You can configure a multi-voice agent framework in LiveKit to include an agent that only performs function calls and does not produce TTS or any audio output. This is supported through the

RoomOutputOptions

configuration, where you can set

audio_enabled=False

. When set this way, the agent will not publish any audio to the room and will only send text responses (if needed) to the

lk.transcription

text stream topic, without speech synchronization or TTS output. This setup is ideal for orchestrator or monitoring agents that coordinate workflows, execute function calls, or route tasks between other agents, without ever speaking or producing audio. LiveKit Agents 1.0 is designed to support these multi-agent orchestration patterns, allowing you to build flexible agent workflows where only selected agents handle voice interactions. Example configuration:

await session.start(
    room=ctx.room,
    room_output_options=RoomOutputOptions(
        audio_enabled=False,
    ),
    # ... agent, room_input_options, etc....
)

This approach lets you create backend coordination agents or workflow orchestrators that never produce audio output, while other agents in the system handle user-facing voice interactions. Sources: Virtual avatar integrations | LiveKit Docs | Building the all-in-one platform for voice AI agents | README.md | Text and transcriptions | LiveKit Docs