Hey there! To tackle the race condition in your voice agent prototype—where the chat context isn’t updating properly in the

on_user_turn_completed

callback—here’s a friendly step-by-step guide: 1. Use the Temporary Context: It’s best to work with the

turn_ctx

parameter that’s passed into the callback, instead of changing the main chat context directly. This way, your changes stay nicely contained within the current turn. 2. Atomic Updates: When you want to save changes, make sure all your updates to

turn_ctx

are done before you call

await self.update_chat_ctx(turn_ctx)

. This keeps everything tidy and consistent. 3. Handle Exceptions: Since the framework catches exceptions in

on_user_turn_completed

, it’s a good idea to write your logic so it handles errors smoothly—this helps avoid any unexpected hiccups. 4. StopResponse for Early Termination: If you need to stop generation early, especially when the user’s message isn’t complete or valid, using

StopResponse

is a safe way to do that. Putting these tips together with LiveKit’s handy features like turn interruption detection and locks for updating the chat context should help you sort out the issue. If you want me to walk you through any part or share some examples, just let me know—I’m here to help! Sources: Voice pipeline nodes | LiveKit Docs | livekit-agents/livekit/agents/voice/agent_activity.py