Thanks — you're asking how to ensure the agent only ends the call after finishing any pending speech, and whether to use a tool call for that. Short answer and options: • You do not need tool calling just to wait for speech to finish. Capture the speech handle and await playback, then shut down. ◦ Example (Python):

handle = session.say("Goodbye! Thank you for calling.", allow_interruptions=False)
    await handle  # waits for playout
    await session.shutdown()

• Example (Node):

const handle = session.say('Goodbye!', { allowInterruptions: false });
    await handle.waitForPlayout();
    await ctx.shutdown();

• Use tool calling only if you want the LLM to decide when to end the call (e.g., after a final confirmation). In that case create an

end_call

tool that: a. Calls

ctx.wait_for_playout()

(or

waitForPlayout()

on the handle) to wait for current/queued speech. b. Calls session shutdown. ◦ Example tool (Python):

@function_tool
    async def end_call(self, ctx: RunContext):
        await ctx.wait_for_playout()
        await self.session.shutdown()

Key settings: • Set

allow_interruptions=False

on the final message so it cannot be cut off. • If multiple queued speech items exist, awaiting the handle or

wait_for_playout()

ensures all queued playout completes before shutdown. If you want, tell me whether your code runs inside an LLM tool runner (so I can give the exact tool signature) or share which SDK/language you're using and I’ll provide a concrete snippet for your setup. Sources: Agent speech and audio | LiveKit docs