Thanks — I can help you set up a user_away_timeout so the system detects silence and prompts the user to confirm they’re still there. Below is a concise approach you can follow. I’m responding to your request to mark users as “away” after N seconds of silence and then prompt them. 1) Configure the timeout on the AgentSession • Set

user_away_timeout

to the number of seconds of silence before the user is considered away. Example:

session = AgentSession(
    vad=silero.VAD.load(),
    llm=openai.LLM(model="gpt-4o-mini"),
    stt=deepgram.STT(),
    tts=cartesia.TTS(),
    user_away_timeout=12.5,  # seconds of silence before user is marked "away"
)

2) Listen for the user_state_changed event and prompt when the user becomes away • When you receive

user_state_changed

with

new_state == "away"

, generate a polite prompt asking if they’re still present:

@session.on("user_state_changed")
async def _user_state_changed(ev: UserStateChangedEvent):
    if ev.new_state == "away":
        # ask the user if they're still there
        await session.generate_reply(
            instructions="The user has been silent for a bit. Politely ask if they are still there."
        )

3) Cancel the inactivity prompt if the user speaks again • If state changes from

away

back to

speaking

, cancel any pending inactivity prompt or task so you don’t interrupt the resumed conversation:

# track the pending task somewhere accessible
if ev.new_state == "away":
    inactivity_task = asyncio.create_task(user_presence_task())
    return

# user came back
if inactivity_task is not None:
    inactivity_task.cancel()
    inactivity_task = None

Notes and suggestions • Choose a timeout value that fits your UX (e.g., 10–15s for short prompts, longer for slower interactions). • The prompt content can be simple and friendly: “Are you still there?” or include an offer to reconnect/help if needed. • If you need different behavior (e.g., send a visual UI notification instead of voice, retry prompts, or escalate after repeated no-responses), tell me which option you prefer and I’ll outline the code changes. If you want, share: • the timeout value you prefer, and • whether you want voice, visual, or both types of prompts I’ll provide a tailored snippet for that setup. Sources: How to detect and handle silence during calls | LiveKit