Start a new response generation.

Helper method to implement synthesize() using stream() for TTS providers that only support streaming inference. This creates a stream

Update an item with the same ID if it exists, otherwise append it.

Build the outer WS message dict expected by ``_handle_transcript_data``. Default shape mirrors a Saaras v3 streaming final-transcript chunk.

Clear the buffer, stopping playback immediately

Assert existence of a function call event matching criteria. Args: name (str, optional): Function name to match.

When timeout isn't None, it will raise asyncio.TimeoutError if the processes didn't finish in time.

Mark the end of the current segment

Force callback to be called with current total if non-zero

Get the default inference URL based on the environment. Priority: 1. LIVEKIT_INFERENCE_URL if set 2. If LIVEKIT_URL contains '.staging.li

Given a word, returns a list of pieces, broken at the possible hyphenation points.

forcefully kill the supervised process

Pre-warm connection to the LLM service

the number of elements queued (unread) in the channel buffer

Yield a function that starts an `AgentSession` for a given agent. Handles LLM lifecycle, the `_TASK_TRANSITION_DELAY` after `start()`, and a

Initialize the MCP server connection and fetch available tools. If the MCP server is not yet connected, this will call ``MCPServer.in

Wait until the current activity is idle and return it. Re-targets on handoff. Raises ``ActivityClosedError`` if the session is closing,

Schedule filler speech while a long-running step blocks the tool. While the context is open, a background scheduler waits for the session to

Build a _TTSOptions with sensible test defaults, overriding as needed.

Create a new instance of xAI STT. Args: enable_interim_results (bool, optional): Whether to return interim (non-final) transcript

Create a new instance of OpenAI LLM. ``api_key`` must be set to your OpenAI API key, either using the argument or by setting the

All audio publications the user sees from the agent.

Map each tool to its owning executor: AsyncToolset tools route to that toolset's own executor; everything else falls back to ``default``.

How many times the caller has spoken so far - the signal for whether a booking flow was actually driven by the caller or silently re-run by the mo

Detach reusable resources for handoff to *new_activity*.

Disable VAD interruption after the backchannel boundary expires.

Flush held transcripts. When ``force`` is True, all buffered events are emitted unconditionally; this is used during interruption-det

Interrupt the current speech or generation, and optionally ignore the user transcript until the given timestamp. Args: i