repeater

title	category	tags	difficulty	description	demonstrates
Repeater	basics	repeater deepgram openai cartesia	beginner	Shows how to create an agent that can repeat what the user says.	Using the `on_user_input_transcribed` event to listen to the user's input Using the `say` method to respond to the user with the same input

This example shows how to build a simple repeater: when the user finishes speaking, the agent says back exactly what it heard by listening to the user_input_transcribed event.

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 and define an AgentServer

Load your .env so the media plugins can authenticate and initialize the AgentServer.

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

load_dotenv()

server = AgentServer()

Prewarm VAD for faster connections

Preload the VAD model once per process to reduce connection latency.

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

server.setup_fnc = prewarm

Define the rtc session with transcript handler

Create the session with interruptions disabled so playback is not cut off mid-echo. Attach a handler to user_input_transcribed; once a transcript is marked final, echo it back with session.say.

@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-5-mini"),
        tts=inference.TTS(model="cartesia/sonic-3", voice="9626c31c-bec5-4cca-baa8-f8ba9e84c8bc"),
        vad=ctx.proc.userdata["vad"],
        allow_interruptions=False,
    )

    @session.on("user_input_transcribed")
    def on_transcript(transcript):
        if transcript.is_final:
            session.say(transcript.transcript)

    await session.start(
        agent=Agent(
            instructions="You are a helpful assistant that repeats what the user says."
        ),
        room=ctx.room
    )
    await ctx.connect()

Run the server

Start the agent server with the CLI runner.

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

Run it

python repeater.py console

How it works

1. The VAD is prewarmed once per process for faster connections.
2. A session-level event emits transcripts as the user speaks.
3. When the transcript is final, the handler calls session.say with the same text.
4. Because interruptions are disabled, the echoed audio plays fully.
5. This pattern is a starting point for building more advanced post-processing on transcripts.

Full example

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

load_dotenv()

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-5-mini"),
        tts=inference.TTS(model="cartesia/sonic-3", voice="9626c31c-bec5-4cca-baa8-f8ba9e84c8bc"),
        vad=ctx.proc.userdata["vad"],
        allow_interruptions=False,
    )

    @session.on("user_input_transcribed")
    def on_transcript(transcript):
        if transcript.is_final:
            session.say(transcript.transcript)

    await session.start(
        agent=Agent(
            instructions="You are a helpful assistant that repeats what the user says."
        ),
        room=ctx.room
    )
    await ctx.connect()

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