Thanks — I’m responding to your question about whether LiveKit sends a session state signal for the start and end of tool execution. LiveKit does not send a single session-state signal specifically named “tool execution started” or “tool execution ended.” Instead, it emits detailed events that mark those phases: • When a tool is invoked, LiveKit emits a conversation_item_added event containing a

FunctionCall

item — this serves as the signal that tool execution has started. • When the tool finishes, LiveKit emits another conversation_item_added event containing a

FunctionCallOutput

item — this signals the end of the tool execution and includes the result. You can observe these lifecycle events programmatically (for example, during testing) using assertion helpers like

is_function_call()

and

is_function_call_output()

to detect the start and end of tool runs. If you’d like, I can: • Show example event payloads for

FunctionCall

and

FunctionCallOutput

. • Suggest how to wire these events into your monitoring/logging or agent workflow. Which would help you most right now? Sources: Testing and evaluation | LiveKit Documentation | Agent session | LiveKit Documentation