Calls

1
2
3
4
5
6

{
  "title": "Calls",
  "audience": "everyone",
  "summary": "Audio, video, screen share, recording, and live transcription — inside any room.",
  "sort_order": 8
}

Calls#

Every room is call-capable. Click the phone (audio) or camera (video) icon in the chat header to start one. Other room members get a ringing notification with Answer / Reject buttons.

Stack#

ScaiWave uses LiveKit as the SFU. Each call gets its own room on the LiveKit server, with a one-to-one mapping between the ScaiWave room and the LiveKit call session. Tokens are short-lived; ScaiWave mints fresh ones per join.

Features#

• Audio and video with adaptive bitrate.
• Screen share — any participant can share at any time. The client picks a single source (display, window, tab) per share session.
• Picture-in-Picture (PiP) — pop the call into a floating window that stays on top while you keep working in ScaiWave. Chromium-only (uses the Document Picture-in-Picture API); falls back gracefully elsewhere.
• Recording — admins can start a recording on any active call. Recordings land in MinIO/S3 and show up as media attachments in the room. Start with POST /v1/calls/{call_id}/recording/start.
• Live transcription — recordings are transcribed asynchronously via ScaiGrid's audio model and posted as a transcript message in the room.

Permissions#

• Start a call: every room member.
• Start a recording: requires power_level >= 50 (default).
• End a call for everyone: the call initiator or any member with power_level >= 50.

Calls and federation#

Calls work across federated rooms — the LiveKit server is your local one, and federated peers receive call invites the same way they receive messages. Token minting happens locally; the foreign participant gets a LiveKit token signed by your server.

AI voice calls#

When you start a call in a room with an AI participant, the call goes active immediately (no ringing) and a voice bot worker joins LiveKit as the AI. This gives you a live voice conversation with the AI — full tool calling, context from the chat history, and on/off-the-record control.

See Voice mode for the full architecture.

Where to go next#

• Tutorial: Talk to your AI
• Concepts: Voice mode
• API: Calls
• API: Voice