LiveKit Agents

In this catalog it is the Voice Agents option when you need to turn speech into structured agent turns and back. Keep LiveKit Agents on the observe/propose side of the loop: useful signals and drafts, with human or policy authority still required before anything irreversible runs.

LiveKit Agents run realtime voice/multimodal agents on WebRTC with turn-taking and in-call tools.

1. Connect a LiveKit room/session.
2. Audio streams in; STT produces text turns.
3. Agent/LLM may call tools mid-session.
4. TTS streams audio back to participants.

Related: pipecat · whisper · topics/03-channels

• Use when: LiveKit Agents builds realtime voice (and multimodal) agents on LiveKit’s WebRTC transport: low-latency audio, turn-taking, and tool calls during live sessions matches the bottleneck and you specifically need to turn speech into structured agent turns and back.
• Skip when: another pack already covers the need with less ops surface, or LiveKit Agents would become a second place people retype truth.
• Prefer instead: compare Pipecat, Whisper before locking the cell.

• Ops: confirm license, hosting, and release cadence for LiveKit Agents (livekit/agents) before relying on it in a live corridor.
• Integration: bad credentials, scopes, or schema drift produce empty retrievals or runaway tool/model spend without an obvious UI alarm.
• Scope: Voice Agents components support agents; they are not the company ledger or the approval authority.