exit_message

title	category	tags	difficulty	description	demonstrates
Exit Message	basics	exit message deepgram openai cartesia	beginner	Shows how to use the on_exit method to take an action when the agent exits.	Using the on_exit method to take an action when the agent exits Using function tools to end sessions gracefully

This example demonstrates how to add a clean shutdown flow to an agent. The agent exposes a function tool to end the session and speaks a goodbye message from on_exit.

Prerequisites

• Add a .env in this directory with your LiveKit credentials:

  LIVEKIT_URL=your_livekit_url
  LIVEKIT_API_KEY=your_api_key
  LIVEKIT_API_SECRET=your_api_secret
  

• Install dependencies:

  pip install "livekit-agents[silero]" python-dotenv

Load environment, logging, and define an AgentServer

Start by importing the required modules and setting up logging to watch the session lifecycle. The AgentServer wraps your application and manages the worker lifecycle.

import logging
from dotenv import load_dotenv
from livekit.agents import JobContext, JobProcess, Agent, AgentSession, AgentServer, cli, inference, function_tool
from livekit.plugins import silero

load_dotenv()

logger = logging.getLogger("exit-message")
logger.setLevel(logging.INFO)

server = AgentServer()

Prewarm VAD for faster connections

Preload the VAD model once per process. This runs before any sessions start and stores the VAD instance in proc.userdata so it can be reused, cutting down on connection latency.

def prewarm(proc: JobProcess):
    proc.userdata["vad"] = silero.VAD.load()

server.setup_fnc = prewarm

Define an agent with an end-session tool

The agent is instructed to call a function tool when the user wants to end the conversation. The end_session tool drains any pending work from the session and then closes the media pipelines gracefully.

class GoodbyeAgent(Agent):
    def __init__(self) -> None:
        super().__init__(
            instructions="""
                You are a helpful agent.
                When the user wants to stop talking to you, use the end_session function to close the session.
            """
        )

    @function_tool
    async def end_session(self):
        """When the user wants to stop talking to you, use this function to close the session."""
        await self.session.drain()
        await self.session.aclose()

Say goodbye on exit

Use the on_exit hook to speak a final message after the session closes. This runs even when the tool initiated the close, giving you a chance to say goodbye before the connection ends.

    async def on_exit(self):
        await self.session.say("Goodbye!")

Define the RTC session entrypoint

Create an AgentSession with STT, LLM, TTS, and VAD configuration. Start the session with the agent and connect to the room. When the user asks to hang up, the LLM will invoke end_session to close the call gracefully.

@server.rtc_session()
async def entrypoint(ctx: JobContext):
    ctx.log_context_fields = {"room": ctx.room.name}

    session = AgentSession(
        stt=inference.STT(model="deepgram/nova-3-general"),
        llm=inference.LLM(model="openai/gpt-4.1-mini"),
        tts=inference.TTS(model="cartesia/sonic-3", voice="9626c31c-bec5-4cca-baa8-f8ba9e84c8bc"),
        vad=ctx.proc.userdata["vad"],
        preemptive_generation=True,
    )

    await session.start(agent=GoodbyeAgent(), room=ctx.room)
    await ctx.connect()

Run the server

The cli.run_app() function starts the agent server and manages connections to LiveKit.

if __name__ == "__main__":
    cli.run_app(server)

Run it

python exit_message.py console

How it works

1. A function tool ends the session when the user asks to hang up.
2. The agent drains outstanding work, closes the session, and then runs on_exit.
3. on_exit plays a final TTS message before the connection ends.

Full example

import logging
from dotenv import load_dotenv
from livekit.agents import JobContext, JobProcess, Agent, AgentSession, AgentServer, cli, inference, function_tool
from livekit.plugins import silero

load_dotenv()

logger = logging.getLogger("exit-message")
logger.setLevel(logging.INFO)


class GoodbyeAgent(Agent):
    def __init__(self) -> None:
        super().__init__(
            instructions="""
                You are a helpful agent.
                When the user wants to stop talking to you, use the end_session function to close the session.
            """
        )

    @function_tool
    async def end_session(self):
        """When the user wants to stop talking to you, use this function to close the session."""
        await self.session.drain()
        await self.session.aclose()

    async def on_exit(self):
        await self.session.say("Goodbye!")


server = AgentServer()


def prewarm(proc: JobProcess):
    proc.userdata["vad"] = silero.VAD.load()


server.setup_fnc = prewarm


@server.rtc_session()
async def entrypoint(ctx: JobContext):
    ctx.log_context_fields = {"room": ctx.room.name}

    session = AgentSession(
        stt=inference.STT(model="deepgram/nova-3-general"),
        llm=inference.LLM(model="openai/gpt-4.1-mini"),
        tts=inference.TTS(model="cartesia/sonic-3", voice="9626c31c-bec5-4cca-baa8-f8ba9e84c8bc"),
        vad=ctx.proc.userdata["vad"],
        preemptive_generation=True,
    )

    await session.start(agent=GoodbyeAgent(), room=ctx.room)
    await ctx.connect()


if __name__ == "__main__":
    cli.run_app(server)