    * **Endpoint**: Enter the API endpoint path (e.g., `/me/messages`, `/me/calendar/events`)
    * **Body**: Add any required request body data
    * **Headers**: Add any custom headers if needed
  </Step>

  <Step title="Test your API call">
    Test your configuration to ensure it works correctly before deploying.
  </Step>
</Steps>

### Example: Creating a Calendar Event with Custom Properties

Here's a practical example of using the Microsoft Outlook API Call tool step to create a calendar event with custom properties:

**API Endpoint**: `POST /me/calendar/events`

**Configuration**:

```json theme={null}
{
  "method": "POST",
  "endpoint": "/me/calendar/events",
  "body": {
    "subject": "Team Sync Meeting",
    "body": {
      "contentType": "HTML",
      "content": "Discuss Q4 objectives and team priorities"
    },
    "start": {
      "dateTime": "2024-12-01T14:00:00",
      "timeZone": "Pacific Standard Time"
    },
    "end": {
      "dateTime": "2024-12-01T15:00:00",
      "timeZone": "Pacific Standard Time"
    },
    "location": {
      "displayName": "Conference Room A"
    },
    "attendees": [
      {
        "emailAddress": {
          "address": "colleague@example.com",
          "name": "Colleague Name"
        },
        "type": "required"
      }
    ]
  }
}
```

This configuration:

* Uses the POST method to create a new event
* Specifies the event details including subject, body, and location
* Sets the start and end times with timezone information
* Adds attendees to the meeting

<Info>
  You can find Microsoft Graph API's complete documentation at [https://learn.microsoft.com/en-us/graph/api/overview](https://learn.microsoft.com/en-us/graph/api/overview).
</Info>

### Common Microsoft Graph API Endpoints for Outlook

Here are some commonly used Microsoft Graph API endpoints you can use with the API Call tool step:

<AccordionGroup>
  <Accordion title="Email Operations" icon="envelope">
    * **List messages**: `GET /me/messages`
    * **Get message**: `GET /me/messages/{id}`
    * **Send email**: `POST /me/sendMail`
    * **Reply to email**: `POST /me/messages/{id}/reply`
    * **Forward email**: `POST /me/messages/{id}/forward`
    * **Delete email**: `DELETE /me/messages/{id}`
    * **Mark as read**: `PATCH /me/messages/{id}` with `{"isRead": true}`

    [View API Documentation](https://learn.microsoft.com/en-us/graph/api/resources/message)
  </Accordion>

  <Accordion title="Calendar Operations" icon="calendar">
    * **List events**: `GET /me/calendar/events`
    * **Get event**: `GET /me/events/{id}`
    * **Create event**: `POST /me/calendar/events`
    * **Update event**: `PATCH /me/events/{id}`
    * **Delete event**: `DELETE /me/events/{id}`
    * **Get schedule**: `POST /me/calendar/getSchedule`

    [View API Documentation](https://learn.microsoft.com/en-us/graph/api/resources/event)
  </Accordion>

  <Accordion title="Contact Operations" icon="address-card">
    * **List contacts**: `GET /me/contacts`
    * **Get contact**: `GET /me/contacts/{id}`
    * **Create contact**: `POST /me/contacts`
    * **Update contact**: `PATCH /me/contacts/{id}`
    * **Delete contact**: `DELETE /me/contacts/{id}`

    [View API Documentation](https://learn.microsoft.com/en-us/graph/api/resources/contact)
  </Accordion>

  <Accordion title="Folder Operations" icon="folder-tree">
    * **List folders**: `GET /me/mailFolders`
    * **Get folder**: `GET /me/mailFolders/{id}`
    * **Create folder**: `POST /me/mailFolders`
    * **Move message**: `POST /me/messages/{id}/move`
    * **Copy message**: `POST /me/messages/{id}/copy`

    [View API Documentation](https://learn.microsoft.com/en-us/graph/api/resources/mailfolder)
  </Accordion>

  <Accordion title="Search & Filter" icon="filter">
    * **Search messages**: `GET /me/messages?$search="subject:meeting"`
    * **Filter messages**: `GET /me/messages?$filter=from/emailAddress/address eq 'user@example.com'`
    * **Order results**: `GET /me/messages?$orderby=receivedDateTime desc`
    * **Select fields**: `GET /me/messages?$select=subject,from,receivedDateTime`

    [View API Documentation](https://learn.microsoft.com/en-us/graph/query-parameters)
  </Accordion>
</AccordionGroup>

<Tip>
  Use the `$filter`, `$search`, `$orderby`, and `$select` query parameters to refine your API calls and retrieve exactly the data you need.
</Tip>

## Set up Outlook as a trigger

You can configure your agents to automatically respond to incoming Outlook emails by setting up Outlook as a trigger.

<Warning>
  The Outlook trigger has a 25MB per-attachment limit due to Microsoft Graph API constraints, so emails with attachments larger than 25MB will not be processed by the trigger. To work around this, compress large files before sending, convert large DOCX files to PDF format to reduce file size, or use cloud storage links (OneDrive, SharePoint) instead of large attachments.
</Warning>

### How to set up the Outlook trigger

<Steps>
  <Step title="Open agent settings">
    Navigate to your agent's settings page.
  </Step>

  <Step title="Add Outlook trigger">
    1. Click on "Integrations & API Keys" in the sidebar
    2. Under "Triggers", click on the Outlook button
    3. Select your connected Outlook account
  </Step>

  <Step title="Configure email filters (optional)">
    Choose whether to filter emails that trigger your agent. Without a search filter, your agent will receive every email sent to your connected account.

    You can use Microsoft Graph search query syntax to filter emails:

    * `subject:ask us anything` - Subject contains "ask us anything"
    * `body:book a demo` - Body contains "book a demo"
    * `from:*apple.com` - Only emails from apple.com domain
    * `hasAttachments:true` - Only emails with attachments
    * `received:07/23/2018` - Only emails received on this date

    [View full search syntax documentation](https://learn.microsoft.com/en-us/graph/search-query-parameter)
  </Step>

  <Step title="Create a response tool">
    Build a tool that uses the "Send Email (Outlook)" tool step to respond to emails. Equip your agent with this tool in the agent settings.
  </Step>

  <Step title="Configure tool permissions">
    Set the tool permissions to "approval mode" initially so your agent asks for permission before sending emails. Switch to autopilot once you're confident in the responses.
  </Step>

  <Step title="Write agent instructions">
    In the "Core Instructions" section of your agent settings, write a prompt that guides your agent on how to respond to emails.
  </Step>
</Steps>

<Note>
  It can take up to 30 minutes for the first email to come through after setting up the trigger.
</Note>

## Example use cases

Here are some ways you can leverage the Outlook integration with your agents:

<AccordionGroup>
  <Accordion title="Customer Support Email Agent" icon="headset">
    Create an agent that automatically responds to customer support emails, categorizes inquiries, searches your knowledge base for relevant information, and provides helpful responses. The agent can escalate complex issues to human team members and track response times.
  </Accordion>

  <Accordion title="Meeting Scheduler Assistant" icon="calendar-check">
    Build an agent that manages your calendar by finding available time slots, scheduling meetings with attendees, sending calendar invites, and handling rescheduling requests. The agent can check multiple calendars for conflicts and suggest optimal meeting times.
  </Accordion>

  <Accordion title="Email Triage & Organization" icon="inbox">
    Deploy an agent that automatically organizes incoming emails by moving them to appropriate folders, adding labels based on content, flagging urgent messages, and archiving newsletters or promotional emails according to your preferences.
  </Accordion>

  <Accordion title="Lead Qualification Agent" icon="user-check">
    Create an agent that qualifies leads from incoming emails, extracts key information (company size, budget, timeline), updates your CRM, and schedules follow-up calls with qualified prospects while sending polite responses to unqualified leads.
  </Accordion>

  <Accordion title="Meeting Notes & Follow-up" icon="clipboard-list">
    Build an agent that sends meeting summaries and action items to attendees after calendar events, tracks follow-up tasks, and sends reminders for pending action items. The agent can integrate with your project management tools.
  </Accordion>

  <Accordion title="Email Newsletter Manager" icon="newspaper">
    Deploy an agent that manages your email newsletter by collecting subscriber information, organizing contact lists, scheduling newsletter sends, and tracking engagement metrics like open rates and click-throughs.
  </Accordion>

  <Accordion title="Out-of-Office Assistant" icon="plane-departure">
    Create an agent that handles emails while you're away by sending personalized auto-responses, forwarding urgent messages to colleagues, categorizing emails for later review, and scheduling follow-ups for when you return.
  </Accordion>

  <Accordion title="Invoice & Receipt Processor" icon="file-invoice-dollar">
    Build an agent that automatically processes invoices and receipts received via email, extracts key information (amount, date, vendor), saves attachments to cloud storage, and updates your accounting system or expense tracking tools.
  </Accordion>

  <Accordion title="Event Registration Manager" icon="ticket">
    Deploy an agent that handles event registrations via email, confirms attendance, sends calendar invites with event details, manages waitlists, and sends reminder emails before the event.
  </Accordion>

  <Accordion title="Contact Sync & Management" icon="address-book">
    Create an agent that keeps your contacts synchronized across platforms, enriches contact information from email signatures, updates contact details when changes are detected, and removes duplicate entries.
  </Accordion>
</AccordionGroup>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="I converted my account input to an input variable but can't see my account. What should I do?">
    This issue occurs because new Microsoft Teams scopes have been added to the Outlook integration (`OnlineMeetingTranscript.Read.All` and `MailBoxSettings.Read`).

    **Solution**: Update the permission type filter to select the **Microsoft Teams** option. This will allow you to see and select your Outlook account that now includes the expanded Microsoft Teams permissions.

    If you continue to experience issues, try reconnecting your Outlook account through the Integrations & API Keys page to ensure all the latest permissions are granted.
  </Accordion>

  <Accordion title="What permissions does the Outlook integration require?">
    The integration requires permissions to read and write emails, manage calendar events, access contacts, and read mailbox settings. This includes Microsoft Teams scopes for meeting transcripts and mailbox settings. You can review the specific permissions during the authentication process.
  </Accordion>

  <Accordion title="How long does it take for the email trigger to start working?">
    After setting up the Outlook trigger, it can take up to 30 minutes for the first email to come through to your agent. Subsequent emails should be processed more quickly, typically within a few minutes.
  </Accordion>

  <Accordion title="Why aren't emails with large attachments triggering my agent?">
    The Outlook trigger has a 25MB per-attachment limit due to Microsoft Graph API constraints, so emails with attachments larger than 25MB will not be processed by the trigger. To work around this, ask senders to compress large files before sending, convert large DOCX files to PDF format to reduce file size, or use cloud storage links (OneDrive, SharePoint, Dropbox) instead of attaching large files directly. You can also set up a separate workflow to handle large file transfers outside of email. This is a Microsoft Graph API limitation, not a Relevance AI limitation.
  </Accordion>

  <Accordion title="Can I filter which emails trigger my agent?">
    Yes! You can use Microsoft Graph search query syntax to filter emails. For example:

    * Filter by subject: `subject:customer inquiry`
    * Filter by sender: `from:*company.com`
    * Filter by attachments: `hasAttachments:true`
    * Combine filters: `subject:urgent AND from:*vip.com`

    [View full search syntax documentation](https://learn.microsoft.com/en-us/graph/search-query-parameter)
  </Accordion>

  <Accordion title="Can I use multiple Outlook accounts with my agents?">
    Yes, you can connect multiple Outlook accounts through the Integrations & API Keys page. When building tools or setting up triggers, you can select which account to use from the dropdown menu. This is useful for managing different email addresses or separating personal and business communications.
  </Accordion>

  <Accordion title="How do I prevent my agent from sending emails automatically?">
    Set your email-sending tools to "approval mode" in the agent settings. This ensures your agent will always ask for permission before sending an email. You can review the draft, make edits if needed, and then approve or reject the send action.
  </Accordion>

  <Accordion title="Can my agent access shared mailboxes or distribution lists?">
    Yes, if you have the appropriate permissions, your agent can access shared mailboxes. You'll need to use the Microsoft Graph API Call tool step with endpoints like `/users/{userId}/messages` instead of `/me/messages`. Make sure your connected account has delegated access to the shared mailbox.
  </Accordion>

  <Accordion title="What's the difference between the pre-built Outlook tool steps and the API Call tool step?">
    Pre-built tool steps (like "Send Email" or "Create Event") are designed for specific, common tasks and have simplified interfaces with guided inputs. The Microsoft Outlook API Call tool step gives you full access to Microsoft Graph API, allowing you to implement any functionality available in the API, including advanced operations not covered by pre-built steps.
  </Accordion>

  <Accordion title="Are there rate limits for Outlook API calls?">
    Yes, Microsoft Graph API enforces rate limits. The specific limits depend on your Microsoft 365 subscription and the type of requests being made. Your agents should be designed to handle rate limiting gracefully. Microsoft typically returns a `429 Too Many Requests` status code when limits are exceeded.
  </Accordion>

  <Accordion title="Can I access email attachments with the Outlook integration?">
    Yes, you can access email attachments using the Microsoft Graph API. Use the endpoint `GET /me/messages/{id}/attachments` to list attachments, and `GET /me/messages/{id}/attachments/{attachmentId}` to download specific attachments. The attachment content is returned as base64-encoded data.

    **Note**: When using the Outlook trigger, emails with attachments larger than 25MB will not be processed due to Microsoft Graph API limitations.
  </Accordion>

  <Accordion title="How do I remove the Outlook integration?">
    To remove the Outlook integration:

    1. Open your agent settings
    2. Navigate to "Integrations & API Keys"
    3. Click the three dots menu next to your connected Outlook trigger
    4. Click "Remove"

    Once removed, your agent will no longer receive emails or have access to your Outlook data. You can also disconnect the integration entirely from the main Integrations & API Keys page.
  </Accordion>

  <Accordion title="Can I use the Outlook integration with Office 365 or Microsoft 365?">
    Yes! The Outlook integration works with Outlook.com, Office 365, and Microsoft 365 accounts. All these services use the same Microsoft Graph API, so the integration functions identically across all Microsoft email platforms.
  </Accordion>
</AccordionGroup>


# Salesforce
Source: https://relevanceai.com/docs/integrations/popular-integrations/salesforce

Interact with your Salesforce CRM with Relevance.

Salesforce is designed to help businesses manage their sales pipeline, primarily through customer relationship tracking activities. With Relevance AI, your agents can interact with Salesforce much like your human sales team would.

Below, we'll show you how to get started with the Salesforce integration, including how to connect your Salesforce account, how to trigger agentic workflows based on Salesforce events, and how to complete common CRM activities with your agents.

## How to connect your Salesforce account

To add Salesforce as an integration, all you need to do is sign-into your Salesforce account via the integrations page:

1. Go to the "Integrations" page in the sidebar.
2. Click on "Salesforce".
3. Click on the Add Integration button.
4. Sign-in to your Salesforce account.
5. Click "Allow access" to Relevance AI.

## Setup Salesforce as an Agent Trigger

A trigger is an event that triggers your agent to start working. In the case of Salesforce, a trigger might be new leads added to your CRM, demos have been booked, a lead has moved down the sales funnel (e.g. Prospect -> Qualified lead) or other events unique to your sales workflows.

### Add a Salesforce trigger to your agent.

After you have created a new agent, you can add Salesforce as a trigger:

1. Open your agent settings (Agent profile tab).
2. Scroll down to Integrations > Triggers.
3. Click on Salesforce.
4. Select your connected Salesforce account, or sign-in if not already added via the integrations page (previous section).
5. Click "Allow access" to Relevance AI.
6. Write an SOQL query to retrieve the data you want to pull in from your Salesforce account.

Here is an example on an SOQL query that pulls in new leads:

The following SOQL query pulls in Lead objects with the selected properties, where the "Relevance\_Outreached\_\_c" custom object has been set to "true".

```SOQL theme={null}
SELECT Id, Name, Email, Phone, AccountId, Account.Name, Title
FROM Lead
WHERE Relevance_Outreached__c = true
```

Now, any new Leads that are added to the CRM since the last time we checked that meet the query conditions, will be sent to the agent.

<Info>Make sure that you are writing [Salesforce Object Query Language (SOQL) queries](https://developer.salesforce.com/docs/atlas.en-us.soql_sosl.meta/soql_sosl/sforce_api_calls_soql.htm), and not SQL queries here. SQL queries that use order-by statements for example, will fail.</Info>

7. Advanced option (Optional): Specify a frequency, which is how often you want to fetch data from Salesforce. If you don't specify a frequency, it'll check every 1 minute.

8. Set a Cadence (Optional): Specificy a cadence, which is how often one of the items fetched from Salesforce triggers the agent (if 100 new leads are added, you might not want your agent to outreach to them all the same day, especially if you're in the middle of warming up your inboxes). The rest will wait in a queue.

## Get your agent to complete common Salesforce activities

Once your account is connected, you can use the Salesforce API call step in the Tool builder. This tool step allows you to complete Salesforce-specific activities, like managing Leads, Contacts, Notes, Tasks and more.

### Use Salesforce API tool-step

You can build custom tools that perform Salesforce activities, by using the Salesforce API Call tool-step:

1. Create a new tool.
2. Scroll down to Tool-steps.
3. Add Salesforce API tool-step.
4. Select your connected Salesforce account in the dropdown.

Below are examples of common Salesforce activities your agents can perform.

### Get an existing Contact's details.

<img />

### Create a new Lead.

<img />

### Create a note (e.g. lead qual research).

<img />

### Create a task

<img />

### Mark a task as complete

<img />

## Troubleshooting

<AccordionGroup>
  <Accordion title="OAuth approval error when connecting">
    If you encounter an `OAUTH_APPROVAL_ERROR_GENERIC` error when authenticating with Salesforce, this is typically due to Salesforce security updates implemented in September 2024 that require explicit installation of connected apps.

    This issue commonly occurs with:

    * Sandbox environments created before September 8th, 2024
    * New Salesforce instances
    * Users without System Administrator permissions

    To resolve this issue:

    1. **Verify System Administrator access**
       * The first user to authenticate must have System Administrator permissions or the "Approve Uninstalled Connected Apps" permission
       * Have a System Administrator complete the initial authentication with Relevance AI
    2. **For Sandbox environments created before September 8th:**
       * Go to **Setup** → **Company Information**
       * Click **"Match Production Licenses"** to update the sandbox with the necessary infrastructure
    3. **Install the Relevance AI connected app:**
       * After a System Administrator successfully authenticates once, go to **Setup**
       * In Quick Find, type **"OAuth"**
       * Navigate to **Connected Apps OAuth Usage**
       * Locate the Relevance AI app in the list
       * Click **"Install"** to approve the connected app for your organization
    4. **Grant permissions to other users (optional):**
       * System Administrators can grant the **"Approve Uninstalled Connected Apps"** permission to other users who need to authenticate

    <Note>Scratch orgs created before September 8th, 2024 cannot have connected apps added and must be recreated. If issues persist after following these steps, contact Salesforce support for assistance with connected app authentication in your organization.</Note>

    For more information, see the [Salesforce documentation](https://help.salesforce.com/s/articleView?id=005132365\&type=1).
  </Accordion>
</AccordionGroup>

## Remove Salesforce integration

You can remove Salesforce as a Trigger in your agent settings, by clicking the three dots next to your connected account, and then "Remove".

If you want to remove the Salesforce integration completely, you can:

1. Go to the "Integrations" page from the sidebar
2. Select Salesforce from the list
3. Click "..." on the account you want to remove
4. Click remove


# Slack
Source: https://relevanceai.com/docs/integrations/popular-integrations/slack

Relevance AI's integration for Slack allows you to connect your Relevance AI Agents with Slack, enabling seamless communication between your AI Workforce and your team's Slack workspace.

<Note>
  The Relevance AI team are constantly making improvements to the Slack integrations. If a new Slack feature isn't showing on your account, try to reconnect your Slack integration. To see our changes, sometimes integrations need a reconnect or marketplace re-approval.
</Note>

To add Slack as an integration, follow these simple steps:

1. Log into Relevance AI: [https://app.relevanceai.com](https://app.relevanceai.com)
2. Go to the **Integrations & API Keys** page in the sidebar of your Relevance AI dashboard.
3. Click on "Slack" from the available integrations.
4. Click on the "Add Integration" button.
5. In the pop-up window, sign into your Slack workspace and authorize Relevance AI to access your Slack account.
6. Select the channels you want to grant access to.
7. Click "Allow" to complete the connection.

<div>
  <iframe title="Relevanceai Demo" />
</div>

<Callout icon="triangle-exclamation">
  This integration for Slack was developed by our team and is not affiliated with Slack. Please contact our Support for any questions/queries you may have.
</Callout>

## Linking your account via magic link

<Note>Both the magic link flow and the standard OAuth flow achieve the same outcome — your Slack account is linked to your Relevance AI account.</Note>

You may receive a magic link from Slack or via email inviting you to connect your accounts. This is an alternative to the standard OAuth flow above and achieves the same result.

1. Click the magic link in Slack or your email.
2. If you're not logged in to Relevance AI, you'll be redirected to sign in first. After signing in, you'll be returned to the consent page automatically.
3. On the consent page, review your Relevance AI account email and your Slack account details shown.
4. Click **Confirm** to link the accounts, or **Cancel** to abort the connection.

## Understanding Slack connections in Relevance AI

Relevance AI uses two distinct Slack authentication methods depending on which part of the platform you're using. They are independent of each other, and your Slack workspace admin can allow or block each type separately.

<CardGroup>
  <Card title="Builder: Agents, Tools, Triggers, Workforces" icon="robot">
    Uses a workspace-level bot token through the Relevance AI Slack app. Installed once for the whole workspace, it lets agents post messages, respond to triggers, and send notifications. Requires a Slack workspace admin to approve the Relevance AI app.
  </Card>

  <Card title="Per-user OAuth" icon="bolt">
    Each user connects their own Slack account individually through OAuth, granting Relevance AI access to act on that user's behalf in Slack.
  </Card>
</CardGroup>

<Info>
  If Slack works in one part of Relevance AI but not another, your workspace admin has likely approved one connection type but not the other. Ask your admin to check permissions for both the Relevance AI Slack app and per-user OAuth app installations.
</Info>

Both connection types can coexist in the same workspace.

## Triggering Agents or Workforces from Slack

<img alt="Slack Triggers New Pn" />

You can trigger your **Agents** or **Workforces** directly from Slack channels or your own DM. This will start your Agent or Workforce in Relevance AI based on a keyword of your choice, and then will reply to your message in Slack.

You can now trigger entire Workforces from Slack, not just individual Agents. This allows you to kick off complex multi-agent workflows directly from your Slack conversations. Learn more about [Workforce Triggers](/docs/build/workforces/build-an-ai-workforce/add-triggers).

### Setting up triggers for channels

To trigger from a **channel**, first **invite Relevance AI to the Slack channel** by typing `/invite @Relevance AI` into the channel.

### Setting up triggers for DMs

To trigger from your **own DM**, open a DM with the Relevance AI bot in Slack and **send it any message**. This establishes the DM connection and makes your DM appear in the channel selection dropdown.

<Tip>
  **Channel vs DM setup:**

  1. **For channels:** Use `/invite @Relevance AI` in the channel
  2. **For your DM:** Send a message to the Relevance AI bot (any message works)
</Tip>

Then, set up your Trigger from Relevance AI by following the steps below:

<div>
  <iframe title="How to set up the Slack trigger" />
</div>

1. **To trigger your Agent:** enter your Triggers page in your Agent, and create a new Trigger. **To trigger your Workforce:** enter your Workforce page, select a Trigger.

2. Select Slack.

3. Connect or select your Slack account, then click 'Continue'.

4. Select the channel or DM you want your Agent to be triggered on.

5. Enter the keyword you want the Agent to be triggered on - if you want the Agent to be triggered on all messages that tag `@Relevance AI`, leave this blank.

6. Click 'Continue'.

7. Queue work hours if needed.

8. Click 'Setup trigger'.

<Note>
  Once setup, you can trigger your Agent from Slack by tagging `@Relevance AI` and mentioning the keyword you've set (if you set a keyword)! You can also keep the conversation rolling in threads - simply `@Relevance AI` in thread for your agent to respond, with full context of the conversation.
</Note>

### Advanced Trigger Settings

#### Live Status Updates

When your Agent or Workforce is triggered from Slack, you'll receive live status updates as it progresses through tasks. These updates are automatically enabled and appear directly in the Slack thread, showing you what your Agent is doing in real-time.

#### Exclude Keywords

You can specify keywords that will prevent your Agent from triggering, even when messages mention your trigger keyword or tag `@Relevance AI`.

**To configure:**

1. In your Slack trigger settings, expand **Advanced Settings**
2. Find the **Exclude Keywords** field
3. Enter keywords or phrases (comma-separated)
4. Save your configuration

If a message contains any of your exclude keywords, the Agent will not trigger. Keywords are case-insensitive.

#### No Agent Reply

Enable this setting to have your Agent process Slack messages without posting a response back to the channel. This is useful for background processing, data collection, or triggering workflows that complete elsewhere.

**To enable:**

1. In your Slack trigger settings, expand **Advanced Settings**
2. Toggle on **"No Agent Reply"** or **"Disable Agent responses"**
3. Save your configuration

<Warning>
  When "No Agent Reply" is enabled, you won't receive confirmation in Slack that your Agent processed the message. Monitor your Agent's task history in Relevance AI to verify it's working correctly.
</Warning>

### Customize Message Formatting

When using the Send Message (advanced) tool step, you can send messages through Slack with enhanced formatting. This includes:

1. **Emojis:** Add emojis to make your messages more engaging and expressive.
2. **Bolded text:** Use bold formatting to highlight important information.
3. **Images:** Include images to provide visual context or instructions.

## Escalate your Agent to Slack

<Warning>
  This feature may be deprecated in the future and an Agent Notification feature will be added instead.
</Warning>

To escalate your agent to Slack, follow these steps:

1. Go to the "Agents" page in the sidebar of your Relevance AI dashboard.
2. Select the agent you want to escalate.
3. Navigate to the Build tab of your agent (top center)
4. Click on "Escalations" in the left sidebar.

<div>
  <iframe title="Escalate to Slack Channel" />
</div>

## Agent Notifications

<Warning>
  This feature is currently in beta for some users
</Warning>

1. Under "Agent Notifications", click "Add agent notification"
2. Configure your notification settings:
   * Select "Slack" as the platform
   * Choose the specific task statuses you want to trigger the notifications with (e.g., "Running" to trigger whenever the agent runs)
   * Select a Slack account that you have previously authed or add a new Slack account
   * Select a Slack channel you want to notify with new agent messages
3. Invite the Relevance AI agent to the Slack channel you want to notify by typing "/invite @RelevanceAI" in the channel.
4. Publish changes to your agent by clicking "Publish changes"
5. Run your agent by selecting "Run" at the top of the page.
6. Prompt the agent with some text, e.g. "Give me a summary of what Slack does".
7. You can reply to the agent's notification directly in Slack, or if you'd prefer you can click on "View task" to navigate back to the task in Relevance AI and perform further actions.

<div>
  <iframe title="Connect Slack Agent Notifications" />
</div>

Once connected, your integration for Slack will be available for use with your Agents and Tools.

## Making API calls with Slack using Tool steps

You can configure your Slack agent to make API calls to Slack via Relevance AI's tool builder. This allows you to perform actions such as posting Slack messages to channels, or reading messages from channels. These are just a few examples of the many Slack actions available. Your agents can combine these actions with their reasoning capabilities to create sophisticated workflows that enhance team communication and productivity.

<Tip>
  Type in "Slack" in the search bar to see all the available Slack actions.
</Tip>

<div>
  <iframe title="Slack Tools - Send a Slack Message via a Tool Step" />
</div>

### All Slack Tool steps

You can learn more about each of the Slack Tool steps on the following pages.

<Snippet />

## Remove integration for Slack

If you need to remove the integration for Slack:

1. Go to the **Integrations & API Keys** page from the sidebar.
2. Search for Slack from the list.
3. Click "..." on the account you want to remove.
4. Click "Remove" and confirm your choice.

This will disconnect your Slack workspace from Relevance AI, and your agents will no longer be able to interact with Slack until you reconnect the integration for Slack.

<div>
  <iframe title="Remove an Integration for Slack" />
</div>

## Troubleshooting

<AccordionGroup>
  <Accordion title="Magic link expired or not working">
    Magic links are time-limited and may expire before you click them. If your magic link no longer works, connect your Slack account directly through the standard OAuth flow: go to **Integrations & API Keys** in the sidebar, find Slack, and click **Add Integration**.
  </Accordion>

  <Accordion title="Consent page shows wrong email address">
    The consent page shows two email addresses: your Relevance AI account email and your Slack account email. If either address looks incorrect, click **Cancel** and verify that you're logged in to the correct Relevance AI account before trying again. To switch accounts, sign out of Relevance AI and sign in with the correct account, then click the magic link again.
  </Accordion>
</AccordionGroup>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can I use my own Slack DM as a trigger?">
    Yes! You can use your own Slack DM as a trigger source. Here's how to set it up:

    1. Open a DM with the Relevance AI bot in Slack
    2. Send the bot any message (this establishes the DM connection)
    3. Your DM will then appear in the channel selection dropdown when setting up triggers in Relevance AI

    **Note:** For DMs, you don't use the `/invite` command. Simply sending a message to the bot is enough to establish the connection. If your DM doesn't appear in the channel list, make sure you've sent at least one message to the Relevance AI bot first.
  </Accordion>

  <Accordion title="What permissions do I need to trigger an agent via Slack DM?">
    To trigger an agent via direct message (DM) in Slack, you must have Editor permissions or higher on that specific agent.
  </Accordion>

  <Accordion title="How do I send specific messages to someone on Slack?">
    If you want to be able to send a message to a specific person in Slack, that person will first need to send a message to the Relevance AI bot in Slack. Once they do this, their name will appear as a destination option when sending Slack messages.
  </Accordion>

  <Accordion title="Slack isn't working as intended, how do I fix it?">
    The most common reason Slack isn't working as expected is that your integration needs to be updated. We occasionally release improvements to our Slack app, and older installations don't pick these up automatically. Updating only takes a moment - go to your project's integrations page, find your Slack connection and click 'Reconnect'.
  </Accordion>

  <Accordion title="When I setup a Slack trigger, I can't find channels in my workspace.">
    This usually happens for one of two reasons. First, the email address you use in Slack may not match the email you used when signing up to Relevance - we require these to be the same. Second, the Relevance AI app might not have been invited to the channel yet. To fix this, open the channel and run `/invite @RelevanceAI`, and it should appear.
  </Accordion>

  <Accordion title="What's the easiest way to install the Relevance AI Slack App?">
    The easiest and most reliable method is to install Slack directly through Relevance. Start the Slack integration connection flow in Relevance and a request will automatically be sent to your Slack admin for approval. In some cases, admins install the Relevance AI app in Slack before the integration is set up in Relevance, which can create sync issues. Installing via Relevance avoids this.
  </Accordion>

  <Accordion title="Can I see what my Agent is doing when triggered from Slack?">
    Yes! Live status updates are automatically enabled for all Slack triggers. You'll see real-time progress updates posted in the Slack thread as your Agent works through its tasks.
  </Accordion>

  <Accordion title="How do I prevent my Agent from responding to certain messages?">
    Use the **Exclude Keywords** feature in your Slack trigger's advanced settings. Add keywords or phrases that should prevent your Agent from triggering, even if the message mentions your trigger keyword or tags `@Relevance AI`.
  </Accordion>

  <Accordion title="Can my Agent process Slack messages without replying?">
    Yes! Enable the **"No Agent Reply"** setting in your trigger's advanced settings. This allows your Agent to process messages without posting a response back to Slack.
  </Accordion>

  <Accordion title="What's the difference between Agent triggers and Workforce triggers in Slack?">
    Agent triggers start a single AI Agent, while Workforce triggers kick off an entire multi-agent workflow with multiple Agents working together. Use Workforce triggers for complex, multi-step processes that require different specialized Agents to collaborate.
  </Accordion>

  <Accordion title="What is the magic link I received from Slack?">
    A magic link is an alternative way to connect your Slack account to Relevance AI. When you click the link, you're taken to a consent page in Relevance AI that shows your Relevance AI account email and your Slack account details. Click **Confirm** to complete the connection or **Cancel** to abort it. The magic link method achieves the same result as connecting via the standard OAuth flow on the Integrations page.
  </Accordion>

  <Accordion title="Do I need to use the magic link or can I connect directly from Relevance AI?">
    Both methods work and achieve the same result. You can connect your Slack account through the standard OAuth flow on the **Integrations & API Keys** page, or by clicking a magic link you received from Slack or via email. Use whichever is more convenient.
  </Accordion>

  <Accordion title="What happens if I'm not logged in when I click the magic link?">
    If you're not logged in to Relevance AI when you click the magic link, you'll be redirected to the sign-in page. After signing in, you'll be automatically returned to the consent page to complete the account linking.
  </Accordion>

  <Accordion title="Why does Slack work in one part of Relevance AI but not another?">
    Relevance AI uses two different Slack authentication methods. Builder features (Agents, Tools, Triggers, Workforces) use a workspace-level bot token through the Relevance AI Slack app, which requires a workspace admin to approve the app installation. The other method uses per-user OAuth, where each user connects their own Slack account individually.

    If Slack works in one area but not the other, your workspace admin has likely approved one connection type without the other. Ask your admin to check the app permission settings for both the Relevance AI Slack app and per-user OAuth app installations. Both connection types can coexist once both are approved.
  </Accordion>
</AccordionGroup>

## Disclaimer

Relevance AI enables users to access verified and genuine generative models from secure and trusted vendors. However, generative AI technologies still leave the potential for hallucinations leading to potentially inaccurate responses. Please review and verify the agent's output and use a combination of deterministic and agentic behaviour to achieve the most reliable results.

## Privacy Policy

To find out more about how Relevance AI handles your data, please see our [Privacy Policy](https://relevanceai.com/privacy-policy).

## Follow along on YouTube

<Frame>
  <iframe title="How to Manage Your AI Agents Directly in Slack" />
</Frame>


# Telegram
Source: https://relevanceai.com/docs/integrations/popular-integrations/telegram

Interact with your Telegram account with Relevance.

# Telegram Integration

Connect your AI agents with Telegram to automate conversations and workflows directly through the popular messaging platform. The Telegram integration allows your agents to receive messages, respond intelligently, and execute actions based on user interactions.

## How the Telegram Integration Works

The Relevance AI Telegram integration connects your **personal Telegram account** (or a dedicated Telegram account you create). When you set up this integration:

* Your AI agent sends and receives messages **as that Telegram account**
* You authenticate using Telegram's standard login process
* The agent can interact with your contacts, groups, and channels

## Connect the integration

Setting up the Telegram integration with Relevance AI is straightforward:

1. Go to the "Integrations & API Keys" page in the sidebar of your Relevance AI dashboard
2. Click on "Telegram" from the available integrations
3. Click the "Add Integration" button
4. You'll be prompted to authenticate with your Telegram account
5. Follow the on-screen instructions to authorize the connection
6. Once connected, you'll see your Telegram account listed under connected accounts

## Setting up triggers

The Telegram integration can be configured as a trigger for your AI agents, allowing them to automatically respond when messages are received.

<Warning>The Telegram Trigger is a Premium trigger, available on the Pro plan and above.</Warning>

### Create a Telegram trigger

1. Navigate to your agent's profile page and "Triggers"
2. Choose "Telegram" from the list of available triggers
3. Select your connected Telegram account
4. Configure the trigger settings:
   * **Chat Type**: Choose between private chats, group chats, or both
   * **Message Filter**: Optionally set up filters to only trigger on specific message types or content
   * **Response Mode**: Select whether your agent should respond automatically or require approval

### Common trigger use cases

* **Customer Support**: Automatically respond to customer inquiries received via Telegram

* **Lead Qualification**: Trigger an agent to qualify leads that reach out through Telegram

* **Information Requests**: Set up an agent to provide information when users ask specific questions

* **Appointment Scheduling**: Allow users to schedule appointments through Telegram conversations

* **Order Processing**: Process orders and transactions initiated through Telegram messages

## Tools & Tool Steps

The Telegram integration provides several powerful actions that your agents can use as part of their workflows. These actions enable your agents to interact with Telegram in various ways.

### Available Telegram actions

* **Send Message**: Send a text message to a specific chat or user
* **Send Photo**: Share images with optional captions
* **Send Document**: Share files and documents of various formats
* **Send Location**: Share geographical locations
* **Create Poll**: Create and send polls to gather feedback
* **Get Chat Information**: Retrieve details about a specific chat
* **Get User Profile**: Retrieve information about a Telegram user
* **Pin Message**: Pin important messages in a chat
* **Delete Message**: Remove messages from a conversation
* **Forward Message**: Forward messages between chats
* **Create Chat Invite Link**: Generate invitation links for groups or channels

These are just some of the many actions available. The Telegram integration offers a comprehensive set of tools that allow your agents to leverage the full functionality of the Telegram platform.

### Example workflow

Here's an example of how you might use Telegram actions in a customer support workflow:

1. **Trigger**: Agent is triggered by an incoming Telegram message
2. **Analysis**: Agent analyzes the message content to determine the customer's intent
3. **Response**: Agent uses the "Send Message" action to respond with an appropriate greeting
4. **Information Gathering**: If additional information is needed, agent asks follow-up questions
5. **Resolution**: Once the issue is understood, agent provides a solution
6. **Documentation**: Agent uses "Send Document" action to share relevant documentation if needed
7. **Follow-up**: Agent schedules a follow-up message for later if the issue requires monitoring

## Use the integration's API tool step (advanced)

In addition to the pre-built actions, advanced users can access additional Telegram functionality through the Telegram API tool step:

1. Create a new tool
2. Scroll down to Tool-steps
3. Add Telegram API tool-step
4. Select your connected Telegram account in the dropdown
5. Configure the API endpoint and parameters according to your needs

**Note**: This tool step uses your connected personal Telegram account's authentication. While it provides access to Telegram's API methods, it operates within the context of your authenticated account, not as a BotFather bot. The available methods depend on what Telegram allows for user accounts versus bots.

For full Telegram Bot API functionality with a BotFather bot, use the [API tool step](https://relevanceai.com/docs/tool/tool-steps/api) with your bot token instead.

### Example API usage

```json theme={null}
{
  "method": "sendMessage",
  "chat_id": "{{chatId}}",
  "text": "This is a custom formatted message with *bold* and _italic_ text",
  "parse_mode": "Markdown"
}
```

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Can I connect multiple Telegram accounts to Relevance AI?">Yes, you can connect multiple Telegram accounts to Relevance AI.</Accordion>
  <Accordion title="Does the Telegram integration work with both personal and business Telegram accounts?">Yes, the integration works with both personal accounts and business accounts on Telegram.</Accordion>
  <Accordion title="Can my agent respond to messages in Telegram groups?">Yes, your agent can be configured to respond to messages in private chats, groups, or both, depending on your trigger settings.</Accordion>
  <Accordion title="Is there a limit to how many messages my agent can process through Telegram?">Message processing is counted against your overall credit usage. The specific limits depend on your subscription plan.</Accordion>
  <Accordion title="Can I customize how my agent appears in Telegram conversations?">Yes, you can customize the name and profile picture of your Telegram bot that represents your agent in conversations.</Accordion>

  <Accordion title="Can I use a BotFather bot token with this integration?">
    No, the native Telegram integration does not support BotFather bot tokens. This integration connects your personal Telegram account (or a dedicated account), and your agent operates as that account. If you want to use BotFather, you should build a custom Agent with a webhook trigger and custom API calls.
  </Accordion>
</AccordionGroup>


# Trello
Source: https://relevanceai.com/docs/integrations/popular-integrations/trello

Enhance your day-to-day team and personal coordination with the Trello integration for Relevance AI.

Trello is a flexible personal productivity and work management tool that helps individuals and teams visually organize tasks, ideas, and workflows using boards, lists, and cards. It's designed to adapt to a wide range of use cases — from planning daily to-dos and tracking habits to managing team coordination or creative pipelines. With its drag-and-drop interface, customizable labels, and integrations with tools like Slack, Google Drive, and now Relevance AI, Trello makes it easy to stay on top of work, whether you're managing solo projects or collaborating asynchronously.

<Warning>
  The Trello integration in Relevance AI was built by Relevance AI, and is therefore supported by our team, not Atlassian (Trello's parent company). If you have a question / issue with using Trello in Relevance AI, please [reach out to our support team](/docs/get-started/support). If you have a question / issue that is only about Trello, you can [reach out to Atlassian support](https://support.atlassian.com/contact/).
</Warning>

## Connect the integration

Setting up the Trello integration with Relevance AI is straightforward:

1. Navigate to the Integrations & App Keys section in your Relevance AI dashboard
2. Find and select "Trello" from the available integrations
3. Click "Connect"
4. You'll be redirected to Trello to authorize the connection
5. Log in to your Trello account (or create one if needed)
6. Review and approve the permissions requested
7. Once authorized, you'll be redirected back to Relevance AI with the integration active

After connecting, your Relevance AI agents will be able to interact with your Trello account, accessing boards, creating cards, and managing your Trello workflows.

## Trello API authentication

In addition to OAuth authentication, you can use Trello's native API key and token authentication method. This approach is useful for server-to-server integrations, automation scripts, or when you need more granular control over API access.

### When to use API key + token authentication

* **OAuth (recommended for most users)**: Best for user-facing integrations where you want users to authorize access through Trello's interface
* **API key + token**: Ideal for backend automation, service accounts, or when you need persistent access without user interaction

### Getting your API key

<Steps>
  <Step title="Access Trello Power-Ups Admin">
    Navigate to [https://trello.com/power-ups/admin](https://trello.com/power-ups/admin) while logged into your Trello account.
  </Step>

  <Step title="Create a new Power-Up or API key">
    If you haven't created a Power-Up before, click "New" to create one. You'll need to provide a name for your integration (e.g., "Relevance AI Integration").
  </Step>

  <Step title="Copy your API key">
    Once created, you'll see your API key displayed. Copy this key and store it securely.
  </Step>
</Steps>

### Generating a token

After obtaining your API key, you need to generate a token:

<Steps>
  <Step title="Generate token URL">
    Visit the following URL in your browser, replacing `YOUR_API_KEY` with your actual API key:

    ```
    https://trello.com/1/authorize?expiration=never&name=RelevanceAI&scope=read,write&response_type=token&key=YOUR_API_KEY
    ```
  </Step>

  <Step title="Authorize the token">
    You'll be prompted to authorize the token. Review the permissions and click "Allow" to generate your token.
  </Step>

  <Step title="Copy your token">
    After authorization, Trello will display your token. Copy and store it securely alongside your API key.
  </Step>
</Steps>

### Using API credentials in Relevance AI

When configuring Trello tool steps in Relevance AI, you can provide your API key and token directly in the tool configuration. This allows your agents to authenticate with Trello without requiring OAuth authorization.

<Warning>
  Store your API key and token securely. Never commit them to version control or share them publicly. Treat them like passwords.
</Warning>

<Info>
  For more details on Trello's API authentication, see the [official Trello API documentation](https://developer.atlassian.com/cloud/trello/guides/rest-api/api-introduction/).
</Info>

## Available triggers

The Trello integration allows you to set up triggers that automatically activate your AI agents when specific events occur in your Trello boards. Most Trello triggers are **instant** (webhook-based), meaning your agents respond in real-time as soon as changes happen — whether it's a new card being created, a comment being added, or a card moving between lists. This enables powerful automation workflows that keep your team productive without manual intervention.

The following trigger options are available, organized by category:

<AccordionGroup>
  <Accordion title="Card-Related Triggers" icon="square-check">
    Monitor card lifecycle events to automate workflows around task management.

    * **New Card (Instant)** — Triggers immediately when a new card is created on a board. Perfect for automatically assigning tasks, sending notifications, or initiating approval workflows.

    * **Card Updated (Instant)** — Triggers whenever any changes are made to a card (title, description, due date, etc.). Useful for tracking modifications and keeping external systems synchronized.

    * **Card Moved (Instant)** — Triggers when a card is moved to a different list. Ideal for stage-based workflows like moving from "In Progress" to "Review" or "Done."

    * **Card Archived (Instant)** — Triggers when a card is archived. Use this to log completed work, update external databases, or trigger cleanup processes.

    * **Card Due Date Reminder** — Triggers at a specified time before a card's due date. Configure how far in advance you want to be notified (e.g., 1 day, 1 hour before due).
  </Accordion>

  <Accordion title="Comment & Collaboration Triggers" icon="comments">
    Stay informed about team discussions and collaboration activities.

    * **New Comment Added to Card (Instant)** — Triggers immediately when someone adds a comment to any card. Great for monitoring discussions, extracting action items, or routing questions to the right team members.

    * **New Member on Card (Instant)** — Triggers when a team member is added to a card. Use this to send welcome messages, provide context, or update assignment tracking systems.
  </Accordion>

  <Accordion title="Label Triggers" icon="tags">
    Automate workflows based on card categorization and organization.

    * **New Label Created (Instant)** — Triggers when a new label is created on a board. Useful for maintaining label consistency across multiple boards or updating documentation.

    * **New Label Added To Card (Instant)** — Triggers when a label is applied to a card. Perfect for routing cards to specific workflows based on priority, category, or department labels.
  </Accordion>

  <Accordion title="Checklist Triggers" icon="list-check">
    Monitor task breakdown and subtask management.

    * **New Checklist (Instant)** — Triggers when a new checklist is added to any card on a board. Use this to track task decomposition or ensure proper subtask templates are being used.
  </Accordion>

  <Accordion title="Board & Activity Triggers" icon="table-columns">
    Track high-level board changes and comprehensive activity monitoring.

    * **New Board (Instant)** — Triggers when a new board is created in your Trello workspace. Ideal for setting up board templates, applying default settings, or notifying administrators.

    * **New Board Activity (Instant)** — Triggers for any activity on a board with extensive filtering options. This is the most flexible trigger, allowing you to monitor multiple activity types simultaneously (card creation, updates, comments, member changes, etc.).

    * **New Notification** — Triggers when you receive a new Trello notification. Use this to centralize all your Trello alerts or create custom notification routing.
  </Accordion>

  <Accordion title="Attachment Triggers" icon="paperclip">
    Monitor file and document additions to cards.

    * **New Attachment (Instant)** — Triggers when a file or link is attached to any card on a board. Perfect for processing uploaded documents, backing up files, or extracting information from attachments.
  </Accordion>

  <Accordion title="Advanced Triggers" icon="wand-magic-sparkles">
    Highly customizable triggers for complex automation scenarios.

    * **Custom Webhook Events (Instant)** — The most powerful trigger option, allowing you to create highly specific conditions by combining multiple filters:
      * Filter by specific boards
      * Filter by event types (create, update, delete, etc.)
      * Filter by specific lists
      * Filter by specific cards
      * Combine multiple conditions for precise control

    This trigger is ideal when you need granular control over exactly which events should activate your agent.
  </Accordion>
</AccordionGroup>

## Practical use cases

Here are some powerful ways to use Trello triggers with Relevance AI:

<Columns>
  <Card title="Automated Task Assignment" icon="user-plus">
    When a new card is created in your "Incoming Requests" list, automatically analyze the card content and assign it to the appropriate team member based on keywords or workload.
  </Card>

  <Card title="Status Update Notifications" icon="bell">
    When a card moves to your "Completed" list, automatically send a summary email to stakeholders and update your project management dashboard.
  </Card>

  <Card title="Comment Analysis" icon="message-bot">
    When a new comment is added, use AI to detect questions, action items, or escalations, then route them appropriately or create follow-up tasks.
  </Card>

  <Card title="Due Date Management" icon="clock">
    Set up reminders that trigger 24 hours before a card is due, sending personalized notifications and checking if all checklist items are complete.
  </Card>

  <Card title="Label-Based Routing" icon="route">
    When a "Priority" or "Urgent" label is added to a card, automatically notify relevant team members and escalate to management if needed.
  </Card>

  <Card title="Document Processing" icon="file-lines">
    When an attachment is added to a card, automatically extract key information, generate summaries, or validate document completeness.
  </Card>
</Columns>

<Tip>
  **Pro tip:** Combine multiple Trello triggers with different agents to create sophisticated workflows. For example, use one agent to process new cards and another to handle comments, creating a complete automation ecosystem for your Trello boards.
</Tip>

## Tool steps for Trello

The Trello integration provides a comprehensive set of actions that your agents can use to interact with your boards and workflows. These actions can be incorporated into your agent's workflows as tool steps, enabling automation capabilities.

### Card Management

<CardGroup>
  <Card title="Create Card" icon="square-plus">
    Create a new card on a Trello board
  </Card>

  <Card title="Update Card" icon="pen-to-square">
    Update an existing card's properties
  </Card>

  <Card title="Get Card" icon="square">
    Retrieve details about a specific card
  </Card>

  <Card title="Archive Card" icon="box-archive">
    Archive a card from a board
  </Card>
</CardGroup>

### Board & List Operations

<CardGroup>
  <Card title="Get Boards" icon="table-columns">
    List all boards accessible to your account
  </Card>

  <Card title="Get List Cards" icon="list">
    Retrieve all cards from a specific list
  </Card>

  <Card title="Create List" icon="list-ul">
    Create a new list on a board
  </Card>
</CardGroup>

### Comments & Collaboration

<CardGroup>
  <Card title="Add Comment to Card" icon="comment">
    Post a comment on a card
  </Card>

  <Card title="Add Comment" icon="message">
    Add a comment to a card (legacy action)
  </Card>
</CardGroup>

### Checklist Management

<CardGroup>
  <Card title="Add Checklist" icon="list-check">
    Add a checklist to a card
  </Card>

  <Card title="Create a Checklist Item" icon="square-check">
    Add an item to an existing checklist
  </Card>
</CardGroup>

### Advanced Operations

<CardGroup>
  <Card title="Trello API Call" icon="code">
    Make custom API calls to any Trello endpoint
  </Card>
</CardGroup>

<Tip>
  Type "Trello" in the tool step search bar to see all available Trello actions when building your tools.
</Tip>

## Use the integration's API tool step (advanced)

In addition to the pre-built actions available in the tool directory, you can build custom tools that perform Trello-specific activities using the Trello API Call tool step.

### How to use the Trello API Call tool step

<Steps>
  <Step title="Create a new tool">
    Create a new tool in Relevance AI or open an existing tool you want to add Trello functionality to.
  </Step>

  <Step title="Add the Trello API Call tool step">
    1. Scroll down to Tool steps
    2. Search for "Trello API Call" in the tool step search bar
    3. Add the Trello API Call tool step to your workflow
  </Step>

  <Step title="Select your authentication method">
    Choose between OAuth (connected account) or API key + token authentication.
  </Step>

  <Step title="Configure the API endpoint">
    Configure the API endpoint, method, and parameters:

    * **Method**: Select the HTTP method (GET, POST, PUT, DELETE)
    * **Endpoint**: Enter the API endpoint path (e.g., `/1/boards/{id}/cards`)
    * **Body**: Add any required request body data
    * **Query Parameters**: Add URL parameters as needed
  </Step>

  <Step title="Test your API call">
    Test your configuration to ensure it works correctly before deploying.
  </Step>
</Steps>

### Example: Creating a card with custom fields

Here's a practical example of using the Trello API Call tool step to create a card with specific properties:

**API Endpoint**: `POST /1/cards`

**Configuration**:

```json theme={null}
{
  "method": "POST",
  "endpoint": "/1/cards",
  "body": {
    "idList": "5f8a1b2c3d4e5f6g7h8i9j0k",
    "name": "New Task from Agent",
    "desc": "This card was created automatically by Relevance AI",
    "pos": "top",
    "due": "2024-12-31",
    "idLabels": ["5f8a1b2c3d4e5f6g7h8i9j0k"]
  }
}
```

This configuration:

* Uses the POST method to create a new card
* Specifies the list ID where the card should be created
* Sets the card name and description
* Positions the card at the top of the list
* Adds a due date
* Applies labels to the card

### Common Trello API endpoints

Here are some commonly used Trello API endpoints you can use with the API Call tool step:

<AccordionGroup>
  <Accordion title="Cards" icon="square">
    * **Get card**: `GET /1/cards/{id}`
    * **Create card**: `POST /1/cards`
    * **Update card**: `PUT /1/cards/{id}`
    * **Delete card**: `DELETE /1/cards/{id}`
    * **Add comment**: `POST /1/cards/{id}/actions/comments`

    [View API Documentation](https://developer.atlassian.com/cloud/trello/rest/api-group-cards/)
  </Accordion>

  <Accordion title="Boards" icon="table-columns">
    * **Get board**: `GET /1/boards/{id}`
    * **Get board lists**: `GET /1/boards/{id}/lists`
    * **Get board cards**: `GET /1/boards/{id}/cards`
    * **Create board**: `POST /1/boards`

    [View API Documentation](https://developer.atlassian.com/cloud/trello/rest/api-group-boards/)
  </Accordion>

  <Accordion title="Lists" icon="list">
    * **Get list**: `GET /1/lists/{id}`
    * **Create list**: `POST /1/lists`
    * **Update list**: `PUT /1/lists/{id}`
    * **Get list cards**: `GET /1/lists/{id}/cards`

    [View API Documentation](https://developer.atlassian.com/cloud/trello/rest/api-group-lists/)
  </Accordion>

  <Accordion title="Checklists" icon="list-check">
    * **Get checklist**: `GET /1/checklists/{id}`
    * **Create checklist**: `POST /1/checklists`
    * **Add checklist item**: `POST /1/checklists/{id}/checkItems`
    * **Update checklist item**: `PUT /1/cards/{id}/checkItem/{idCheckItem}`

    [View API Documentation](https://developer.atlassian.com/cloud/trello/rest/api-group-checklists/)
  </Accordion>

  <Accordion title="Labels" icon="tags">
    * **Get label**: `GET /1/labels/{id}`
    * **Create label**: `POST /1/labels`
    * **Add label to card**: `POST /1/cards/{id}/idLabels`
    * **Remove label from card**: `DELETE /1/cards/{id}/idLabels/{idLabel}`

    [View API Documentation](https://developer.atlassian.com/cloud/trello/rest/api-group-labels/)
  </Accordion>
</AccordionGroup>

<Info>
  You can find Trello's complete API documentation at [https://developer.atlassian.com/cloud/trello/rest/](https://developer.atlassian.com/cloud/trello/rest/).
</Info>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What permissions does the Trello integration require?">
    The integration requires read and write access to your Trello boards, lists, and cards. When you connect your Trello account, you'll be asked to authorize Relevance AI to access your Trello data. You can review the specific permissions during the OAuth authorization process.
  </Accordion>

  <Accordion title="Should I use OAuth or API key + token authentication?">
    **OAuth** is recommended for most users as it's simpler to set up and provides user-level permissions. Use **API key + token** when you need:

    * Server-to-server integrations
    * Persistent access without user interaction
    * Service account functionality
    * More granular control over API access
  </Accordion>

  <Accordion title="Can I restrict my agent to only access certain boards?">
    Yes, when configuring triggers, you can specify which boards to monitor. For tool steps, you control which boards your agent interacts with through the board ID parameters in your workflows. Your agent will only access the boards you explicitly configure.
  </Accordion>

  <Accordion title="Why isn't my trigger firing when I make changes in Trello?">
    If your trigger isn't activating, check the following:

    1. Verify your Trello integration is still connected (re-authenticate if needed)
    2. Confirm the trigger is configured for the correct board
    3. Check that the specific event type matches your trigger (e.g., "Card Moved" vs "Card Updated")
    4. For non-instant triggers, ensure the polling frequency has elapsed

    If issues persist, try disconnecting and reconnecting your Trello integration.
  </Accordion>

  <Accordion title="What's the difference between 'Card Updated' and 'Card Moved' triggers?">
    **Card Updated** triggers whenever any property of a card changes (title, description, due date, attachments, etc.), including when a card moves. **Card Moved** is more specific and only triggers when a card moves from one list to another. Use Card Moved for stage-based workflows; use Card Updated for comprehensive change monitoring.
  </Accordion>

  <Accordion title="Can I use the Trello integration with multiple Trello accounts?">
    Yes, you can connect multiple Trello accounts by adding separate integrations for each account. Each integration will appear with its account name in your integrations list, and you can select the appropriate account when configuring triggers or tool steps.
  </Accordion>

  <Accordion title="Are there rate limits for Trello API calls?">
    Yes, Trello enforces API rate limits (typically 100 requests per 10 seconds per token). Relevance AI manages these limits automatically for pre-built tool steps. If you're using the API Call tool step for high-volume operations, be mindful of rate limiting and implement appropriate delays in your workflows.
  </Accordion>

  <Accordion title="How do I find my board, list, or card IDs for API calls?">
    The easiest way to find IDs is to open the board, list, or card in Trello and add `.json` to the end of the URL. For example, `https://trello.com/b/abc123/board-name.json` will show the board's data including its ID. Alternatively, use the Trello API to list your boards and retrieve IDs programmatically.
  </Accordion>

  <Accordion title="Can I create custom workflows that span multiple Trello boards?">
    Absolutely! You can configure triggers to monitor one board and use tool steps to create or update cards on different boards. This enables powerful cross-board automation, such as syncing tasks between team boards or escalating items to management boards.
  </Accordion>
</AccordionGroup>


# Twilio
Source: https://relevanceai.com/docs/integrations/popular-integrations/twilio

Connect Twilio to send SMS, make voice calls, and automate communications with AI agents

Twilio is a cloud communications platform that provides APIs for SMS, voice calls, video, and other messaging channels. With the Twilio integration in Relevance AI, your agents can send text messages, make and manage phone calls, look up phone numbers, handle media, and run SMS verification flows.

<Note>
  This integration is built by Relevance AI. For integration-specific support, contact [Relevance AI support](https://relevanceai.com/docs/get-started/support). For Twilio platform issues, visit [Twilio Support](https://help.twilio.com/).
</Note>

## Connect the integration

You authenticate with your Twilio Account SID and Auth Token. Both are available in the **Account Info** panel on your [Twilio Console](https://console.twilio.com/) dashboard.

<Steps>
  <Step title="Navigate to Integrations">
    Go to the **Integrations & API Keys** page in your Relevance AI dashboard.
  </Step>

  <Step title="Select Twilio">
    Find and click on **Twilio** from the list of available integrations.
  </Step>

  <Step title="Add integration">
    Click the **Add Integration** button to begin the connection process.
  </Step>

  <Step title="Enter your Account SID">
    Paste your Twilio **Account SID**. In the Twilio Console, this is shown under **Account Info** on the main dashboard (format: `ACxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx`).
  </Step>

  <Step title="Enter your Auth Token">
    Paste your Twilio **Auth Token**. This is shown directly below the Account SID in the **Account Info** panel. Click the eye icon in the Twilio Console to reveal it.
  </Step>

  <Step title="Confirm connection">
    Click **Connect**. Once authenticated, your Twilio account will appear as connected in your integrations list.
  </Step>
</Steps>

<Tip>
  If you manage multiple Twilio projects, you can connect each project separately using its own Account SID and Auth Token. Each connection appears as a distinct entry in your integrations list.
</Tip>

## Setting up triggers

Twilio does not currently support event-based triggers in Relevance AI. To react to inbound SMS or calls, set up a [webhook trigger](/docs/build/agents/build-your-agent/triggers) on your agent and configure your Twilio phone number's webhook URL to point to it.

## Tools & tool steps

The Twilio integration provides 17 operations across seven categories. To find them in the tool step search, type "Twilio" in the search bar.

### SMS operations

<AccordionGroup>
  <Accordion title="Send SMS" icon="message">
    Send an SMS message from a Twilio phone number to any destination number. Specify the `from` number (must be a Twilio number on your account), the `to` number, and the message body. Supports plain text up to 1,600 characters; longer messages are split automatically.
  </Accordion>

  <Accordion title="Send MMS" icon="image">
    Send an MMS message with one or more media attachments. Provide the `from` and `to` numbers, an optional text body, and one or more publicly accessible media URLs. Supported formats include JPEG, PNG, GIF, and MP4.
  </Accordion>

  <Accordion title="Schedule SMS" icon="clock">
    Schedule an SMS to be sent at a future date and time. Specify the message content, the send time in ISO 8601 format, and the messaging service SID to use. Scheduling requires a Twilio Messaging Service with scheduling enabled.
  </Accordion>

  <Accordion title="Cancel scheduled message" icon="ban">
    Cancel a previously scheduled SMS before it is sent. Provide the message SID of the scheduled message. The message must still be in `scheduled` status — messages already sent cannot be recalled.
  </Accordion>
</AccordionGroup>

### Voice call operations

<AccordionGroup>
  <Accordion title="Make call" icon="phone">
    Initiate an outbound phone call from a Twilio number. Provide the `from` and `to` numbers and a TwiML URL or TwiML string that defines how the call behaves (e.g., playing a message, collecting input, or connecting to another number).
  </Accordion>

  <Accordion title="Make TTS call" icon="volume-high">
    Make a voice call and read out a text message using Twilio's text-to-speech engine. Specify the `from` and `to` numbers, the text to speak, and optionally the language and voice. No TwiML knowledge required — the integration handles the TwiML generation.
  </Accordion>

  <Accordion title="Hang up call" icon="phone-slash">
    End an in-progress call by updating its status to `completed`. Provide the call SID. Useful for programmatically terminating calls from within an agent workflow.
  </Accordion>
</AccordionGroup>

### Call & message management

<AccordionGroup>
  <Accordion title="List calls" icon="list">
    Retrieve a list of calls from your Twilio account. Filter by `from` number, `to` number, status, or date range. Returns call metadata including SID, direction, duration, status, and timestamps.
  </Accordion>

  <Accordion title="Get call" icon="circle-info">
    Fetch details about a specific call using its call SID. Returns the call's status, direction, duration, from/to numbers, price, and any associated recordings.
  </Accordion>

  <Accordion title="List messages" icon="messages">
    Retrieve a list of messages from your Twilio account. Filter by `from` number, `to` number, date sent, or message status. Returns message metadata including SID, body, direction, status, and timestamps.
  </Accordion>

  <Accordion title="Get message" icon="envelope-open">
    Fetch details about a specific message using its message SID. Returns the full message body, status, direction, error codes (if any), and pricing information.
  </Accordion>
</AccordionGroup>

### Phone number lookup

<AccordionGroup>
  <Accordion title="Lookup phone number" icon="magnifying-glass">
    Look up carrier and line-type information for any phone number. Returns whether the number is mobile, landline, or VoIP, along with the carrier name and country. Useful for validating numbers before sending SMS or making calls. This operation uses Twilio Lookup and may incur additional Twilio charges.
  </Accordion>

  <Accordion title="Fetch available numbers" icon="hashtag">
    Search for available Twilio phone numbers to purchase in a given country or area code. Filter by capabilities (SMS, voice, MMS), area code, or locality. Returns a list of available numbers with their capabilities and monthly costs.
  </Accordion>
</AccordionGroup>

### Transcription services

<AccordionGroup>
  <Accordion title="Transcribe recording" icon="microphone">
    Submit a Twilio call recording for transcription. Provide the recording SID. Twilio processes the audio and returns the transcription text once complete. Transcription is an asynchronous operation — poll the Get Transcription step or use a webhook to retrieve the result.
  </Accordion>

  <Accordion title="Get transcription" icon="file-lines">
    Retrieve the text and metadata for a completed transcription using its transcription SID. Returns the transcription body, status, duration, and the associated recording SID.
  </Accordion>
</AccordionGroup>

### SMS verification

<AccordionGroup>
  <Accordion title="Send verification code" icon="shield-check">
    Send a one-time passcode (OTP) to a phone number via SMS or voice call using Twilio Verify. Provide the destination number and the channel (`sms` or `call`). Requires a Twilio Verify Service SID configured in your Twilio account.
  </Accordion>

  <Accordion title="Check verification code" icon="check-circle">
    Verify a code entered by the user against the OTP sent by Twilio Verify. Provide the phone number, the code, and the Verify Service SID. Returns `approved` if the code matches and is still within its validity window.
  </Accordion>
</AccordionGroup>

### Media handling

<AccordionGroup>
  <Accordion title="List media" icon="photo-film">
    List all media attachments associated with a specific MMS message. Provide the message SID. Returns media SIDs, content types, and URLs for each attachment.
  </Accordion>

  <Accordion title="Delete media" icon="trash">
    Delete a media attachment from your Twilio account using its media SID and parent message SID. Deleted media is no longer accessible via its Twilio URL. This action is permanent.
  </Accordion>
</AccordionGroup>

## Use the integration's API tool step (advanced)

For operations not covered by the pre-built tool steps, use the **Twilio API Call** tool step to make raw requests to any Twilio REST API endpoint.

<Steps>
  <Step title="Add the tool step">
    In your agent or tool builder, add the **Twilio API Call** tool step.
  </Step>

  <Step title="Select HTTP method">
    Choose the appropriate HTTP method — most Twilio write operations use `POST`, and reads use `GET`.
  </Step>

  <Step title="Enter the endpoint">
    Provide the API path relative to `https://api.twilio.com/2010-04-01/Accounts/{AccountSid}/`, for example `/Messages.json` or `/Calls/{CallSid}.json`.
  </Step>

  <Step title="Add request parameters">
    Provide the request body as form-encoded key-value pairs (Twilio's REST API uses `application/x-www-form-urlencoded`, not JSON).
  </Step>

  <Step title="Test the request">
    Run the tool step to confirm it returns the expected response before using it in a live workflow.
  </Step>
</Steps>

### Example: sending an SMS via API tool step

```json theme={null}
{
  "method": "POST",
  "endpoint": "/Messages.json",
  "body": {
    "From": "+15551234567",
    "To": "+15559876543",
    "Body": "Your appointment is confirmed for tomorrow at 10am."
  }
}
```

For the full endpoint reference, see the [Twilio REST API documentation](https://www.twilio.com/docs/usage/api).

## Example use cases

<CardGroup>
  <Card title="Appointment reminders" icon="calendar-check">
    Send automated SMS reminders to customers before scheduled appointments. Reduce no-shows by triggering reminder messages 24 hours and 1 hour before the appointment time.
  </Card>

  <Card title="Two-factor authentication" icon="shield-halved">
    Add SMS-based two-factor authentication to your workflows using Twilio Verify. Send a one-time code and verify it before granting access or confirming high-value actions.
  </Card>

  <Card title="Customer support automation" icon="headset">
    Automatically respond to inbound SMS inquiries using an AI agent. Route complex issues to human agents while resolving common questions instantly.
  </Card>

  <Card title="Call center automation" icon="phone-volume">
    Trigger outbound voice calls for order confirmations, collections follow-ups, or survey outreach. Use text-to-speech to deliver dynamic messages without pre-recording audio.
  </Card>

  <Card title="Lead qualification by SMS" icon="user-check">
    Qualify inbound leads via SMS conversation. An agent can ask qualifying questions, score responses, and pass qualified leads to your CRM automatically.
  </Card>

  <Card title="Delivery and logistics alerts" icon="truck">
    Send real-time SMS notifications for shipment status updates, delivery confirmations, or exception alerts — triggered automatically as order status changes in your systems.
  </Card>
</CardGroup>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Where do I find my Account SID and Auth Token?" icon="key">
    Both are displayed in the **Account Info** panel on your Twilio Console dashboard at console.twilio.com. Click the eye icon next to the Auth Token to reveal it. Keep your Auth Token secret — anyone with it can make API calls on your behalf.
  </Accordion>

  <Accordion title="Do I need a Twilio phone number to use this integration?" icon="hashtag">
    For sending SMS or making outbound calls, yes — you need at least one Twilio phone number on your account. You can purchase numbers in the Twilio Console under **Phone Numbers** > **Manage** > **Buy a number**. For Twilio Verify (SMS verification), you need a Verify Service instead of a regular phone number.
  </Accordion>

  <Accordion title="Can I connect multiple Twilio accounts?" icon="building">
    Yes. Each Twilio project has its own Account SID and Auth Token. Add a separate integration for each project, and each will appear as a distinct connected account in your integrations list.
  </Accordion>

  <Accordion title="Why is my SMS not being delivered?" icon="triangle-exclamation">
    Common causes include: the destination number is a landline (not SMS-capable), the `from` number does not have SMS capability enabled, your Twilio account balance is insufficient, or the message was filtered as spam. Check the message status and error code using the Get Message tool step, then refer to the [Twilio error code reference](https://www.twilio.com/docs/api/errors) for resolution steps.
  </Accordion>

  <Accordion title="Are there rate limits on Twilio API calls?" icon="gauge-high">
    Twilio enforces rate limits per account and per phone number. The limits vary by Twilio account type and the specific API. If your workflow sends a high volume of messages, consider using a Twilio Messaging Service to distribute load across multiple numbers. Consult [Twilio's rate limit documentation](https://www.twilio.com/docs/usage/api/rate-limits) for specifics.
  </Accordion>

  <Accordion title="Can I receive inbound SMS or calls?" icon="inbox">
    Inbound messages and calls are handled via webhooks. Configure your Twilio phone number's inbound webhook URL to point to a [webhook trigger](/docs/build/agents/build-your-agent/triggers) on your Relevance AI agent. Your agent will then be activated each time a message or call arrives.
  </Accordion>

  <Accordion title="How do I use Twilio Verify for SMS verification?" icon="shield-check">
    First, create a Verify Service in your Twilio Console under **Verify** > **Services**. Note the Service SID. Then use the **Send verification code** tool step with your Verify Service SID to send an OTP, and the **Check verification code** tool step to validate what the user enters.
  </Accordion>
</AccordionGroup>

***

<CardGroup>
  <Card title="Need help?" icon="life-ring" href="https://relevanceai.com/docs/get-started/support">
    Contact our support team for assistance with the Twilio integration.
  </Card>

  <Card title="Explore more integrations" icon="puzzle-piece" href="/integrations/introduction">
    Discover other integrations to enhance your workflows.
  </Card>
</CardGroup>


# Vercel
Source: https://relevanceai.com/docs/integrations/popular-integrations/vercel

Vercel is the platform for frontend developers, providing the speed and reliability innovators need to create at the moment of inspiration.

Vercel is the platform for frontend developers, providing the speed and reliability innovators need to create at the moment of inspiration. With Relevance AI's Vercel integration, you can seamlessly connect your Vercel projects to your AI agents, enabling them to manage deployments, monitor project status, and automate your development workflows, making your deployment processes more efficient and intelligent.

<Warning>
  The Vercel integration in Relevance AI was built by Relevance AI, and is therefore supported by our team, not Vercel. If you have a question or issue with using Vercel in Relevance AI, please [reach out to our support team](/docs/get-started/support). If you have a question or issue that is only about Vercel, you can [reach out to Vercel support](https://vercel.com/support).
</Warning>

## Connect the integration

Connecting your Vercel account to Relevance AI is a straightforward process:

1. Go to the "Integrations" page in the sidebar of your Relevance AI dashboard.
2. Click on "Vercel" from the available integrations.
3. Click on the "Add Integration" button.
4. In the pop-up window, sign into your Vercel account.
5. Grant the necessary permissions for Relevance AI to access your Vercel projects and deployments.
6. Once authenticated, your Vercel account will appear as a connected integration.

## Tool steps for Vercel

The Vercel integration provides a comprehensive set of actions that your agents can use to interact with your deployments and development workflows. These actions can be incorporated into your agent's workflows as tool steps, enabling sophisticated deployment automation capabilities.

### Deployment Management

<CardGroup>
  <Card title="Create Deployment" icon="rocket">
    Create a new deployment for a Vercel project with custom configurations
  </Card>

  <Card title="List Deployments" icon="list">
    Retrieve a list of deployments with filtering options
  </Card>

  <Card title="Cancel Deployment" icon="circle-xmark">
    Cancel an ongoing or queued deployment
  </Card>
</CardGroup>

### Advanced Operations

<CardGroup>
  <Card title="Vercel API Call" icon="code">
    Make authorized requests to any Vercel API endpoint
  </Card>
</CardGroup>

<Tip>
  Type "Vercel" in the tool step search bar to see all available Vercel actions when building your tools.
</Tip>

## Use the integration's API tool step (Advanced)

In addition to the pre-built actions available in the tool directory, you can build custom tools that perform Vercel-specific activities using the Vercel API Call tool step.

### How to use the Vercel API Call tool step

<Steps>
  <Step title="Create a new tool">
    Create a new tool in Relevance AI or open an existing tool you want to add Vercel functionality to.
  </Step>

  <Step title="Add the Vercel API Call tool step">
    1. Scroll down to Tool-steps
    2. Search for "Vercel API Call" in the tool step search bar
    3. Add the Vercel API Call tool step to your workflow
  </Step>

  <Step title="Select your Vercel account">
    Select your connected Vercel account from the dropdown menu.
  </Step>

  <Step title="Configure the API endpoint">
    Configure the API endpoint, method, and parameters according to your needs:

    * **Method**: Select the HTTP method (GET, POST, PUT, DELETE, PATCH)
    * **Path**: Enter the API endpoint path (e.g., `/v13/deployments`)
    * **Body**: Add any required request body data
    * **Headers**: Add any custom headers if needed
  </Step>

  <Step title="Test your API call">
    Test your configuration to ensure it works correctly before deploying.
  </Step>
</Steps>

### Example: Getting Deployment Details

Here's a practical example of using the Vercel API Call tool step to retrieve details about a specific deployment:

**API Endpoint**: `GET /v13/deployments/{id}`

**Configuration**:

```json theme={null}
{
  "method": "GET",
  "path": "/v13/deployments/dpl_ABC123xyz",
  "headers": {}
}
```

This configuration:

* Uses the GET method to retrieve deployment information
* Specifies the deployment ID in the path
* Returns comprehensive details about the deployment including status, creator, and metadata

<Info>
  You can find Vercel's complete API documentation at [https://vercel.com/docs/rest-api](https://vercel.com/docs/rest-api).
</Info>

### Common Vercel API Endpoints

Here are some commonly used Vercel API endpoints you can use with the API Call tool step:

<AccordionGroup>
  <Accordion title="Deployments" icon="rocket">
    * **List deployments**: `GET /v6/deployments`
    * **Get deployment**: `GET /v13/deployments/{id}`
    * **Create deployment**: `POST /v13/deployments`
    * **Cancel deployment**: `PATCH /v12/deployments/{id}/cancel`
    * **Delete deployment**: `DELETE /v13/deployments/{id}`

    [View API Documentation](https://vercel.com/docs/rest-api/endpoints/deployments)
  </Accordion>

  <Accordion title="Projects" icon="folder">
    * **List projects**: `GET /v9/projects`
    * **Get project**: `GET /v9/projects/{id}`
    * **Create project**: `POST /v10/projects`
    * **Update project**: `PATCH /v9/projects/{id}`
    * **Delete project**: `DELETE /v9/projects/{id}`

    [View API Documentation](https://vercel.com/docs/rest-api/endpoints/projects)
  </Accordion>

  <Accordion title="Domains" icon="globe">
    * **List domains**: `GET /v5/domains`
    * **Get domain**: `GET /v5/domains/{domain}`
    * **Add domain**: `POST /v10/projects/{id}/domains`
    * **Remove domain**: `DELETE /v9/projects/{id}/domains/{domain}`

    [View API Documentation](https://vercel.com/docs/rest-api/endpoints/domains)
  </Accordion>

  <Accordion title="Environment Variables" icon="key">
    * **List env variables**: `GET /v9/projects/{id}/env`
    * **Create env variable**: `POST /v10/projects/{id}/env`
    * **Update env variable**: `PATCH /v9/projects/{id}/env/{id}`
    * **Delete env variable**: `DELETE /v9/projects/{id}/env/{id}`

    [View API Documentation](https://vercel.com/docs/rest-api/endpoints/projects#environment-variables)
  </Accordion>

  <Accordion title="Teams" icon="users">
    * **List teams**: `GET /v2/teams`
    * **Get team**: `GET /v2/teams/{id}`
    * **List team members**: `GET /v2/teams/{id}/members`

    [View API Documentation](https://vercel.com/docs/rest-api/endpoints/teams)
  </Accordion>
</AccordionGroup>

## Example use cases

Here are some ways you can leverage the Vercel integration with your agents:

<AccordionGroup>
  <Accordion title="Automated Deployment Pipeline" icon="diagram-project">
    Create an agent that automatically triggers deployments when specific conditions are met, such as successful test runs, code reviews, or scheduled times. The agent can monitor your repository, validate changes, and deploy to the appropriate environment (production, staging, or preview) based on your workflow rules.
  </Accordion>

  <Accordion title="Multi-Environment Deployment Manager" icon="layer-group">
    Build an agent that manages deployments across multiple environments simultaneously. The agent can coordinate staging deployments for testing, automatically promote successful builds to production, and maintain environment-specific configurations and variables.
  </Accordion>

  <Accordion title="Deployment Status Monitor" icon="chart-line">
    Deploy an agent that continuously monitors deployment status, tracks build times, and alerts your team when deployments fail or take longer than expected. The agent can provide detailed reports on deployment health and performance metrics.
  </Accordion>

  <Accordion title="Rollback Automation Agent" icon="clock-rotate-left">
    Create an agent that automatically detects deployment issues through error monitoring or performance degradation and initiates rollbacks to the last stable version. The agent can analyze logs, monitor error rates, and make intelligent decisions about when to rollback.
  </Accordion>

  <Accordion title="Performance-Based Deployment Scheduler" icon="calendar-clock">
    Build an agent that analyzes traffic patterns and schedules deployments during low-traffic periods to minimize user impact. The agent can optimize deployment timing based on historical data and current system load.
  </Accordion>

  <Accordion title="CI/CD Integration Agent" icon="arrows-spin">
    Deploy an agent that bridges your CI/CD pipeline with Vercel deployments, automatically triggering builds when tests pass, managing preview deployments for pull requests, and coordinating multi-stage deployment workflows.
  </Accordion>

  <Accordion title="Deployment Notification System" icon="bell">
    Create an agent that sends customized notifications to different stakeholders (developers, QA, product managers) about deployment events. The agent can provide context-aware updates via Slack, email, or other channels based on deployment status and team preferences.
  </Accordion>

  <Accordion title="Preview Deployment Manager" icon="eye">
    Build an agent that automatically creates and manages preview deployments for feature branches, assigns unique URLs, notifies reviewers, and cleans up old preview deployments to optimize resource usage.
  </Accordion>
</AccordionGroup>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What permissions does the Vercel integration require?">
    The integration requires permissions to manage deployments, access project information, and read deployment logs. You can review the specific permissions during the authentication process. Vercel uses OAuth for secure authentication, ensuring your credentials are never stored directly.
  </Accordion>

  <Accordion title="Can I restrict my Agent to only access certain projects?">
    Yes, you can configure your tools and triggers to only interact with specific projects by setting the appropriate project parameters. Additionally, you can use Vercel's team-based access controls to limit which projects are accessible through the integration.
  </Accordion>

  <Accordion title="How do I authenticate with Vercel for API calls?">
    When you connect your Vercel account through the Integrations page, Relevance AI handles authentication automatically using OAuth. When using the Vercel API Call tool step, simply select your connected Vercel account from the dropdown, and authentication will be handled for you.
  </Accordion>

  <Accordion title="What's the difference between the pre-built Vercel tool steps and the API Call tool step?">
    Pre-built tool steps (like "Create Deployment" or "List Deployments") are designed for specific, common tasks and have simplified interfaces with guided inputs. The Vercel API Call tool step gives you full access to Vercel's REST API, allowing you to implement any functionality available in the API, including advanced or custom operations not covered by pre-built steps.
  </Accordion>

  <Accordion title="Can I use webhooks to trigger my agents based on Vercel events?">
    Yes! You can set up Vercel webhooks to trigger your agents when specific events occur in your projects (like deployment started, deployment completed, or deployment failed). Configure webhooks in your Vercel project settings to point to your agent's webhook URL in Relevance AI.
  </Accordion>

  <Accordion title="Are there rate limits for Vercel API calls?">
    Yes, Vercel enforces rate limits on API calls. The specific limits depend on your Vercel plan. Your agents should be designed to handle rate limiting gracefully. You can check your current rate limit status in the API response headers.
  </Accordion>

  <Accordion title="Can I manage environment variables through the integration?">
    Yes! You can use the Vercel API Call tool step to create, update, and delete environment variables for your projects. This allows your agents to dynamically manage configuration across different environments.
  </Accordion>

  <Accordion title="How do I cancel a deployment that's already in progress?">
    Use the "Cancel Deployment" tool step or the Vercel API Call with the `PATCH /v12/deployments/{id}/cancel` endpoint. You'll need the deployment ID, which you can obtain from the "List Deployments" tool step or from deployment creation responses.
  </Accordion>

  <Accordion title="Can I deploy to specific branches or environments?">
    Yes! When creating a deployment, you can specify the branch, environment (production, preview, or development), and other configuration options. This allows your agents to manage multi-environment deployment strategies effectively.
  </Accordion>
</AccordionGroup>


# WhatsApp
Source: https://relevanceai.com/docs/integrations/popular-integrations/whatsapp

Connect your WhatsApp account to Relevance AI and enable your agents to communicate directly with customers through one of the world's most popular messaging platforms.

<Warning>
  Our WhatsApp integration provider is shipping a change in mid-March 2026 that requires action from existing users. If you currently use the WhatsApp integration, you will need to reconnect your account once the change takes effect. Additionally, previously threaded conversations will start on a new thread due to a change in identifier format. Affected users have been emailed with further details.
</Warning>

## Connect the integration

Connecting your WhatsApp account to Relevance AI is a straightforward process:

1. Go to the "Integrations & API Keys" page in the sidebar.
2. Click on "WhatsApp".
3. Click on the "Add Integration" button.
4. In the pop-up window, sign into your WhatsApp Business account.
5. Follow the authorization steps to grant Relevance AI access to your WhatsApp account.
6. Once connected, you'll see your WhatsApp account listed under active integrations.

## Setting up triggers

WhatsApp Integration can be configured as a trigger for your AI agents, allowing them to automatically respond when messages are received through WhatsApp.

<Warning>The WhatsApp Trigger is a Premium trigger, available on the Pro plan and above.</Warning>

<div>
  <iframe title="WhatsApp Trigger Setup" />
</div>

To set up WhatsApp as a trigger:

1. Create a new agent or edit an existing one.
2. Navigate to the "Triggers" section in "Agent Profile".
3. Select "WhatsApp" from the available triggers.
4. Choose the WhatsApp account you want to connect to this agent.
5. Configure trigger conditions (optional):
   * Respond to all incoming messages
   * Respond only to specific contacts or groups
   * Respond based on message content filters
6. Save your trigger configuration.

Once set up, your agent will automatically activate when new WhatsApp messages that match your trigger conditions are received, enabling real-time customer engagement.

## Tools & Tool Steps

WhatsApp integration provides powerful tools and actions that your agents can use to engage with customers. These actions can be incorporated into your agent's workflow at any point, not just as triggers.

Here are some of the key WhatsApp actions available:

### Send WhatsApp Message

Send text messages to individual contacts or groups.

Input parameters:

* Recipient phone number or group ID
* Message content
* Optional: Attachments

### Send WhatsApp Media

Share images, documents, or other media files with your contacts.

Input parameters:

* Recipient phone number or group ID
* Media type (image, document, audio, video)
* Media URL or file
* Optional: Caption

### Send WhatsApp Template Message

Use pre-approved message templates for business communications.

Input parameters:

* Recipient phone number
* Template name
* Template parameters

### Get WhatsApp Conversation History

Retrieve previous messages from a specific conversation.

Input parameters:

* Contact phone number
* Optional: Time range

### Mark WhatsApp Message as Read

Update the read status of messages.

Input parameters:

* Message ID

These are just a few examples of the many WhatsApp actions available. Your agents can leverage these tools to create sophisticated customer engagement workflows, from automated support responses to personalized marketing campaigns.

## Use the integration's API tool step (advanced)

In addition to the tools and actions available in the directory, you can build custom tools that perform WhatsApp activities by using the WhatsApp API Call tool step:

1. Create a new tool.
2. Scroll down to Tool-steps.
3. Add WhatsApp API tool-step.
4. Select your connected WhatsApp account in the dropdown.

This advanced approach gives you direct access to the WhatsApp Business API, allowing for more complex and customized interactions:

### Example API call:

```json theme={null}
{
  "endpoint": "/messages",
  "method": "POST",
  "body": {
    "to": "{{recipient_number}}",
    "type": "text",
    "text": {
      "body": "{{message_content}}"
    }
  }
}
```

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What types of WhatsApp accounts are supported?">Relevance AI supports both WhatsApp Personal and Business accounts. To learn more about the Business accounts integration, head [here](/docs/integrations/popular-integrations/whatsapp-for-business).</Accordion>
  <Accordion title="Are there message volume limitations?">Message volumes are subject to WhatsApp's own policies and rate limits. The Relevance AI platform itself doesn't impose additional limits beyond your plan's credit allocation.</Accordion>
  <Accordion title="Can I use WhatsApp for bulk messaging?">Yes, but you must comply with WhatsApp's business policies regarding bulk messaging and ensure you have proper opt-ins from recipients.</Accordion>
  <Accordion title="How secure are WhatsApp conversations through Relevance AI?">All WhatsApp communications maintain end-to-end encryption. Relevance AI adheres to strict security protocols to ensure your conversation data remains protected.</Accordion>
  <Accordion title="Can multiple agents use the same WhatsApp account?">Yes, you can connect multiple agents to the same WhatsApp account, but you should configure triggers carefully to avoid conflicts.</Accordion>
  <Accordion title="I received an email about a change to the WhatsApp integration. What do I need to do?">Our WhatsApp integration provider is shipping a change in mid-March 2026. If you're an existing user, you'll need to reconnect your WhatsApp account once the change takes effect. Please also note that previously threaded conversations will start on a new thread, as the underlying identifier format is changing. No action is needed until the change goes live — we'll follow up with further instructions.</Accordion>
</AccordionGroup>


# Whatsapp for Business
Source: https://relevanceai.com/docs/integrations/popular-integrations/whatsapp-for-business

Interact with your WhatsApp Business account with Relevance.

Connect your AI agents with WhatsApp Business to engage with customers directly through one of the world's most popular messaging platforms. The WhatsApp Business integration enables your agents to handle customer inquiries, provide support, and deliver personalized experiences through WhatsApp's familiar interface.

## Connect the integration

Setting up the WhatsApp Business integration with Relevance AI is straightforward:

1. Navigate to the "Integrations & API Keys" page in the sidebar of your Relevance AI dashboard
2. Locate and click on "WhatsApp Business"
3. Click the "Add Integration" button
4. Follow the authentication flow to connect your WhatsApp Business account
5. Once connected, you'll see your WhatsApp Business account listed under active integrations

## Setting up triggers

The WhatsApp Business integration allows you to create powerful triggers that automatically activate your AI agents when specific events occur. This enables seamless customer engagement without manual intervention.

<Warning>The WhatsApp for Business Trigger is a Premium trigger, available on the Pro plan and above.</Warning>

To set up a WhatsApp trigger:

1. Go to your agent's profile page
2. Under "Triggers", select "WhatsApp Business"
3. Configure the trigger settings:
   * **Trigger on new messages**: Activate your agent when a customer sends a message to your WhatsApp Business number
   * **Trigger on specific keywords**: Set your agent to respond only when certain keywords or phrases are detected
   * **Trigger based on customer attributes**: Configure triggers based on customer data or conversation context

Once configured, your agent will automatically engage with customers through WhatsApp when the trigger conditions are met.

## Tools & Tool Steps

The WhatsApp Business integration provides a rich set of actions that your agents can use to interact with customers. These actions can be incorporated into your agent's workflow to create sophisticated conversation flows and deliver exceptional customer experiences.

Here are some of the key actions available:

### Send Message

Send text messages to customers through WhatsApp. This action supports rich text formatting, allowing your agent to send well-structured, easy-to-read messages.

### Send Media

Share images, documents, videos, or audio files with customers. This is perfect for sending product catalogs, instructional materials, or visual content to enhance the conversation.

### Send Template Message

Utilize pre-approved message templates for common scenarios like welcome messages, order confirmations, appointment reminders, and more. Templates ensure compliance with WhatsApp's policies while maintaining consistent communication.

### Send Interactive Messages

Create interactive experiences with buttons, list messages, and reply buttons that guide customers through specific flows or help them make selections without typing lengthy responses.

### Get Customer Profile

Retrieve customer profile information to personalize interactions based on available data, such as name, profile picture, or previous interaction history.

### Mark Messages as Read

Update message status to "read" to provide customers with confirmation that their messages have been received and processed.

### Create Groups

Create WhatsApp groups for specific purposes, such as product launches, support communities, or team discussions.

### Manage Contacts

Add, update, or retrieve contact information to maintain an accurate customer database.

These are just a few examples of the many actions available through the WhatsApp Business integration. The platform offers numerous additional capabilities to support complex business workflows and customer engagement strategies.

## Use the integration's API tool step (advanced)

For advanced users who need more customized functionality, Relevance AI provides direct access to the WhatsApp Business API through a dedicated API tool step:

1. Create a new tool in your Relevance AI dashboard
2. Scroll down to Tool-steps
3. Add the WhatsApp Business API tool-step
4. Select your connected WhatsApp Business account from the dropdown

With the API tool step, you can build custom tools that leverage the full power of the WhatsApp Business API, including:

### Custom Message Formatting

Create highly customized message formats with advanced formatting options not available through standard actions.

### Batch Operations

Send messages to multiple recipients simultaneously or perform bulk operations on conversations.

### Advanced Media Handling

Implement sophisticated media processing workflows, such as image analysis or document processing before sending responses.

### Integration with External Systems

Connect WhatsApp conversations with external databases, CRMs, or business systems to create unified customer experiences.

### Custom Analytics

Build specialized analytics tools to track conversation metrics, customer engagement patterns, or agent performance.

The API tool step gives you complete flexibility to extend the capabilities of the WhatsApp Business integration to meet your specific business requirements.

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What are the requirements for using the WhatsApp Business integration?">You need a WhatsApp Business account and must comply with WhatsApp's Business Policy. Your account must be verified and in good standing.</Accordion>
  <Accordion title="Can I use the integration with multiple WhatsApp Business numbers?">Yes, you can connect multiple WhatsApp Business numbers to Relevance AI and configure different agents to work with specific numbers.</Accordion>
  <Accordion title="Is the integration compliant with WhatsApp's messaging policies?">Yes, the integration is designed to comply with WhatsApp's Business Messaging Policy. However, it's your responsibility to ensure that your messaging practices adhere to these policies and local regulations.</Accordion>
</AccordionGroup>


# WorldNewsAPI
Source: https://relevanceai.com/docs/integrations/popular-integrations/worldnewsapi

Access thousands of news sources in 86+ languages from 210+ countries with semantic search, sentiment analysis, and real-time updates

WorldNewsAPI provides access to thousands of news sources across 210+ countries in 86+ languages, with 170,000+ news articles added daily in real-time. The platform offers semantic tagging, sentiment analysis, author extraction, and structured data extraction including images, videos, titles, descriptions, and publication dates.

With the WorldNewsAPI integration in Relevance AI, you can search news with advanced filters, extract structured data from news URLs, and convert location names to coordinates for location-based news searches—all directly within your AI workflows.

<Warning>
  The WorldNewsAPI integration in Relevance AI was built by Relevance AI, and is therefore supported by our team, not WorldNewsAPI. If you have a question or issue with using WorldNewsAPI in Relevance AI, please [reach out to our support team](/docs/get-started/support). If you have a question or issue that is only about WorldNewsAPI, you can [reach out to WorldNewsAPI support](https://worldnewsapi.com/).
</Warning>

## About WorldNewsAPI

WorldNewsAPI is a comprehensive news data platform that provides:

* **Global Coverage**: Access to thousands of news sources from 210+ countries
* **Multi-language Support**: Content in 86+ languages
* **Real-time Updates**: 170,000+ new articles added daily
* **Advanced Features**: Semantic tagging, sentiment analysis, author extraction
* **Rich Media**: Automatic extraction of images, videos, titles, descriptions, and dates
* **Powerful Search**: Semantic search with filters for country, language, time, sentiment, entities, sources, and authors

<Note>
  This is a native integration that replaces the previous Pipedream version, offering improved reliability and performance.
</Note>

## Connect the integration

Connecting your WorldNewsAPI account to Relevance AI requires an API key:

1. Go to the "Integrations" page in the sidebar of your Relevance AI dashboard.
2. Click on "WorldNewsAPI" from the available integrations.
3. Click on the "Add Integration" button.
4. Enter your WorldNewsAPI API key when prompted.
5. Once authenticated, your WorldNewsAPI account will appear as a connected integration.

<Tip>
  To get your WorldNewsAPI API key:

  1. Sign up at [https://worldnewsapi.com/](https://worldnewsapi.com/)
  2. Confirm your email address
  3. Go to your Profile page
  4. Click "Show/Hide API Key" to reveal your key
</Tip>

## Tool steps for WorldNewsAPI

The WorldNewsAPI integration provides ready-made actions that your agents can use to search news, extract article data, and work with location-based searches. These actions can be incorporated into your agent's workflows as tool steps.

### News Search & Discovery

<CardGroup>
  <Card title="Search News" icon="magnifying-glass">
    Search thousands of news sources with filters for 50+ languages, 150+ countries, date ranges, sentiment, entities, sources, and authors
  </Card>

  <Card title="Extract News from URL" icon="file-lines">
    Extract structured data from any news article URL including title, text, images, videos, publish date, authors, language, source country, and sentiment
  </Card>
</CardGroup>

### Location Services

<CardGroup>
  <Card title="Get Geo Coordinates" icon="location-dot">
    Convert location names to latitude and longitude coordinates for location-based news searches
  </Card>
</CardGroup>

### Advanced API Access

<CardGroup>
  <Card title="WorldNewsAPI API Call" icon="code">
    Make custom API calls to access any WorldNewsAPI endpoint for advanced use cases
  </Card>
</CardGroup>

<Tip>
  Type "WorldNewsAPI" in the tool step search bar to see all available actions when building your tools.
</Tip>

## Use the integration's API tool step (Advanced)

In addition to the pre-built actions available in the tool directory, you can build custom tools that perform WorldNewsAPI-specific activities using the WorldNewsAPI API Call tool step.

### How to use the API Call tool step

<Steps>
  <Step title="Create a new tool">
    Create a new tool in Relevance AI or open an existing tool you want to add WorldNewsAPI functionality to.
  </Step>

  <Step title="Add the WorldNewsAPI API Call tool step">
    1. Scroll down to Tool-steps
    2. Search for "WorldNewsAPI API Call" in the tool step search bar
    3. Add the WorldNewsAPI API Call tool step to your workflow
  </Step>

  <Step title="Select your WorldNewsAPI account">
    Select your connected WorldNewsAPI account from the dropdown menu.
  </Step>

  <Step title="Configure the API endpoint">
    Configure the API endpoint, method, and parameters according to your needs:

    * **Method**: Select the HTTP method (GET, POST)
    * **Endpoint**: Enter the API endpoint path (e.g., `/search-news`)
    * **Parameters**: Add query parameters for filtering and search criteria
  </Step>

  <Step title="Test your API call">
    Test your configuration to ensure it works correctly before deploying.
  </Step>
</Steps>

### Example: Search News with Custom Filters

Here's a practical example of using the WorldNewsAPI API Call tool step to search for technology news with sentiment filtering:

**API Endpoint**: `GET /search-news`

**Configuration**:

```json theme={null}
{
  "method": "GET",
  "endpoint": "/search-news",
  "params": {
    "text": "artificial intelligence",
    "language": "en",
    "sentiment": "positive",
    "earliest-publish-date": "2026-03-01",
    "number": 10
  }
}
```

This configuration:

* Searches for articles about "artificial intelligence"
* Filters for English language content
* Returns only positive sentiment articles
* Limits results to articles published after March 1, 2026
* Returns up to 10 results

<Info>
  You can find WorldNewsAPI's complete API documentation at [https://worldnewsapi.com/docs/](https://worldnewsapi.com/docs/).
</Info>

### Common API endpoints

<AccordionGroup>
  <Accordion title="Search News" icon="magnifying-glass">
    **Endpoint**: `GET /search-news`

    Search news articles with advanced filtering options:

    * **text**: Search query text
    * **language**: Filter by language code (e.g., "en", "es", "fr")
    * **source-countries**: Filter by country codes
    * **sentiment**: Filter by sentiment (positive, negative, neutral)
    * **entities**: Filter by named entities
    * **earliest-publish-date**: Start date for results
    * **latest-publish-date**: End date for results
    * **number**: Number of results to return

    Returns: Array of news articles with full metadata

    [View API Documentation](https://worldnewsapi.com/docs/)
  </Accordion>

  <Accordion title="Extract News" icon="file-lines">
    **Endpoint**: `GET /extract-news`

    Extract structured data from a news article URL:

    * **url**: The news article URL to extract from
    * **analyze**: Whether to include sentiment analysis

    Returns: Title, text, images, videos, publish date, authors, language, source country, sentiment

    [View API Documentation](https://worldnewsapi.com/docs/)
  </Accordion>

  <Accordion title="Geo Coordinates" icon="location-dot">
    **Endpoint**: `GET /geo-coordinates`

    Convert location names to coordinates:

    * **location**: Location name to geocode

    Returns: Latitude and longitude coordinates

    Useful for location-based news searches and mapping.

    [View API Documentation](https://worldnewsapi.com/docs/)
  </Accordion>

  <Accordion title="Top News" icon="star">
    **Endpoint**: `GET /top-news`

    Get top news headlines:

    * **source-country**: Country code for news sources
    * **language**: Language code
    * **date**: Date for top news

    Returns: Top headlines for the specified criteria

    [View API Documentation](https://worldnewsapi.com/docs/)
  </Accordion>
</AccordionGroup>

## Example use cases

Here are some ways you can leverage the WorldNewsAPI integration with your agents:

<AccordionGroup>
  <Accordion title="Brand Monitoring Agent" icon="chart-line">
    Create an agent that monitors news mentions of your brand, products, or key executives. The agent searches for relevant articles, analyzes sentiment, and sends alerts when negative coverage is detected or when your brand appears in major publications.
  </Accordion>

  <Accordion title="Competitive Intelligence Bot" icon="chess">
    Build an agent that tracks competitor mentions across global news sources. The agent monitors product launches, executive changes, funding announcements, and market moves, then compiles daily intelligence briefings for your team.
  </Accordion>

  <Accordion title="Market Research Analyzer" icon="magnifying-glass-chart">
    Deploy an agent that analyzes news trends in your industry. The agent tracks emerging topics, sentiment shifts, and geographic patterns, providing insights into market dynamics and consumer sentiment.
  </Accordion>

  <Accordion title="Content Aggregation System" icon="newspaper">
    Create an agent that builds curated news feeds for specific topics, industries, or regions. The agent filters by language, sentiment, and relevance, then formats articles for your website, newsletter, or internal dashboard.
  </Accordion>

  <Accordion title="Crisis Monitoring System" icon="triangle-exclamation">
    Build an agent that monitors for breaking news and crisis situations relevant to your organization. The agent searches for keywords related to potential risks, analyzes sentiment and urgency, and sends real-time alerts to stakeholders.
  </Accordion>

  <Accordion title="Multi-language News Digest" icon="globe">
    Deploy an agent that aggregates news from multiple languages and regions. The agent searches news in different languages, translates key articles, and creates a comprehensive international news digest for global teams.
  </Accordion>

  <Accordion title="Sentiment Analysis Dashboard" icon="face-smile">
    Create an agent that tracks sentiment trends for specific topics over time. The agent collects articles, analyzes sentiment patterns, and generates reports showing how public perception is evolving.
  </Accordion>

  <Accordion title="Location-based News Alerts" icon="location-dot">
    Build an agent that monitors news for specific geographic locations. The agent uses geo-coordinates to find local news, tracks regional events, and alerts relevant teams about location-specific developments.
  </Accordion>
</AccordionGroup>

## Best practices

<AccordionGroup>
  <Accordion title="API Rate Limits" icon="gauge">
    **Manage your API usage effectively:**

    * Rate limits depend on your WorldNewsAPI plan
    * Implement retry logic with exponential backoff for rate limit errors
    * Cache search results when appropriate to reduce API calls
    * Use specific filters to reduce result set sizes
    * Monitor your API usage through the WorldNewsAPI dashboard
    * Consider upgrading your plan if you consistently hit limits
  </Accordion>

  <Accordion title="Search Optimization" icon="magnifying-glass">
    **Get better search results:**

    * Use specific search terms rather than broad queries
    * Combine multiple filters (language, country, date) to narrow results
    * Use sentiment filters to focus on relevant coverage
    * Specify date ranges to get timely results
    * Use entity filters to find articles mentioning specific people or organizations
    * Test different search queries to find the most effective terms
  </Accordion>

  <Accordion title="Data Quality" icon="star">
    **Ensure high-quality data:**

    * Validate extracted article data before using it
    * Handle cases where extraction fails or returns incomplete data
    * Check sentiment scores for reliability
    * Verify publication dates are within expected ranges
    * Filter out duplicate articles from different sources
    * Implement fallback logic for missing data fields
  </Accordion>

  <Accordion title="Performance" icon="bolt">
    **Optimize workflow performance:**

    * Limit the number of results per search to what you actually need
    * Use pagination for large result sets
    * Process articles in batches rather than one at a time
    * Cache frequently accessed data
    * Schedule regular searches rather than real-time queries when possible
    * Use webhooks or scheduled runs instead of continuous polling
  </Accordion>
</AccordionGroup>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What languages are supported?">
    WorldNewsAPI supports 86+ languages including English, Spanish, French, German, Chinese, Japanese, Arabic, Portuguese, Russian, and many more. You can filter search results by language code (e.g., "en" for English, "es" for Spanish).
  </Accordion>

  <Accordion title="How many news sources are available?">
    WorldNewsAPI provides access to thousands of news sources across 210+ countries. The exact number varies as sources are continuously added and updated. Coverage includes major international publications, regional news outlets, and specialized industry sources.
  </Accordion>

  <Accordion title="What's the difference between the native integration and the Pipedream version?">
    The native WorldNewsAPI integration is built directly into Relevance AI, offering better reliability, performance, and support. The Pipedream version (now renamed "World News API (Pipedream)") is being deprecated. We recommend using the native integration for all new workflows.
  </Accordion>

  <Accordion title="How do I get a WorldNewsAPI API key?">
    To get your API key:

    1. Visit [https://worldnewsapi.com/](https://worldnewsapi.com/)
    2. Sign up for an account
    3. Confirm your email address
    4. Log in and go to your Profile page
    5. Click "Show/Hide API Key" to reveal your key

    Keep your API key secure and never share it publicly.
  </Accordion>

  <Accordion title="Are there rate limits?">
    Yes, rate limits depend on your WorldNewsAPI subscription plan. Free plans have lower limits, while paid plans offer higher request volumes. Check your WorldNewsAPI account dashboard for your specific limits. If you hit rate limits, implement retry logic in your workflows.
  </Accordion>

  <Accordion title="How accurate is the sentiment analysis?">
    WorldNewsAPI uses advanced natural language processing for sentiment analysis. While generally accurate, sentiment detection can vary based on article complexity, language, and context. We recommend reviewing sentiment scores and implementing confidence thresholds for critical use cases.
  </Accordion>

  <Accordion title="Can I search for news in multiple languages simultaneously?">
    Yes, you can search across multiple languages by making separate API calls for each language or by not specifying a language filter (which returns results in all languages). You can then filter or process the results based on your needs.
  </Accordion>

  <Accordion title="How real-time is the news data?">
    WorldNewsAPI adds 170,000+ articles daily in real-time. New articles typically appear in search results within minutes of publication. However, the exact timing depends on when the source publishes and how quickly it's indexed.
  </Accordion>

  <Accordion title="Can I extract data from any news URL?">
    The Extract News tool works with most news article URLs. However, some sites may block automated extraction or have paywalls. The tool extracts publicly accessible content and structured data like titles, images, videos, authors, and publication dates.
  </Accordion>

  <Accordion title="What are geo coordinates used for?">
    Geo coordinates (latitude and longitude) allow you to search for news from specific geographic locations. This is useful for finding local news, tracking regional events, or monitoring news from areas relevant to your business operations.
  </Accordion>

  <Accordion title="Can I filter by specific news sources?">
    Yes, you can filter search results by specific news sources or domains using the source filters in your search queries. This is useful when you want to monitor specific publications or exclude certain sources.
  </Accordion>

  <Accordion title="How far back does the news archive go?">
    WorldNewsAPI maintains an extensive archive of historical news articles. The exact coverage period varies by source, but many sources have archives going back several years. Use the date range filters to search historical content.
  </Accordion>
</AccordionGroup>


# Zendesk
Source: https://relevanceai.com/docs/integrations/popular-integrations/zendesk

Connect Zendesk to automate customer support workflows and manage tickets with AI agents

Zendesk is a leading customer service platform that helps businesses manage support tickets, customer communications, and service operations. With the Zendesk integration in Relevance AI, you can automate ticket management, streamline support workflows, and enhance customer service operations using AI agents. The integration connects via OAuth, so tokens refresh automatically and there is no manual credential management required.

<Note>
  This integration is built by Relevance AI. For integration-specific support, contact [Relevance AI support](https://relevanceai.com/docs/get-started/support). For Zendesk platform issues, visit [Zendesk Support](https://support.zendesk.com/).
</Note>

## Connect the Integration

Follow these steps to connect your Zendesk account to Relevance AI:

<Steps>
  <Step title="Navigate to Integrations">
    Go to the **Integrations & API Keys** page in your Relevance AI dashboard.
  </Step>

  <Step title="Select Zendesk">
    Find and click on **Zendesk** from the list of available integrations.
  </Step>

  <Step title="Add Integration">
    Click the **Add Integration** button to begin the connection process.
  </Step>

  <Step title="Enter your Zendesk subdomain">
    Provide your Zendesk subdomain — the part before `.zendesk.com` in your Zendesk URL (e.g., if your Zendesk URL is `yourcompany.zendesk.com`, enter `yourcompany`).
  </Step>

  <Step title="Authorize via OAuth">
    Click **Authorize** to be redirected to Zendesk. Sign in if prompted, then click **Allow** to grant Relevance AI access to your Zendesk account.
  </Step>

  <Step title="Confirm connection">
    You'll be redirected back to Relevance AI. A confirmation message will indicate that your Zendesk account is successfully connected. Tokens refresh automatically — no manual management needed.
  </Step>
</Steps>

## Setting Up Triggers

<Warning>
  **Enterprise Plan Required for Triggers** - Zendesk triggers are only available on Enterprise plans. Tool steps (actions) are available on all plans. [Contact sales](https://relevanceai.com/book-a-demo) to learn more.
</Warning>

Zendesk triggers allow you to automatically activate agents when specific events occur in your Zendesk account, such as:

* New ticket created
* Ticket status changed
* Ticket priority updated
* New comment added to ticket
* Ticket assigned to agent

To set up a Zendesk trigger (Enterprise plans only):

<Steps>
  <Step title="Navigate to Agent Settings">
    Open your agent and go to the **Triggers** tab in the agent settings.
  </Step>

  <Step title="Select Zendesk Trigger">
    Choose **Zendesk** from the list of available trigger integrations.
  </Step>

  <Step title="Configure Trigger Event">
    Select the specific Zendesk event that should activate your agent (e.g., "New Ticket Created").
  </Step>

  <Step title="Map Variables">
    Map Zendesk ticket data (subject, description, requester, priority, etc.) to your agent's input variables.
  </Step>

  <Step title="Test and Activate">
    Test your trigger to ensure it works correctly, then activate it to start automating your support workflows.
  </Step>
</Steps>

## Tool Steps for Zendesk

Zendesk tool steps allow you to incorporate Zendesk actions into your agent workflows. These actions are available on all plans and can be used to automate ticket management and customer support operations.

### Ticket Management

<AccordionGroup>
  <Accordion title="Create Ticket" icon="ticket">
    Create a new support ticket in Zendesk with specified details including subject, description, priority, type, and assignee.
  </Accordion>

  <Accordion title="Get Ticket Info" icon="circle-info">
    Retrieve detailed information about a specific ticket using its ticket ID, including status, priority, assignee, requester, and all comments.
  </Accordion>

  <Accordion title="List Tickets" icon="list">
    Retrieve a list of tickets from your Zendesk account with optional filtering by status, priority, assignee, or date range.
  </Accordion>

  <Accordion title="Search Tickets" icon="magnifying-glass">
    Search for tickets using Zendesk's search query language to find tickets matching specific criteria.
  </Accordion>

  <Accordion title="Update Ticket" icon="pen-to-square">
    Update an existing ticket's properties such as status, priority, assignee, tags, or add internal notes and public comments.
  </Accordion>

  <Accordion title="Delete Ticket" icon="trash">
    Delete a ticket from your Zendesk account. Note that this permanently removes the ticket.
  </Accordion>
</AccordionGroup>

<Tip>
  To find these tool steps when building your agent or tool, search for "Zendesk" in the tool step search bar.
</Tip>

## Use the Integration's API Tool Step (Advanced)

For advanced use cases, you can use the **Zendesk API Call** tool step to make custom API requests to any Zendesk endpoint. This gives you access to the full Zendesk API beyond the pre-built tool steps.

### How to use the Zendesk API Call tool step

<Steps>
  <Step title="Add the Tool Step">
    In your agent or tool builder, add the **Zendesk API Call** tool step.
  </Step>

  <Step title="Select HTTP Method">
    Choose the appropriate HTTP method (GET, POST, PUT, PATCH, DELETE).
  </Step>

  <Step title="Enter the Endpoint">
    Provide the API endpoint path (e.g., `/api/v2/users.json` or `/api/v2/tickets/12345.json`).
  </Step>

  <Step title="Add Request Body">
    For POST, PUT, or PATCH requests, provide the request body in JSON format.
  </Step>

  <Step title="Configure Response Handling">
    Map the API response to variables in your workflow for further processing.
  </Step>

  <Step title="Test the Request">
    Test your API call to ensure it returns the expected data.
  </Step>
</Steps>

### Example: Adding Tags to a Ticket

```json theme={null}
{
  "method": "PUT",
  "endpoint": "/api/v2/tickets/12345.json",
  "body": {
    "ticket": {
      "tags": ["urgent", "vip", "billing"]
    }
  }
}
```

### Common Zendesk API Endpoints

**Ticket Operations**

* `GET /api/v2/tickets.json` - List all tickets
* `GET /api/v2/tickets/{id}.json` - Get ticket details
* `POST /api/v2/tickets.json` - Create a ticket
* `PUT /api/v2/tickets/{id}.json` - Update a ticket
* `DELETE /api/v2/tickets/{id}.json` - Delete a ticket

**User Management**

* `GET /api/v2/users.json` - List users
* `GET /api/v2/users/{id}.json` - Get user details
* `POST /api/v2/users.json` - Create a user
* `PUT /api/v2/users/{id}.json` - Update a user

**Organization Management**

* `GET /api/v2/organizations.json` - List organizations
* `GET /api/v2/organizations/{id}.json` - Get organization details

**Ticket Comments**

* `GET /api/v2/tickets/{id}/comments.json` - List ticket comments
* `POST /api/v2/tickets/{id}/comments.json` - Add a comment

**Search**

* `GET /api/v2/search.json?query={query}` - Search across Zendesk

For complete API documentation, visit the [Zendesk API Reference](https://developer.zendesk.com/api-reference/).

## Example Use Cases

<CardGroup>
  <Card title="Automated Ticket Triage" icon="filter">
    Automatically categorize and prioritize incoming support tickets based on content analysis, customer history, and urgency indicators. Route tickets to the appropriate team or agent.
  </Card>

  <Card title="Intelligent Response Suggestions" icon="message-bot">
    Analyze ticket content and customer history to generate suggested responses for support agents, reducing response time and improving consistency.
  </Card>

  <Card title="Customer Sentiment Analysis" icon="face-smile">
    Monitor ticket sentiment in real-time and automatically escalate tickets with negative sentiment or frustrated customers to senior support staff.
  </Card>

  <Card title="SLA Monitoring & Alerts" icon="clock">
    Track ticket response and resolution times, automatically escalate tickets approaching SLA breaches, and notify managers of potential violations.
  </Card>

  <Card title="Knowledge Base Automation" icon="book">
    Automatically suggest relevant knowledge base articles to customers based on their ticket content, or create new articles from frequently asked questions.
  </Card>

  <Card title="Multi-Channel Support Sync" icon="arrows-rotate">
    Synchronize support requests from multiple channels (email, chat, social media) into Zendesk tickets and maintain consistent customer communication.
  </Card>

  <Card title="Customer Data Enrichment" icon="database">
    Automatically enrich ticket data with information from CRM systems, previous interactions, purchase history, and customer profiles for better context.
  </Card>

  <Card title="Automated Follow-ups" icon="paper-plane">
    Schedule and send automated follow-up messages to customers based on ticket status, resolution time, or customer satisfaction scores.
  </Card>
</CardGroup>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="What permissions are required for the Zendesk integration?" icon="lock">
    The integration uses OAuth, so you authorize access through your Zendesk account directly. For full functionality, the Zendesk account you authorize with should have Admin or Agent permissions.
  </Accordion>

  <Accordion title="Can I access tickets from multiple Zendesk accounts?" icon="building">
    Yes, you can connect multiple Zendesk accounts by adding separate integrations for each account. Each integration will appear with its account name in your integrations list.
  </Accordion>

  <Accordion title="How do I troubleshoot authentication errors?" icon="triangle-exclamation">
    If you encounter authentication errors, check the following: ensure your Zendesk subdomain is entered correctly (just the part before `.zendesk.com`), verify that the Zendesk account you authorized has Admin or Agent permissions, and try disconnecting and reconnecting the integration to trigger a fresh OAuth flow. If issues persist, contact [Relevance AI support](/docs/get-started/support).
  </Accordion>

  <Accordion title="Can I use custom ticket fields?" icon="sliders">
    Yes! When using the Zendesk API Call tool step, you can access and update custom ticket fields by including them in your API requests. Reference your custom field IDs from your Zendesk Admin Center.
  </Accordion>

  <Accordion title="Are there rate limits for Zendesk API calls?" icon="gauge-high">
    Yes, Zendesk enforces API rate limits based on your Zendesk plan. The integration respects these limits automatically. If you encounter rate limit errors, consider implementing delays between API calls or upgrading your Zendesk plan.
  </Accordion>

  <Accordion title="Can I automate ticket creation from other sources?" icon="wand-magic-sparkles">
    Absolutely! You can use Relevance AI agents to monitor various sources (email, forms, chat, social media) and automatically create Zendesk tickets using the Create Ticket tool step. This enables centralized support management across all channels.
  </Accordion>

  <Accordion title="How do I handle ticket attachments?" icon="paperclip">
    Ticket attachments can be accessed and managed through the Zendesk API Call tool step. Use the appropriate endpoints to upload attachments when creating tickets or retrieve attachments from existing tickets.
  </Accordion>
</AccordionGroup>

***

<CardGroup>
  <Card title="Need Help?" icon="life-ring" href="https://relevanceai.com/docs/get-started/support">
    Contact our support team for assistance with the Zendesk integration
  </Card>

  <Card title="Explore More Integrations" icon="puzzle-piece" href="/integrations/introduction">
    Discover other integrations to enhance your workflows
  </Card>
</CardGroup>


# Zoom
Source: https://relevanceai.com/docs/integrations/popular-integrations/zoom

Adding support for generating Zoom meeting links

Zoom offers the ability to create cloud web meetings.

### How to add integration?

1. Go to the **Integrations & API Keys** page from the sidebar
2. Select Zoom from the list
3. Click "+ Integration"
4. Complete OAuth flow by logging into Zoom account

### How to use?

Once your account is connected, you can use Zoom steps in the Tool builder. You will have access to a "Zoom API Call" step which you will be able to use to create a meeting, update a meeting, delete a meeting, and more. You can run that step with the required API call for scheduling a meeting.

#### Subdomain field

The Zoom API Call step includes an optional **Subdomain** field for enterprise Zoom deployments that use a custom subdomain. If your organization's Zoom instance is accessed at a URL like `myorg.zoom.us` rather than the standard `zoom.us`, enter the subdomain portion only (e.g., `myorg`, not `myorg.zoom.us`). Leave this field empty if you use the standard zoom.us endpoint — it defaults to zoom.us when not specified.

### How to remove integration?

1. Go to the **Integrations & API Keys** page from the sidebar
2. Select Zoom from the list
3. Click "..." on the account you want to remove
4. Click remove

### Frequently asked questions

<AccordionGroup>
  <Accordion title="Why am I getting an authentication error saying only the account owner can connect Zoom?">
    This issue occurs when you're trying to connect Zoom as an admin user (not the account owner) and your admin role lacks the necessary permissions for external API connections.

    Two specific permissions must be enabled in Zoom's admin role settings:

    * **Chat messages** — allows API access to all users' chat messages
    * **External Connections** — allows admin accounts to view and manage external connections

    To fix this:

    1. Log into your Zoom account as an admin.
    2. Navigate to **Admin > Role Management**.
    3. Select your admin role.
    4. Check the boxes for both **Chat messages** and **External Connections**.
    5. Save your changes.
    6. Retry connecting your Zoom account in Relevance AI.

    This applies whether you're connecting directly through Relevance AI or via third-party services like Pipedream.
  </Accordion>
</AccordionGroup>


# ZoomInfo
Source: https://relevanceai.com/docs/integrations/popular-integrations/zoominfo

Search ZoomInfo's B2B database and enrich contact and company records from your AI agents.

ZoomInfo is a B2B database and intelligence platform that provides contact and company information for sales and marketing teams. With the Relevance AI ZoomInfo integration, your AI agents can search ZoomInfo's database for free and enrich records with detailed contact and company data using ZoomInfo credits. This integration is compatible with GTM agents.

<Warning>
  The ZoomInfo integration in Relevance AI was built by Relevance AI, and is therefore supported by our team, not ZoomInfo. If you have a question or issue with using ZoomInfo in Relevance AI, please [reach out to our support team](/docs/get-started/support). If you have a question or issue that is only about ZoomInfo, you can reach out to ZoomInfo support directly.
</Warning>

## Connect the integration

The ZoomInfo integration uses credential-based authentication. You will need three values from your ZoomInfo account: your ZoomInfo email address, a client ID, and a private key.

<Steps>
  <Step title="Find your ZoomInfo API credentials">
    Log in to your ZoomInfo account and navigate to the API settings or developer portal. Locate your **Client ID** and **Private key**. These are generated in your ZoomInfo account under the API or integrations section. If you do not see them, contact your ZoomInfo account administrator.
  </Step>

  <Step title="Open the Integrations page in Relevance AI">
    Go to **Integrations & API Keys** in the sidebar of your Relevance AI dashboard.
  </Step>

  <Step title="Select ZoomInfo">
    Find and click on **ZoomInfo** from the available integrations list.
  </Step>

  <Step title="Click Add Integration">
    Click **Add Integration** to open the credentials form.
  </Step>

  <Step title="Enter your credentials">
    Fill in the three required fields:

    * **ZoomInfo email**: The email address associated with your ZoomInfo account
    * **Client ID**: Your ZoomInfo API client ID
    * **Private key**: Your ZoomInfo API private key
  </Step>

  <Step title="Save the connection">
    Click **Save** to connect your account. Once authenticated, your ZoomInfo account will appear as a connected integration.
  </Step>
</Steps>

<Warning>
  Keep your private key secure. Do not share it publicly or commit it to version control. If your private key is compromised, regenerate it from your ZoomInfo account settings.
</Warning>

## Understanding credit consumption

ZoomInfo uses a credit-based system for data enrichment. It is important to understand which operations are free and which consume credits before building your workflows.

### Free operations

**Searching does not consume ZoomInfo credits.** The following operations are free:

* Searching for contacts by name, title, company, location, or other criteria
* Searching for companies by name, industry, size, location, or other filters
* Browsing and filtering ZoomInfo's database to build prospect lists

### Credit-consuming operations

**Enrichment operations consume ZoomInfo credits.** Credits are used when you retrieve detailed data for a specific record, including:

* Email addresses
* Phone numbers
* Company revenue figures
* Employee counts
* Other detailed contact and company attributes

The integration includes built-in credit safeguards to help prevent unexpected credit usage. Monitor your credit balance in your ZoomInfo dashboard and design workflows to enrich only when necessary.

<Tip>
  Use search operations to filter and qualify prospects first, then enrich only the records that meet your criteria. This approach minimizes credit consumption.
</Tip>

## Tools & tool steps

The ZoomInfo integration provides actions for both free search and credit-based enrichment. You can add these as tool steps when building tools in Relevance AI.

### Search operations (free)

<CardGroup>
  <Card title="Search contacts" icon="magnifying-glass">
    Search for contacts matching criteria such as job title, company, location, or industry. Does not consume credits.
  </Card>

  <Card title="Search companies" icon="buildings">
    Find companies matching filters like industry, employee count, revenue range, or location. Does not consume credits.
  </Card>
</CardGroup>

### Enrichment operations (consumes credits)

<CardGroup>
  <Card title="Enrich contact" icon="user-plus">
    Retrieve detailed contact data including email addresses, phone numbers, and professional details for a specific person. Consumes credits.
  </Card>

  <Card title="Enrich company" icon="building">
    Retrieve detailed company data including revenue, employee count, technologies used, and firmographic details. Consumes credits.
  </Card>
</CardGroup>

### How to add ZoomInfo tool steps to your agent

1. Create a new tool in your Relevance AI workspace or open an existing one.
2. Scroll down to **Tool steps**.
3. Search for "ZoomInfo" in the tool step search bar.
4. Select the desired ZoomInfo action.
5. Select your connected ZoomInfo account from the dropdown.
6. Configure the action parameters and connect the tool to your agent.

## Use the ZoomInfo API tool step (advanced)

Advanced users can make custom API calls to any ZoomInfo endpoint using the ZoomInfo API Call tool step.

<Steps>
  <Step title="Create or open a tool">
    Create a new tool in Relevance AI or open an existing tool you want to add ZoomInfo functionality to.
  </Step>

  <Step title="Add the ZoomInfo API Call tool step">
    Scroll down to Tool steps, search for "ZoomInfo API Call", and add it to your workflow.
  </Step>

  <Step title="Select your ZoomInfo account">
    Select your connected ZoomInfo account from the dropdown.
  </Step>

  <Step title="Configure the API endpoint">
    Set the HTTP method, endpoint path, and request body parameters according to the ZoomInfo API documentation.
  </Step>

  <Step title="Test your configuration">
    Test your API call before deploying to confirm it returns the expected data.
  </Step>
</Steps>

<Accordion title="Example: Custom company enrichment">
  ```json theme={null}
  {
    "endpoint": "/companies/enrich",
    "method": "POST",
    "body": {
      "company_name": "{{input.company_name}}",
      "domain": "{{input.domain}}",
      "include_attributes": ["industry", "revenue", "employee_count", "technologies", "location"]
    }
  }
  ```

  This call retrieves specific company attributes for use in lead qualification. Note that enrichment calls consume ZoomInfo credits.
</Accordion>

## Example use cases

<AccordionGroup>
  <Accordion title="GTM agent prospect research" icon="magnifying-glass-chart">
    Build a GTM agent that searches ZoomInfo for companies matching your ideal customer profile, then enriches only the top-scoring matches with detailed contact and company data. The agent can prepare outreach briefs and route leads to the appropriate sales representatives.
  </Accordion>

  <Accordion title="Credit-efficient lead enrichment" icon="coins">
    Create an agent that first searches ZoomInfo to filter and qualify leads based on free criteria (company size, industry, location), then enriches only the records that pass qualification with email addresses and phone numbers. This pattern keeps credit usage low while maximizing data quality.
  </Accordion>

  <Accordion title="Account-based marketing research" icon="bullseye">
    Deploy an agent that searches ZoomInfo for decision-makers at target accounts, enriches their contact details, and compiles account briefs for your marketing team without consuming credits on records outside your target list.
  </Accordion>

  <Accordion title="CRM data enrichment" icon="database">
    Build an agent that identifies incomplete records in your CRM and uses ZoomInfo to fill in missing email addresses, phone numbers, and company details. Implement logic to skip records that already have complete data to avoid unnecessary credit consumption.
  </Accordion>
</AccordionGroup>

## Best practices

<AccordionGroup>
  <Accordion title="Minimize credit consumption" icon="coins">
    Use search operations to filter your prospect pool before enriching. Only enrich records that meet your qualification criteria, and implement checks to avoid enriching the same record twice. Monitor credit usage in your ZoomInfo dashboard and set up alerts if available.
  </Accordion>

  <Accordion title="Secure your credentials" icon="shield-check">
    Store your ZoomInfo client ID and private key securely. Do not hard-code credentials in tool configurations visible to other team members. Rotate your private key if you suspect it has been compromised.
  </Accordion>

  <Accordion title="Data privacy and compliance" icon="scale-balanced">
    Only use ZoomInfo data for legitimate business purposes. Comply with GDPR, CCPA, and other applicable privacy regulations in your region. Review ZoomInfo's terms of service and acceptable use policy before building automated outreach workflows.
  </Accordion>
</AccordionGroup>

## Frequently asked questions (FAQs)

<AccordionGroup>
  <Accordion title="Does using ZoomInfo consume credits?">
    It depends on the operation. **Search operations are free** — searching for contacts or companies does not consume ZoomInfo credits. **Enrichment operations consume credits** — retrieving detailed data such as email addresses, phone numbers, revenue figures, or employee counts for a specific record uses credits. Design your workflows to search first and enrich only when needed.
  </Accordion>

  <Accordion title="What credentials do I need to connect ZoomInfo?">
    You need three credentials from your ZoomInfo account: your ZoomInfo email address, a client ID, and a private key. These are available in the API or integrations section of your ZoomInfo account settings. Contact your ZoomInfo account administrator if you do not have API access.
  </Accordion>

  <Accordion title="Does ZoomInfo work with GTM agents?">
    Yes. The ZoomInfo integration is compatible with GTM agents in Relevance AI. You can use ZoomInfo search and enrichment tool steps in GTM agent workflows.
  </Accordion>

  <Accordion title="What happens if I run out of ZoomInfo credits?">
    Enrichment operations will fail when your ZoomInfo credit balance is exhausted. Search operations will continue to work since they do not consume credits. Monitor your credit balance in your ZoomInfo dashboard and top up as needed. The integration includes built-in credit safeguards to help prevent unexpected over-usage.
  </Accordion>

  <Accordion title="Do I need a ZoomInfo account to use this integration?">
    Yes. You need an active ZoomInfo account with API access and valid API credentials (email, client ID, and private key) to connect the integration.
  </Accordion>
</AccordionGroup>


# Remove Integrations
Source: https://relevanceai.com/docs/integrations/remove-integrations

Manage your connected services by removing integrations that are no longer needed.

## Overview

While integrations enhance your AI agents' capabilities by connecting them to external services, there may be times when you need to remove these connections. Whether you're switching service providers, or simply cleaning up unused connections, Relevance AI provides a straightforward process for removing integrations.

## How to Remove Integrations

Follow these steps to remove an integration from your Relevance AI account:

1. Navigate to the **Integrations** page from the left-hand menu of your Relevance AI account.
2. Locate the integration you want to remove in your list of connected services.
3. Click on the integration to expand its details.
4. Click the three dots settings button and click the **Remove** button.

Once disconnected, the integration will no longer appear in your list of available connections, and your AI agents will no longer have access to that service.

## Important Considerations

Before removing an integration, consider the following:

### Impact on Agents

* **Check Dependencies**: Verify which agents are currently using the integration before removing it.
* **Update Agent Workflows**: Agents that rely on the removed integration may need reconfiguration to prevent workflow disruptions.
* **Tool Availability**: Tools that depend on the removed integration will no longer function properly.

## Reconnecting Integrations

If you need to reconnect an integration after removing it:

1. Navigate to the **Integrations** page.
2. Browse the available integrations and select the service you want to reconnect.
3. Follow the authentication steps to reauthorize the connection.
4. Once reconnected, the integration will appear in your list of available connections.

## Frequently asked questions (FAQs)

**Q: How do I know which agents are using a particular integration?**\
A: Before removing an integration, go to each agent's settings and check the "Connected Resources" or "Tools" section to see which integrations they're using.


# Agents
Source: https://relevanceai.com/docs/sdk/agents

Load agents, read their metadata, send messages, and list tasks.

The `Agent` class represents a single agent on the Relevance AI platform. Use it to start conversations, fetch existing tasks, and read configuration set in the dashboard.

## Load a single agent

Fetch an agent by its ID. The agent ID is visible on the agent's page in the dashboard.

```typescript theme={null}
import { Agent } from "@relevanceai/sdk";

const agent = await Agent.get("<agent-id>");
```

To use a specific client instead of the default singleton:

```typescript theme={null}
const agent = await Agent.get("<agent-id>", client);
```

## List all agents

Fetch a paginated list of every agent in the project:

```typescript theme={null}
const agents = await Agent.getAll();
```

Pagination is controlled with `pageSize` and `page`:

```typescript theme={null}
const agents = await Agent.getAll({
  pageSize: 50,
  page: 2,
});
```

The defaults are `pageSize: 20` and `page: 1`.

## Read agent metadata

Once loaded, an agent exposes a number of fields:

```typescript theme={null}
const agent = await Agent.get("<agent-id>");

console.log(agent.name);        // display name
console.log(agent.description); // agent description
console.log(agent.avatar);      // avatar emoji or URL
console.log(agent.id);          // unique identifier
console.log(agent.region);      // deployment region
console.log(agent.project);     // project identifier
console.log(agent.createdAt);   // Date object
console.log(agent.updatedAt);   // Date object
```

`name`, `description`, and `avatar` may be `undefined` if they haven't been configured in the dashboard.

## Start a conversation

Send a message with `sendMessage`. This creates a new task (the conversation thread) and returns it immediately. The method doesn't wait for the agent to respond.

```typescript theme={null}
const task = await agent.sendMessage("What can you help me with?");
```

To continue an existing conversation, pass the task back:

```typescript theme={null}
await agent.sendMessage("Tell me more about that", task);
```

File attachments can be included as well:

```typescript theme={null}
const file = new File([bytes], "report.pdf", {
  type: "application/pdf",
});

const task = await agent.sendMessage("Summarize this report", [file]);
```

For a full walkthrough of message types, attachments, and response handling, see [Messaging](/docs/sdk/messaging).

## Retrieve tasks

### A single task

Fetch a specific task by ID:

```typescript theme={null}
const task = await agent.getTask("<task-id>");
```

### A list of tasks

Fetch a paginated, sortable, and filterable list of tasks for the agent:

```typescript theme={null}
const tasks = await agent.getTasks({
  pageSize: 10,
  page: 1,
  sort: { updatedAt: "desc" },
  filter: { status: ["running", "queued"] },
  search: "quarterly report",
});
```

All options are optional. The defaults are `pageSize: 100`, `page: 1`, and `sort: { createdAt: "asc" }`.

Filtering by status accepts an array of simplified task statuses. For the full lifecycle and status values, see [Tasks](/docs/sdk/tasks).


# Authentication
Source: https://relevanceai.com/docs/sdk/authentication

API keys for server code, embed keys for the browser, and region selection.

The SDK supports two authentication methods. Each one is designed for a specific environment.

## API keys vs. embed keys

<CardGroup>
  <Card title="API keys" icon="server">
    Grant full access to a project. Use in server-side applications, background workers, and other secure environments where the key is never exposed to end users.
  </Card>

  <Card title="Embed keys" icon="browser">
    Scoped to a single public agent or workforce. Use in browser applications and other client-side contexts where the key is visible in network traffic.
  </Card>
</CardGroup>

Rule of thumb: if the code runs on a server, use an API key. If it runs in a browser, use an embed key.

## Regions

Every Relevance AI project is deployed to a specific region. Match the region where your project was created — check the project settings in the dashboard.

| Constant    | Region        |
| ----------- | ------------- |
| `REGION_US` | United States |
| `REGION_EU` | Europe        |
| `REGION_AU` | Australia     |

```typescript theme={null}
import { REGION_US, REGION_EU, REGION_AU } from "@relevanceai/sdk";
```

## API key authentication

API keys are available in the project settings on the dashboard. They follow the `sk-...` format and grant unrestricted access to every resource in the project.

The fastest way to authenticate is with `createClient`:

```typescript theme={null}
import { createClient, REGION_AU } from "@relevanceai/sdk";

const client = createClient({
  apiKey: "sk-...",
  region: REGION_AU,
  project: "<project-id>",
});
```

For more control, construct a `Key` instance explicitly and pass it to the `Client` constructor:

```typescript theme={null}
import { Client, Key, REGION_AU } from "@relevanceai/sdk";

const key = new Key({
  key: "sk-...",
  region: REGION_AU,
  project: "<project-id>",
});

const client = new Client(key);
```

Both approaches produce an authenticated client. For details on `createClient` vs. direct construction, see [Client](/docs/sdk/client).

## Embed key authentication

Embed keys are generated at runtime and scoped to a single public agent or a single workforce. They're safe to use in browser environments because they can't access resources outside their scope.

### For an agent

```typescript theme={null}
import { Key, createClient, REGION_US } from "@relevanceai/sdk";

const key = await Key.generateEmbedKey({
  region: REGION_US,
  project: "<project-id>",
  agentId: "<public-agent-id>",
});

const client = createClient(key);
```

The agent must be marked as public in the dashboard. Private agents can't be accessed with embed keys.

### For a workforce

```typescript theme={null}
import { Key, REGION_US } from "@relevanceai/sdk";

const key = await Key.generateEmbedKey({
  region: REGION_US,
  project: "<project-id>",
  workforceId: "<workforce-id>",
});
```

Each embed key is bound to exactly one agent or one workforce. To interact with multiple agents or workforces from the client side, generate a separate embed key for each.

## Persisting embed keys

Generating an embed key involves a network request and, importantly, initializes a specific user session. You should persist keys to avoid unnecessary latency and to ensure session continuity; **regenerating a new key on every page load will prevent the user from restoring their past conversations.**

To maintain the session, persist the key with `toJSON()` and restore it using the `Key` constructor.

### Save to localStorage

```typescript theme={null}
const key = await Key.generateEmbedKey({
  region: REGION_US,
  project: "<project-id>",
  agentId: "<agent-id>",
});

localStorage.setItem("relevance_key", JSON.stringify(key.toJSON()));
```

### Restore from localStorage

```typescript theme={null}
import { Key, Client } from "@relevanceai/sdk";

const stored = localStorage.getItem("relevance_key");
if (stored) {
  const key = new Key(JSON.parse(stored));
  const client = new Client(key);
}
```

`toJSON` returns a plain object containing every field needed to reconstruct the key — region, project, scoped agent or workforce ID, and the task prefix used for conversation namespacing.

## Security guidance

<Warning>
  Never embed API keys in client-side code, browser-accessible environment variables, or version control. API keys grant full project access and must remain on the server.
</Warning>

* Embed keys are the only safe authentication method for browser environments. They're scoped to a single agent or workforce and can't access other project resources.
* Immediately revoke any compromised API keys before regenerating. Existing keys do not auto-expire and must be manually decommissioned via the dashboard to terminate access.
* Store embed keys in `localStorage`, a secure cookie, or an equivalent client-side persistence mechanism. Regenerate them if the storage is cleared.


# Client
Source: https://relevanceai.com/docs/sdk/client

Initialize the SDK client, use the default singleton, or manage multiple clients.

The `Client` is the authenticated handle the SDK uses to talk to the Relevance AI platform. Most applications only need one.

## Initialize with createClient

`createClient` is the recommended way to set up the SDK. It constructs a `Client`, registers it as the default singleton, and returns it.

```typescript theme={null}
import { createClient, REGION_EU } from "@relevanceai/sdk";

const client = createClient({
  apiKey: "sk-...",
  region: REGION_EU,
  project: "<project-id>",
});
```

It also accepts a `Key` instance directly, which is useful when working with embed keys:

```typescript theme={null}
import { createClient, Key, REGION_US } from "@relevanceai/sdk";

const key = await Key.generateEmbedKey({
  region: REGION_US,
  project: "<project-id>",
  agentId: "<agent-id>",
});

const client = createClient(key);
```

For full details on keys, see [Authentication](/docs/sdk/authentication).

<Note>
  Call `createClient` exactly once. It throws if a default client already exists. To work with additional projects or authentication scopes, construct `Client` instances directly — see [Using multiple clients](#using-multiple-clients).
</Note>

## The default singleton

Once `createClient` runs, the client is stored as a global default. Every SDK method that takes an optional client parameter falls back to this default when none is passed.

```typescript theme={null}
import { createClient, Client, Agent, REGION_US } from "@relevanceai/sdk";

// Set the default at application startup
createClient({
  apiKey: "sk-...",
  region: REGION_US,
  project: "<project-id>",
});

// Retrieve the default anywhere in the application
const client = Client.default();

// SDK methods pick it up automatically
const agent = await Agent.get("<agent-id>");
```

This keeps authentication centralized. Initialize the client once during startup, then the rest of the application can interact with agents and workforces without passing the client around.

## Using multiple clients

Some apps need to talk to multiple projects, or use different authentication scopes — for example, an API key for admin operations and an embed key for end-user interactions. In those cases, construct `Client` instances directly and pass them explicitly.

```typescript theme={null}
import { Client, Key, Agent, REGION_EU } from "@relevanceai/sdk";

const adminKey = new Key({
  key: "sk-admin-key",
  region: REGION_EU,
  project: "project-one",
});

const userKey = new Key({
  key: "sk-user-key",
  region: REGION_EU,
  project: "project-two",
});

const adminClient = new Client(adminKey);
const userClient = new Client(userKey);

// Pass the client explicitly
const adminAgent = await Agent.get("agent-a", adminClient);
const userAgent = await Agent.get("agent-b", userClient);
```

<Note>
  Only one client can be the default singleton. When using multiple clients, pass the right instance to every SDK method call.
</Note>


# Installation
Source: https://relevanceai.com/docs/sdk/installation

Install @relevanceai/sdk in Node.js, Deno, Bun, Cloudflare Workers, or the browser.

The SDK ships to both npm and JSR, and works in every major JavaScript runtime with no extra configuration.

## Prerequisites

Before installing, make sure you have:

* A Relevance AI account with access to the project dashboard
* An API key or embed key — see [Authentication](/docs/sdk/authentication)
* One of the supported runtimes below

## Node.js, Bun, and Cloudflare Workers

Install from npm with any package manager.

<Tabs>
  <Tab title="npm">
    ```bash theme={null}
    npm install @relevanceai/sdk@latest
    ```
  </Tab>

  <Tab title="yarn">
    ```bash theme={null}
    yarn add @relevanceai/sdk@latest
    ```
  </Tab>

  <Tab title="pnpm">
    ```bash theme={null}
    pnpm add @relevanceai/sdk@latest
    ```
  </Tab>

  <Tab title="Bun">
    ```bash theme={null}
    bun add @relevanceai/sdk@latest
    ```
  </Tab>
</Tabs>

Then import the named exports you need:

```typescript theme={null}
import { createClient, Agent } from "@relevanceai/sdk";
```

## Deno

Add the package from JSR:

```bash theme={null}
deno add jsr:@relevanceai/sdk
```

Or import directly without a local install:

```typescript theme={null}
import { createClient } from "jsr:@relevanceai/sdk";
```

## Browser via CDN

For browser apps that don't use a bundler, load the SDK through an import map:

```html theme={null}
<script type="importmap">
  {
    "imports": {
      "@relevanceai/sdk": "https://esm.run/@relevanceai/sdk"
    }
  }
</script>
<script type="module">
  import { createClient } from "@relevanceai/sdk";
</script>
```

## Browser bundler configuration

When bundling for the browser with Vite, Webpack, or Rollup, you may see a warning about `node:crypto`. The SDK uses this module for UUID generation and falls back to the native browser equivalent. To silence the warning, add a shim and point the bundler at it.

<Steps>
  <Step title="Create the shim">
    Save this as `src/shims/crypto.ts`:

    ```typescript theme={null}
    export default globalThis.crypto;
    ```
  </Step>

  <Step title="Configure the bundler">
    <AccordionGroup>
      <Accordion title="Vite">
        In `vite.config.ts`, add a resolve alias:

        ```typescript theme={null}
        import { defineConfig } from "vite";
        import path from "node:path";

        export default defineConfig({
          resolve: {
            alias: {
              "node:crypto": path.resolve(
                __dirname,
                "src/shims/crypto.ts"
              ),
            },
          },
        });
        ```
      </Accordion>

      <Accordion title="Webpack">
        In `webpack.config.js`, add a resolve alias:

        ```javascript theme={null}
        module.exports = {
          resolve: {
            alias: {
              "node:crypto": path.resolve(
                __dirname,
                "src/shims/crypto.js"
              ),
            },
          },
        };
        ```
      </Accordion>

      <Accordion title="Rollup">
        Use the `@rollup/plugin-alias` plugin:

        ```javascript theme={null}
        import alias from "@rollup/plugin-alias";
        import path from "node:path";

        export default {
          plugins: [
            alias({
              entries: [
                {
                  find: "node:crypto",
                  replacement: path.resolve(
                    __dirname,
                    "src/shims/crypto.js"
                  ),
                },
              ],
            }),
          ],
        };
        ```
      </Accordion>
    </AccordionGroup>
  </Step>
</Steps>

## Verify the installation

Run a minimal script to confirm everything is wired up. Replace the placeholder values with real credentials from the dashboard.

```typescript theme={null}
import { createClient, Agent, REGION_US } from "@relevanceai/sdk";

createClient({
  apiKey: "<api-key>",
  region: REGION_US,
  project: "<project-id>",
});

const agent = await Agent.get("<agent-id>");
console.log("Connected:", agent.name);
```

If the agent's name prints to the console, the SDK is installed and authenticated correctly.


# JavaScript SDK
Source: https://relevanceai.com/docs/sdk/introduction

Build apps on top of Relevance AI agents and workforces with the official JavaScript/TypeScript SDK.

`@relevanceai/sdk` is the official JavaScript and TypeScript SDK for the Relevance AI platform. Use it to start conversations with agents, trigger workforces, stream responses in real time, and handle file attachments — all from your own application code.

## What's in the box

<CardGroup>
  <Card title="Universal runtime support" icon="globe">
    Works in Node.js, Deno, Bun, Cloudflare Workers, and the browser. Zero dependencies — built on web standards.
  </Card>

  <Card title="Event-driven" icon="rss">
    Real-time updates through the native `EventTarget` API. No custom event system to learn.
  </Card>

  <Card title="Type-safe" icon="shield-check">
    Full TypeScript definitions across agents, tasks, messages, and workforces. Type guards narrow message kinds automatically.
  </Card>

  <Card title="Streaming by default" icon="bolt">
    Thinking and typing tokens arrive through the same event stream as final messages — no extra configuration.
  </Card>
</CardGroup>

## When to use the SDK

* Embedding chat into a web or mobile app
* Triggering agents from a backend service, worker, or cron job
* Running workforces from a server-side pipeline
* Building custom dashboards on top of agent output

If you just need to fire a one-off webhook or call from a language the SDK doesn't support, use the [API trigger](/docs/build/agents/build-your-agent/agent-triggers/api-trigger) instead.

## Where to next

<CardGroup>
  <Card title="Quickstart" icon="rocket" href="/sdk/quickstart">
    Install, authenticate, and send your first message in under a minute.
  </Card>

  <Card title="Authentication" icon="key" href="/sdk/authentication">
    API keys for server code, embed keys for the browser, and region selection.
  </Card>

  <Card title="Agents" icon="robot" href="/sdk/agents">
    Load agents, send messages, and list tasks.
  </Card>

  <Card title="Workforces" icon="users" href="/sdk/workforces">
    Trigger multi-agent workforces and handle handover messages.
  </Card>
</CardGroup>


# Messaging
Source: https://relevanceai.com/docs/sdk/messaging

Send messages, work with message types, manage file attachments, and handle errors.

Messages are how your application and an agent exchange information inside a task. This page covers sending messages, the different kinds of incoming messages, and how to handle them.

## Send a message

Both agents and workforces use `sendMessage`. It accepts a string and returns a `Task` representing the conversation thread.

```typescript theme={null}
const task = await agent.sendMessage("What can you help me with?");
```

The returned task begins processing immediately. The promise resolves as soon as the message is sent — not when the agent responds. To receive the response, listen for events on the task. See [Tasks](/docs/sdk/tasks) for event handling.

## Continue a conversation

To send a follow-up message within the same conversation, pass the existing task as the second argument:

```typescript theme={null}
const task = await agent.sendMessage("Hello");

// Later, continue the same conversation
await agent.sendMessage("Tell me more", task);
```

Each call with an existing task appends to the conversation history. The agent has full context of the prior messages.

## File attachments

Attach files by passing an array of `File` objects as the second argument. The SDK handles the upload.

```typescript theme={null}
const pdf = new File([bytes], "contract.pdf", {
  type: "application/pdf",
});

const task = await agent.sendMessage("Summarize this contract", [pdf]);
```

Multiple files in a single message:

```typescript theme={null}
const task = await agent.sendMessage(
  "Compare these two documents",
  [fileA, fileB]
);
```

To send attachments while continuing a conversation, pass the attachments array and the task:

```typescript theme={null}
await agent.sendMessage(
  "Here is an updated version",
  [updatedFile],
  task
);
```

Pre-uploaded attachments (objects with `fileName` and `fileUrl` fields) can be mixed in with `File` objects in the same array.

<Note>
  Workforce tasks don't support file attachments. Only agent tasks accept files.
</Note>

## Handle incoming messages

Messages arrive through the `"message"` event on a task. Each event's `detail` includes a `message` property. Use the type guard methods to determine the kind and access the right fields.

```typescript theme={null}
task.addEventListener("message", ({ detail }) => {
  const { message } = detail;

  if (message.isAgent()) {
    console.log("Agent:", message.text);
  } else if (message.isTool()) {
    console.log("Tool:", message.status);
  } else if (message.isThinking()) {
    console.log("Thinking:", message.text);
  } else if (message.isTyping()) {
    console.log("Typing:", message.text);
  }
});
```

The type guards narrow the TypeScript type — properties specific to each message kind become available after the check.

## Agent responses

When `message.isAgent()` returns `true`, the message is an `AgentMessage` carrying the agent's response text.

```typescript theme={null}
if (message.isAgent()) {
  console.log(message.text);
  console.log(message.agentId);
}
```

## Tool executions

When `message.isTool()` returns `true`, the message is a `ToolMessage` representing a tool or subagent execution inside the agent's pipeline.

```typescript theme={null}
if (message.isTool()) {
  console.log("Tool:", message.tool?.name);
  console.log("Status:", message.status);
}
```

The `status` field tracks where the tool is in its lifecycle:

| Status      | Meaning                    |
| ----------- | -------------------------- |
| `pending`   | Scheduled, not yet started |
| `running`   | Currently executing        |
| `completed` | Finished successfully      |
| `error`     | Failed with an error       |
| `cancelled` | Execution was cancelled    |

### Read tool output

Once a tool completes, its output is available:

```typescript theme={null}
if (message.isTool() && message.status === "completed") {
  console.log(message.output);
}
```

### Detect subagents

A tool message may represent a subagent rather than a standalone tool. Use `isSubAgent` to check, and `subAgentTaskId` to access the subagent's task:

```typescript theme={null}
if (message.isTool() && message.isSubAgent()) {
  console.log("Sub-agent task:", message.subAgentTaskId);
}
```

### Tool errors

Check for errors on a tool message with `hasErrors`:

```typescript theme={null}
if (message.isTool() && message.hasErrors()) {
  for (const error of message.errors) {
    console.error(error.stepName, error.message);
  }
}
```

## User messages

User messages represent input sent by the application or end user. They show up in the message history alongside agent responses.

```typescript theme={null}
task.addEventListener("message", ({ detail }) => {
  const { message } = detail;

  if (message.isUser()) {
    console.log("User:", message.text);

    if (message.hasAttachments()) {
      for (const attachment of message.attachments) {
        console.log("File:", attachment.fileName);
      }
    }
  }
});
```

`isTrigger` returns `true` for the first message in a conversation — the one that created the task:

```typescript theme={null}
if (message.isUser() && message.isTrigger()) {
  console.log("Conversation started with:", message.text);
}
```

## Error handling

Agent errors are delivered through the `"error"` event on the task. The event detail contains an `AgentErrorMessage` with one or more error strings.

```typescript theme={null}
task.addEventListener("error", ({ detail }) => {
  const { message } = detail;
  console.error("Last error:", message.lastError);

  // All errors in this cycle
  for (const error of message.errors) {
    console.error(error);
  }
});
```

Error events mean the agent ran into a problem during execution. The task may transition to an `"error"` or `"action"` status depending on the nature of the failure. See [Tasks](/docs/sdk/tasks) for status details.

## Workforce messages

Workforce tasks include two additional message types for multi-agent coordination:

* **`WorkforceAgentMessage`** — an agent within the workforce is executing a subtask. Includes details about which agent is running and the state of its work.
* **`WorkforceAgentHandoverMessage`** — work is being delegated from one agent to another within the workforce. Includes the trigger message and details about the receiving agent.

These messages appear alongside standard agent and tool messages in the event stream. See [Workforces](/docs/sdk/workforces) for workforce-specific behavior.

## Streaming messages

Two message types represent real-time incremental output from the agent:

* **`ThinkingMessage`** — the agent's intermediate reasoning as it processes.
* **`TypingMessage`** — the agent's response text as it's being generated.

Both are identified with the `isThinking` and `isTyping` type guards shown above. For streaming behavior and live typing indicators, see [Streaming](/docs/sdk/streaming).


# Quickstart
Source: https://relevanceai.com/docs/sdk/quickstart

Install the SDK, authenticate, and send your first message in under a minute.

Get from zero to a live agent conversation in five steps. This quickstart uses Node.js — for other runtimes, see [Installation](/docs/sdk/installation).

<Steps>
  <Step title="Install the SDK">
    ```bash theme={null}
    npm install @relevanceai/sdk
    ```

    Yarn, pnpm, Bun, and Deno are all supported. See [Installation](/docs/sdk/installation) for the full matrix.
  </Step>

  <Step title="Grab your credentials">
    From the Relevance AI dashboard, go to **Integrations & API Keys** and generate a Relevance API key. You'll also need your project ID and the region your project is deployed in.

    For a walkthrough, see [API integration](/docs/get-started/core-concepts/api-integration).
  </Step>

  <Step title="Create a client">
    ```typescript theme={null}
    import { createClient, REGION_US } from "@relevanceai/sdk";

    createClient({
      apiKey: process.env.RELEVANCE_API_KEY,
      region: REGION_US,
      project: process.env.RELEVANCE_PROJECT_ID,
    });
    ```

    `createClient` stores the client as the default singleton, so you don't need to pass it around. Replace `REGION_US` with `REGION_EU` or `REGION_AU` to match your project.
  </Step>

  <Step title="Load an agent and send a message">
    ```typescript theme={null}
    import { Agent } from "@relevanceai/sdk";

    const agent = await Agent.get("<agent-id>");
    const task = await agent.sendMessage("What can you help me with?");
    ```

    `sendMessage` returns a `Task` immediately — it doesn't wait for the agent to respond. The agent ID is visible on the agent's page in the dashboard.
  </Step>

  <Step title="Listen for the response">
    ```typescript theme={null}
    task.addEventListener("message", ({ detail: { message } }) => {
      if (message.isAgent()) {
        console.log("Agent:", message.text);
      }
    });
    ```

    Messages arrive through the `"message"` event on the task. Use the type guards (`isAgent`, `isTool`, `isTyping`, and so on) to handle each kind. When the conversation is done, call `task.unsubscribe()` to release the connection.
  </Step>
</Steps>

## What to read next

<CardGroup>
  <Card title="Authentication" icon="key" href="/sdk/authentication">
    Use embed keys for browser apps, or API keys for server code.
  </Card>

  <Card title="Tasks" icon="list-check" href="/sdk/tasks">
    Understand task status, events, and cleanup.
  </Card>

  <Card title="Messaging" icon="messages" href="/sdk/messaging">
    Handle tool calls, user messages, attachments, and errors.
  </Card>

  <Card title="Streaming" icon="wave-sine" href="/sdk/streaming">
    Render responses as the agent types them.
  </Card>
</CardGroup>


# Streaming
Source: https://relevanceai.com/docs/sdk/streaming

Render thinking and typing tokens in real time as the agent produces them.

Streaming delivers incremental thinking and typing tokens as the agent processes — letting your application render live feedback before the final response is complete.

## What streaming provides

As an agent works through a request, it produces two kinds of incremental output:

<CardGroup>
  <Card title="Thinking tokens" icon="brain">
    The agent's intermediate reasoning and planning. Useful for progress indicators or debug views.
  </Card>

  <Card title="Typing tokens" icon="keyboard">
    The agent's response text as it's being generated. Enables character-by-character or chunk-by-chunk rendering.
  </Card>
</CardGroup>

Both arrive through the standard `"message"` event on a task, alongside the final polled messages.

## Streaming is automatic

The SDK enables and manages streaming for you. There's no configuration step. When a task is subscribed to (via `addEventListener`), the SDK opens a streaming connection if one is available. Token refresh, reconnection, and cleanup are handled internally.

Streaming works across every supported runtime: browsers, Node.js, Deno, Bun, and Cloudflare Workers.

## Listen for streaming events

Streaming messages arrive through the same `"message"` event used for every other message type. Identify them with the `isThinking` and `isTyping` type guards:

```typescript theme={null}
task.addEventListener("message", ({ detail }) => {
  const { message } = detail;

  if (message.isThinking()) {
    console.log("[thinking]", message.text);
  }

  if (message.isTyping()) {
    process.stdout.write(message.text);
  }
});
```

Both `ThinkingMessage` and `TypingMessage` expose a `text` property with the incremental content.

## Build a live typing indicator

A common pattern is rendering the agent's response in real time as it's generated. Accumulate typing tokens into a buffer and render them as they arrive:

```typescript theme={null}
let buffer = "";

task.addEventListener("message", ({ detail }) => {
  const { message } = detail;

  if (message.isTyping()) {
    buffer += message.text;
    renderPartialResponse(buffer);
  }

  if (message.isAgent()) {
    // Final response arrived; replace the buffer
    buffer = "";
    renderFinalResponse(message.text);
  }
});
```

The typing tokens represent the same content that will eventually appear in the final `AgentMessage`. Once the final message arrives, replace the accumulated buffer with the complete response.

## Thinking vs. typing

Thinking and typing are different phases of the agent's processing.

* **Thinking** happens while the agent reasons about what to do next, which tools to call, or how to structure its response. You might show this as a "thinking..." indicator, display it in a debug panel, or ignore it entirely.
* **Typing** happens as the agent generates its actual response text. This is the content that ends up in the final agent message.

```typescript theme={null}
task.addEventListener("message", ({ detail }) => {
  const { message } = detail;

  if (message.isThinking()) {
    showThinkingIndicator(message.text);
  }

  if (message.isTyping()) {
    appendToResponse(message.text);
  }
});
```

## Streaming lifecycle

Streaming follows the same lifecycle as task subscriptions:

* **Start** — when the first event listener is added to a task, the SDK subscribes and opens a streaming connection if available.
* **Active** — thinking and typing tokens arrive through the `"message"` event as the agent processes. Connection health and token refresh are managed transparently.
* **End** — when `task.unsubscribe()` is called, the streaming connection is closed along with the polling loop.

There's no separate subscription or connect step for streaming. It's part of the standard task subscription and follows the same cleanup rules described in [Tasks](/docs/sdk/tasks).


# Tasks
Source: https://relevanceai.com/docs/sdk/tasks

Task lifecycle, status model, events, subscriptions, and cleanup.

A task represents a single conversation thread between your application and an agent or workforce. When you send a message, the SDK returns a `Task` — everything that happens next flows through it.

## The shape of a task

```typescript theme={null}
const task = await agent.sendMessage("Hello");

console.log(task.id);     // unique task identifier
console.log(task.name);   // display title
console.log(task.status); // current status
```

## Task status

Every task has a `status` field. The SDK simplifies the platform's internal states into a concise set:

| Status        | Meaning                            |
| ------------- | ---------------------------------- |
| `not-started` | Created locally, not yet sent      |
| `idle`        | Agent is idle, awaiting input      |
| `paused`      | Execution has been paused          |
| `queued`      | Waiting to start (capacity, queue) |
| `running`     | Agent is actively processing       |
| `action`      | Requires approval or has escalated |
| `completed`   | Finished successfully              |
| `cancelled`   | Cancelled before completion        |
| `error`       | Encountered a failure              |

<Accordion title="How internal platform states map to SDK statuses">
  The platform uses more granular internal states. The SDK maps them to the statuses above so application code doesn't need to handle every variation.

  | Platform state             | SDK status  |
  | -------------------------- | ----------- |
  | `idle`                     | `idle`      |
  | `starting-up`              | `queued`    |
  | `waiting-for-capacity`     | `queued`    |
  | `queued-for-approval`      | `queued`    |
  | `queued-for-rerun`         | `queued`    |
  | `running`                  | `running`   |
  | `pending-approval`         | `action`    |
  | `escalated`                | `action`    |
  | `paused`                   | `paused`    |
  | `completed`                | `completed` |
  | `cancelled`                | `cancelled` |
  | `timed-out`                | `error`     |
  | `unrecoverable`            | `error`     |
  | `errored-pending-approval` | `error`     |

  For workforce state mappings, see [Workforces](/docs/sdk/workforces).
</Accordion>

## Is the task still active?

`isRunning` returns `true` when the status is `queued` or `running`:

```typescript theme={null}
if (task.isRunning()) {
  console.log("Agent is working...");
}
```

## Fetch message history

Retrieve every message on a task with `getMessages`:

```typescript theme={null}
const messages = await task.getMessages();
```

To fetch only messages after a specific point, pass an `after` date:

```typescript theme={null}
const recent = await task.getMessages({
  after: new Date("2025-06-01T00:00:00Z"),
});
```

Each message is one of the types described in [Messaging](/docs/sdk/messaging). The same type guard methods (`isAgent`, `isTool`, and so on) work here.

## Listen for events

Tasks emit events in real time as the conversation progresses. There are three event types:

| Event       | When it fires                        |
| ----------- | ------------------------------------ |
| `"message"` | A new message is received            |
| `"error"`   | An agent error occurs                |
| `"update"`  | Task metadata changes (e.g., status) |

### The "message" event

The primary event for agent responses, tool executions, and streaming content:

```typescript theme={null}
task.addEventListener("message", ({ detail }) => {
  const { message } = detail;

  if (message.isAgent()) {
    console.log("Agent:", message.text);
  }
});
```

For the full set of message types and their fields, see [Messaging](/docs/sdk/messaging).

### The "error" event

Fires when the agent runs into a failure:

```typescript theme={null}
task.addEventListener("error", ({ detail }) => {
  console.error(detail.message.lastError);
});
```

### The "update" event

Fires when the task's metadata changes, such as a status transition:

```typescript theme={null}
task.addEventListener("update", () => {
  console.log("New status:", task.status);
});
```

<Warning>
  The current SDK has a TypeScript type mismatch on this event — the declared event map key differs from the string dispatched at runtime. The `"update"` string shown above is correct and events fire as expected. If the type checker complains, add `// @ts-expect-error` above the line until the upstream fix lands.
</Warning>

## Subscriptions are automatic

The first call to `addEventListener` on a task activates its subscription. There's no need to call `subscribe` manually.

Once subscribed, the SDK keeps the task up to date by polling for new messages and metadata changes, and opens a streaming connection when available to deliver real-time thinking and typing events. See [Streaming](/docs/sdk/streaming) for details.

The SDK manages the polling frequency internally — faster when the task is active, slower when idle. This is transparent to the application.

## Clean up

Always call `unsubscribe` when a task is no longer needed. This stops the polling loop and closes any active connections, freeing network and memory resources.

```typescript theme={null}
task.unsubscribe();
```

Failing to unsubscribe from tasks that have left scope is a common source of resource leaks. In component-based UI frameworks, call `unsubscribe` from the cleanup or teardown handler.

<CodeGroup>
  ```typescript React theme={null}
  useEffect(() => {
    const handler = ({ detail }) => {
      setMessages((prev) => [...prev, detail.message]);
    };

    task.addEventListener("message", handler);

    return () => {
      task.unsubscribe();
    };
  }, [task]);
  ```

  ```typescript Vanilla JavaScript theme={null}
  // When navigating away from a conversation view
  function teardown() {
    task.unsubscribe();
  }
  ```
</CodeGroup>


# Workforces
Source: https://relevanceai.com/docs/sdk/workforces

Trigger multi-agent workforces, continue conversations, and handle handover messages.

A workforce is a team of agents that coordinate to handle complex, multistep tasks through delegation and handover. The platform routes messages to the right agent, and agents can pass work between each other as the task progresses — your application doesn't have to manage the coordination.

## Load a workforce

Fetch a workforce by ID. The workforce ID is available on the workforce's page in the dashboard.

```typescript theme={null}
import { Workforce } from "@relevanceai/sdk";

const workforce = await Workforce.get("<workforce-id>");
```

To use a specific client instead of the default singleton:

```typescript theme={null}
const workforce = await Workforce.get("<workforce-id>", client);
```

Once loaded, the workforce exposes its metadata:

```typescript theme={null}
console.log(workforce.name);    // display name
console.log(workforce.id);      // unique identifier
console.log(workforce.region);  // deployment region
console.log(workforce.project); // project identifier
```

## Start a workforce task

Send a message with `sendMessage`. This creates a new task and returns it immediately. The workforce begins routing the message to the appropriate agent.

```typescript theme={null}
const task = await workforce.sendMessage(
  "Analyze last quarter's sales data and prepare a summary"
);
```

As with agent tasks, `sendMessage` resolves when the message is sent, not when the workforce finishes. Listen for events on the returned task to receive results — see [Tasks](/docs/sdk/tasks).

## Continue a conversation

Pass the existing task to `sendMessage` to add follow-up messages:

```typescript theme={null}
await workforce.sendMessage("Include a regional breakdown", task);
```

<Note>
  Unlike agent tasks, workforce tasks don't support file attachments.
</Note>

## Retrieve workforce tasks

### A single task

Fetch a specific task by ID:

```typescript theme={null}
const task = await workforce.getTask("<task-id>");
```

### A list of tasks

Fetch a list of workforce tasks with `getTasks`. Workforce task listing uses cursor-based pagination and supports search:

```typescript theme={null}
const tasks = await workforce.getTasks({
  pageSize: 50,
  search: "quarterly report",
});
```

To paginate through results, pass the cursor from the previous response:

```typescript theme={null}
const firstPage = await workforce.getTasks({ pageSize: 20 });

// Use the cursor from the response to get the next page
const nextPage = await workforce.getTasks({
  pageSize: 20,
  cursor: "<cursor-from-previous-page>",
});
```

### Differences from agent task retrieval

Workforce and agent task listings don't support the same options:

| Feature          | Agent tasks | Workforce tasks |
| ---------------- | ----------- | --------------- |
| Pagination       | Page-based  | Cursor-based    |
| Sort options     | Supported   | Not supported   |
| Status filtering | Supported   | Not supported   |
| Search           | Supported   | Supported       |
| File attachments | Supported   | Not supported   |

## Workforce task states

Workforce tasks use a different set of internal states than agent tasks. The SDK maps them to the same simplified statuses used elsewhere.

| Workforce state            | SDK status  |
| -------------------------- | ----------- |
| `running`                  | `running`   |
| `completed`                | `completed` |
| `execution-limit-reached`  | `error`     |
| `pending-approval`         | `action`    |
| `escalated`                | `action`    |
| `errored-pending-approval` | `error`     |

The `action` status means the workforce needs human intervention — either through an approval step or because the task was escalated.

## Workforce-specific messages

Workforce tasks emit two message types that don't appear in agent tasks.

### Agent run messages

A `WorkforceAgentMessage` (type `"workforce-agent-run"`) indicates that an agent within the workforce is executing a subtask. These messages include details about which agent is running and the current state of its work.

```typescript theme={null}
task.addEventListener("message", ({ detail }) => {
  const { message } = detail;

  if (message.type === "workforce-agent-run") {
    // An agent in the workforce is working
  }
});
```

### Handover messages

A `WorkforceAgentHandoverMessage` (type `"workforce-agent-handover"`) indicates that work is being delegated from one agent to another. The message includes the trigger message that started the handover and details about the receiving agent.

```typescript theme={null}
task.addEventListener("message", ({ detail }) => {
  const { message } = detail;

  if (message.type === "workforce-agent-handover") {
    // Work is being handed to another agent
  }
});
```

These messages appear interleaved with standard agent and tool messages in the event stream. For full details on all message types and type guards, see [Messaging](/docs/sdk/messaging).