Deepgram

Stream meeting audio to a Deepgram Voice Agent and play its responses back into the meeting.

Recall.ai connects to Deepgram's Voice Agent API directly. Configure the agent's speech recognition, language model, and voice in the bot request. Deepgram runs the conversation while Recall.ai handles meeting audio and playback.

Setup

1. Get a Deepgram API key

Create an API key in the Deepgram Console for a project with access to Voice Agent. This integration uses Deepgram's US cloud Voice Agent endpoint.

2. Add your credentials to Recall.ai

Open Agents → Deepgram in the Recall region where you create bots:

Select the workspace used by your Recall API key, then save your Deepgram key. This updates the workspace's default Deepgram credential, which is also used by Deepgram transcription. Changes apply to new connections.

If the default credential uses an EU or self-hosted Deepgram endpoint, use Manage saved Deepgram credentials to select a US cloud credential. Creating the bot in another Recall region does not change the Deepgram endpoint used by this integration.

Quickstart

Call Create Bot with your meeting URL and agent settings.

We recommend setting the relevant meeting platform in variant to web_4_core when pairing this audio agent with a video provider or a kind: "webpage" source. See bot sizing guidance.

{
  "meeting_url": "MEETING_URL",
  "output_media": {
    "audio": {
      "kind": "deepgram",
      "config": {
        "agent": {
          "language": "en",
          "listen": {
            "provider": {
              "type": "deepgram",
              "model": "nova-3"
            }
          },
          "think": {
            "provider": {
              "type": "open_ai",
              "model": "gpt-4o-mini"
            },
            "prompt": "You are a helpful meeting assistant. Keep spoken replies brief."
          },
          "speak": {
            "provider": {
              "type": "deepgram",
              "model": "aura-2-thalia-en"
            }
          },
          "greeting": "Hi, I'm here to help with the meeting."
        }
      }
    }
  }
}

Once the bot joins, Recall.ai connects to Deepgram, sends the agent settings and meeting audio mix, and plays the agent's replies into the meeting.

For recording the agent's voice, see Recording meeting agents. For response controls, see the Meeting agents FAQ.

Additional configurations

The output_media.audio.config object supports:

FieldRequiredDescription
agentYesThe agent object from Deepgram's Settings configuration reference. Must include listen, think, and speak objects.
credential_idNoUUID of a saved Deepgram credential in the same Recall workspace. Defaults to the workspace's default Deepgram credential. The selected credential must use Deepgram US cloud.

The API key is read from saved credentials. api_key is not accepted in this config.

Configure the agent

Recall.ai sends agent as the agent field of Deepgram's Settings message, adding bot context headers to server-side function endpoints as described below. Supply that object directly; Recall.ai creates the surrounding message and manages audio formats.

Common fields inside agent include:

FieldDescription
listenSpeech recognition provider and model settings. Required by Recall.ai.
thinkLanguage model provider, prompt, and optional function definitions. Required by Recall.ai.
speakSpeech synthesis provider and voice settings. Required by Recall.ai.
languageLanguage setting supported by Deepgram.
greetingOptional message the agent speaks when the session starts.

See Configure the Voice Agent for provider-specific options. Recall.ai sends 16 kHz linear PCM in both directions. Settings outside the agent object, such as top-level audio or tags, are not exposed by this integration.

Use Recall bot context in tools

Recall.ai automatically adds these headers to each server-side function endpoint in agent.think.functions[].endpoint.headers. Deepgram includes them when calling your tool server:

HeaderValue
X-Recall-Bot-IdThe Recall bot's ID, matching the ID returned by Create Bot.
X-Recall-Api-Base-UrlThe bot's regional Recall API base URL, without an API version path, such as https://us-west-2.recall.ai.

Your tool server can use these values to send chat messages, start a screenshare, or make the bot leave. It still needs your Recall API key to authenticate those API requests; the context contains no Recall credentials.

These header names are reserved. Recall.ai replaces supplied values regardless of capitalization and preserves other headers, including your tool's authentication headers. The context is added only to server-side function endpoints; client-side functions and the model, speech recognition, and speech synthesis endpoints are unchanged.

Append the endpoint's versioned path to this base URL when making Recall API requests.

Add a camera or avatar

Deepgram is supported under output_media.audio. You can combine it with an independent camera or screenshare source. When audio is configured, the meeting audio mix goes only to the audio agent; camera and screenshare sources do not receive it.

To use 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 create a saved Deepgram agent first?

No. This integration uses the inline agent settings in each request. It does not accept a saved agent ID or a complete Settings message.

Where should I implement tool calls?

Configure server-side functions in your Deepgram agent settings so Deepgram calls your server. Recall.ai does not execute client-side functions: it returns an unsupported-function result and emits a customer-visible warning while the session continues.

To add chat messages, screensharing, and a leave action, see Add meeting tools.


Did this page help you?