Called by the runner to push audio frames to the generator.

Reclaim focus for the agent and notify all participants.

Release a previously reserved slot. Called automatically by acquire_context() on success, or manually on failure.

Judge that evaluates if responses are relevant and on-topic. Checks if the agent directly addresses what the user asked, stays focused on the

Resume the audio playback

The overlap window if another confirmed booking holds this booking's room.

Judge that evaluates if responses are safe, compliant, and appropriate. Checks for unauthorized advice, improper disclosure, failure to escalate,

Send synthesis content to the connection

Set the tracer provider for the livekit-agents. Args: tracer_provider (TracerProvider): The tracer provider to set. metadata (dic

Mark the session as successful. Args: reason: Optional reason for the success (stored separately from the tag).

Synthesize text to speech using chunked stream.

Judge that evaluates if the agent completed its goal based on its instructions. Extracts the agent's instructions from AgentConfigUpdate items in

Judge that evaluates if the agent used tools correctly. Checks tool selection, parameter accuracy, output interpretation, and error handling.

Return tuples of (loan_id, amount, due_date).

Updates the agent's instructions. If the agent is running in realtime mode, this method also updates the instructions for th

Update LLM configuration options. Each option is read on the next ``chat()`` call, so a swap takes effect on the agent's next turn wi

Update the options for the agent session. Args: endpointing_opts (NotGivenOr[EndpointingOptions], optional): Endpointing

Validates card number via the Luhn algorithm

Wait for an `EventEmitter` event to fire. Returns the event payload as a tuple (since `EventEmitter` callbacks may receive multiple positiona

Waits for the sound to finish playing.

FallbackAdapter is an LLM that can fallback to a different LLM if the current LLM fails. Args: llm (list[LLM]): List of LLM insta