Skip to main content
Beta: digital avatars in meetings are in beta. Instead of a plain roster entry, a bot running a Telnyx AI Assistant can appear on camera as that assistant’s avatar. Telnyx renders the avatar — lip-synced video and audio — from the assistant’s own conversation, and the bot streams it as its camera. This suits branded, customer-facing, or demo meetings where a personified presence matters. The avatar is configured on the assistant, not on the meeting session. Add avatar_settings to the assistant first — see Configure an avatar, including its voice requirements. Then opt the session in with assistant.avatar:
The request carries no avatar ID or key. At creation the service reads the assistant and rejects the request if it has no Telnyx avatar configured, before any bot is sent to the meeting. The rest of the assistant parameters work as usual.

How It Works

  • The assistant owns the avatar’s speech. Only the assistant’s conversation is rendered, so speak and stop_speaking return 409 invalid_state, and speak_on_enter is rejected with 400 invalid_request — the assistant greets through its avatar instead. chat_on_enter works as usual.
  • Startup takes a few seconds. The bot shows an Avatar starting frame until the avatar’s video is playing; avatar_state then becomes connected and the assistant’s greeting starts. Read avatar_state rather than assuming the camera is ready.
  • Interruption follows the assistant. With audio_gate: "full_duplex", the assistant’s own interruption handling stops the avatar when a participant genuinely interrupts it; short backchannels such as “mm-hmm” do not cut it off. With half_duplex, the assistant does not hear the room while the avatar speaks, as without an avatar.
  • Immediate REST creates only. join_at is not supported, avatars are not available through calendar auto-join or the MCP server, and avatar mode is chosen at creation and cannot be toggled mid-meeting.
  • Platforms. Google Meet, Zoom, Teams, and Webex. Other meeting URLs are rejected with 422 unsupported_capability.
  • Recording. The avatar’s camera video is not included in the recording; the bot’s audio is.
  • Camera image. The avatar takes precedence over a static camera_image: the image is ignored and its URL is not fetched.

Avatar State

The session response carries the avatar captured from the assistant at creation, and its state:
avatar_state is starting, connected, degraded, or disconnected, and each change is recorded as an avatar.connected, avatar.degraded, or avatar.disconnected session event. All three fields are null on a session without an avatar.

Voice-only Fallback

If the avatar cannot start, or stops working mid-meeting, the session continues voice-only: the bot shows a Voice only frame and plays the assistant’s audio itself.
  • Before the avatar is live, the assistant’s conversation restarts once without the avatar. A second failure fails the assistant as it would without an avatar.
  • After the avatar is live, the remaining speech switches to voice-only without interrupting the conversation. The utterance in flight is not replayed.
Either way avatar_state becomes degraded, and the avatar.degraded event carries one of these reasons: Telnyx ends the avatar shortly before an hour into the session; a longer meeting continues voice-only with telnyx_avatar_ended.

Troubleshooting