Thanks — I’m responding to your question about the

interrupted

flag you’re seeing in llm_node-related events and why you can’t find it on user events. Short answer • The

interrupted

flag indicates a speech segment was cut off mid-utterance because the other party began speaking. • It is a property of speech objects/events (e.g.,

SpeechHandle

,

ChatMessageEvent

) rather than the llm_node metric itself. • The flag applies to both assistant and user speech: for assistant events it will be true when the user interrupts the assistant; for user events it will be true when the assistant interrupts the user. Why you might not see it on user events • Examples and docs commonly show the agent/assistant being checked for

interrupted

, so it’s often demonstrated from the assistant side. • It’s possible the export, logging pipeline, or metric view you’re using does not surface the

interrupted

field for user events even if the system tracks it internally. • The most reliable way to read interruption state for assistant speech is via the

SpeechHandle

returned by

session.say()

or

session.generate_reply()

and calling

wait_for_playout()

— that gives you direct access to the

interrupted

property. Next steps I can help with • If you want, tell me how you’re ingesting or viewing events (SDK stream, logs, metrics export, or a specific dashboard). I can suggest where to look or what to enable so user-event interruption flags are exposed. • If you can share a small anonymized event JSON where you expected

interrupted

but don’t see it, I can review it and point out whether the field is present or being dropped. Reference • How to detect when an agent has finished speaking: https://livekit.io/field-guides/guide/detecting-agent-finished-speaking Would you like me to walk through the exact place in your event stream to check, or help with a snippet to read

interrupted

from

SpeechHandle

? Sources: How to detect when an agent has finished speaking | LiveKit