Thanks — I’m responding about your request to add a ctx shutdown callback that prints the entire transcript when using pipeline mode, and the code you posted that “does not work.” Quick diagnosis • The most likely issue is that your callback cannot see the

session

object when it runs (scope/definition timing). The callback must be defined in the same scope where

session

is created so it can read

session.history

. Also make sure the callback is registered before shutdown and that shutdown is actually triggered. Working minimal example • This example shows the pattern that works in pipeline mode: create

session

inside the entrypoint, define the shutdown callback in that same scope so it can access

session.history

, register it with

ctx.add_shutdown_callback

, then start the session.

@server.rtc_session(agent_name="my-agent")
async def entrypoint(ctx: JobContext):
    # create session in this scope
    session = AgentSession(
        stt=inference.STT(model="deepgram/nova-3-general"),
        llm=inference.LLM(model="openai/gpt-4o"),
        tts=inference.TTS(model="cartesia/sonic-3"),
        vad=silero.VAD.load(),
    )

    # shutdown callback defined in same scope so it can access `session`
    async def print_transcript():
        print("\n================ FULL CALL TRANSCRIPT ================\n")
        history = getattr(session, "history", None)
        if history:
            for item in history:
                role = getattr(item, "role", None) or getattr(item, "speaker", "unknown")
                content = getattr(item, "content", None) or getattr(item, "text", "")
                print(f"{role.upper()}: {content}")
        else:
            print("No conversation history available.")
        print("\n================ END TRANSCRIPT =======================\n")

    # register callback
    ctx.add_shutdown_callback(print_transcript)

    # start the session and connect
    await session.start(
        agent=Agent(instructions="You are a helpful assistant."),
        room=ctx.room,
    )
    await ctx.connect()

Notes and best practices • Use

session.history

(public API) to get the full conversation history. • Avoid relying on private/internal attributes like

_chat_ctx

. • The callback can be

async def

(safe even if not awaiting anything). If it needs to write files or call async APIs, keep it async and await those operations. • Register the callback before you call

session.start()

/

ctx.connect()

so it’s in place for shutdown. • Confirm shutdown is happening (add a simple print at top of callback); if you never see that print, the callback may not be invoked. If this still “does not work” Please share any of the following so I can narrow it down: 1. The exact behavior or error (traceback, logs, or “nothing prints”). 2. The entrypoint function (the surrounding code where you create session and register the callback). 3. Whether you see any prints from the callback at shutdown (to confirm it runs). 4. Whether you use

AgentSession

pipeline mode or a custom flow, and the framework version. If you provide one or two of those items I’ll give a tailored fix or point out the exact line that needs changing. Sources: Building multi-agent architectures with LiveKit agents | LiveKit | Transcriber | LiveKit Documentation | Server startup modes | LiveKit Documentation