Assert RMS stays above `min_rms` for `duration` seconds. Raises `AssertionError` if RMS drops below threshold at any sample.

Graph with both fake LLM and StreamWriter for multi-mode testing.

Construct the expected end state by applying `expected_state` SQL to a fresh seed. The agent's DB is compared against this by *state* (see diff_da

non-strict mode tool description see https://serde.rs/enum-representations.html for the internally tagged representation

Close a specific context

Judge that evaluates if responses are coherent and logical. Checks if the agent presents ideas in an organized manner without contradictions

Commit the user turn and generate a reply. Returns a future that resolves with the user's audio transcript once STT is complete and e

Judge that evaluates if responses are appropriately concise. Critical for voice AI where brevity matters. Checks for unnecessary verbosity, r

Resolve the effective run mode. Under pytest-xdist (`-n`) concurrency can't work -- a group shares one event loop and xdist would split it ac

Configure Hamming monitoring export for LiveKit AgentSession. After configuring, attach to each AgentSession via ``hamming.attach_session(session

CPU usage percentage between 0 and 1

Yield until the outbound channel is empty (sender task has drained it).

Remove parameters that are not supported by the given model. Strips any provider prefix (e.g. ``openai/o3-pro`` -> ``o3-pro``) before matchin

Encode a rtc.VideoFrame to a portable image format (JPEG or PNG). See EncodeOptions for more details.

Mark the end of input, no more audio will be pushed

Evaluate a conversation with all judges. Automatically tags the session with results when called within a job context. Args:

Filter out emojis from the text.

Forward one segment's audio/text to the outputs, then wait for its playout. Returns when the segment has fully played, been interrupted, or never

Initialize the event from a cache entry. Args: entry: The cache entry to initialize the event from. is_interruption:

Get MIME type based on output format

Judge that evaluates context preservation across agent handoffs. Handoffs can be silent (seamless) or explicit ("transferring you to..."). Ei

Return True if the given token marks an end or finalized event.

Lookup a registered function by key. Args: key: Unique identifier for the function. Returns: RegisteredFunction: the functio

Create a FunctionCallResult, handling ToolError, StopResponse, and validation.

Test tool 1 Args: arg1: The first argument opt_arg2: The optional second argument

Mock tool in agent Args: arg1: The first argument opt_arg2: The optional second argument

Strictly-increasing wall-clock read off this loop's clock (see _VirtualClock.read).

Notify the sender that playback has started

End interruption inference when agent is speaking and overlap speech ends.