Pipecat
Stream meeting audio to your Pipecat agent and play its responses back into the meeting.
Connect a Pipecat Cloud or self-hosted agent to a Recall.ai meeting bot. Your Pipecat pipeline handles the conversation, models, and tools. Recall.ai sends the meeting audio to your agent and plays its audio and, optionally, video back into the meeting.
Setup
1. Prepare your Pipecat agent
Configure your agent to use SmallWebRTCTransport with audio input and output enabled. Recall.ai supports agents running on Pipecat Cloud or, alternatively, self-hosted Pipecat agents running on your infrastructure.
- Pipecat Cloud: Deploy your agent and copy its deployment name. Use that exact name for
agent_namebelow. In the Pipecat dashboard, create a public API key under Settings → API Keys → Public. - Self-hosted: Run your agent with an HTTP(S)
SmallWebRTCoffer endpoint that Recall.ai can reach, typicallyhttps://your-agent.example.com/api/offer. Use the full endpoint URL forurlbelow.
A note onSmallWebRTCA Pipecat agent configured only for Daily transport needs SmallWebRTC support to connect.
2. Add your credentials to Recall.ai
Open Agents → Pipecat in the Recall region where you create bots:
Select the workspace used by your Recall API key, then save the credentials for your deployment:
- Pipecat Cloud: Save your public API key and leave Self-hosted offer URL empty.
- Self-hosted with authentication: Save your bearer token and the exact offer URL you will use in
config.url. Recall.ai sendsAuthorization: Bearer <token>only when the URL matches the saved URL exactly. - Self-hosted without authentication: No saved credentials are required.
Each workspace stores one Pipecat credential configuration. Enter a new key or token when changing the saved URL. Credential changes apply to new connections.
Quickstart
Call Create Bot with your meeting URL and Pipecat Cloud deployment name. The variant setting below selects web_4_core on Zoom, Google Meet, Microsoft Teams, and Webex. See bot sizing guidance.
{
"meeting_url": "MEETING_URL",
"variant": {
"zoom": "web_4_core",
"google_meet": "web_4_core",
"microsoft_teams": "web_4_core",
"webex": "web_4_core"
},
"output_media": {
"audio": {
"kind": "pipecat",
"config": {
"agent_name": "meeting-agent"
}
}
}
}Once the bot joins the meeting, Recall.ai starts a Pipecat Cloud session, sends it a mixed meeting audio stream, and plays the agent's replies into the meeting.
For a self-hosted agent, replace the output_media object in the request above with:
{
"audio": {
"kind": "pipecat",
"config": {
"url": "https://your-agent.example.com/api/offer"
}
}
}To add chat messages, screensharing, and a leave action, see Add meeting tools.
Additional configurations
The same config object is supported under output_media.audio, output_media.camera, and output_media.screenshare when kind is pipecat:
| Field | Required | Description |
|---|---|---|
agent_name | For Pipecat Cloud | Your deployment name. Use 1–256 letters, digits, underscores, or hyphens. Requires a saved Pipecat Cloud key. Mutually exclusive with url. |
url | For self-hosted agents | The full HTTP(S) SmallWebRTC offer endpoint, up to 4,096 characters. Must not contain embedded credentials, whitespace, or a fragment. Mutually exclusive with agent_name. |
body | No | A JSON object containing data for your agent, up to 1 MiB when serialized. Its fields are defined by your agent. |
Supply exactly one of agent_name or url. These three fields are the complete accepted configuration: api_key, credential_id, custom headers, and transport are not accepted in the API request. Save authentication credentials in the dashboard.
Pass data to your agent
Use body to provide context or other inputs that your agent is written to read. For example, this config supplies a meeting topic and preferred language:
{
"agent_name": "meeting-agent",
"body": {
"meeting_topic": "Product planning",
"language": "en"
}
}For Pipecat Cloud, Recall.ai forwards this object as body in the session-start request. If omitted, Recall.ai sends an empty object. Recall.ai selects the WebRTC transport and starts the session without creating a Daily room.
For self-hosted agents, Recall.ai forwards the object as requestData in the SmallWebRTC offer request. If omitted, the offer has no requestData field.
The example fields are application-defined. Your agent must read and use them to affect its behavior.
Play agent video through the camera or screenshare
If your agent produces video, enable video output in its SmallWebRTC transport and put the Pipecat source under camera. Replace the quickstart's output_media object with:
{
"camera": {
"kind": "pipecat",
"config": {
"agent_name": "meeting-agent"
}
}
}Recall.ai sends meeting audio to the agent, plays its audio into the meeting, and displays its video through the bot's camera. Use screenshare instead of camera to display the video as the bot's screenshare. The agent receives the meeting audio mix; this integration does not send participant video or meeting screenshares to Pipecat.
When output_media.audio is also configured, the meeting audio mix goes only to that audio source. Camera and screenshare sources no longer receive the meeting mix. Use a single Pipecat source under camera or screenshare when one agent should both hear the meeting and produce audio and video.
For an audio agent with an automatic avatar, combine output_media.audio with automatic_video_output, leaving output_media.camera and output_media.screenshare absent.
FAQ
Do I need to configure Pipecat in each Recall.ai workspace?
Yes, if your agent requires authentication. Save the credentials in each Recall.ai workspace and region where you create bots using Pipecat. Self-hosted agents without authentication can connect without saved credentials.
Why does the bot join but my agent does not respond?
Confirm that your agent is running and supports SmallWebRTC with audio input and output enabled. For Cloud, check that agent_name matches the deployment name and that the saved public key belongs to the same Pipecat Cloud account. For self-hosted agents, check that the offer endpoint is reachable and that any saved bearer token is scoped to the exact config.url.
Where should I implement tool calls?
Implement tools on your Pipecat server. Recall.ai does not execute RTVI client-side function calls: it returns an unsupported-function result for valid requests and keeps the session running.
For recording the agent's voice, see Recording meeting agents. For response controls, see the Meeting agents FAQ.
Updated about 1 hour ago
