To set up a passive agent that does not output audio in LiveKit, you can configure the

RoomOutputOptions

with

audio_enabled=False

when starting your agent session. Here’s an example:

session = AgentSession(
    llm=openai.realtime.RealtimeModel(),
    # Configure other components as needed
)

await session.start(
    agent=Agent(instructions="Your agent instructions here"),
    room=ctx.room,
    room_output_options=RoomOutputOptions(
        audio_enabled=False,  # Disable audio output
        transcription_enabled=True,  # Keep transcriptions enabled if you want text output
    ),
)

This configuration will allow your agent to process inputs and generate responses, but it will only publish text to the

lk.transcription

text stream topic without producing any audio output in the room. The agent will still be able to receive and process audio input from participants if STT (Speech-to-Text) is configured. Sources: Text and transcriptions | LiveKit Docs | Tree View: | dse-workbench