Plugins
LiveWire provides stub classes under the plugins and inference namespaces that
mirror popular livekit-agents plugin constructors so that existing code compiles
unchanged. On SignalWire, the platform handles all STT, TTS, LLM, and VAD
infrastructure automatically. These classes are no-ops that log an informational
message once on first instantiation.
You do not need to install Deepgram, Cartesia, ElevenLabs, Silero, or any other provider SDK. SignalWire manages the entire AI pipeline in its control plane. These stubs exist solely so that imports and constructor calls from existing livekit-agents code do not raise errors.
STT Plugins
plugins.DeepgramSTT
Stub for the LiveKit Deepgram STT plugin.
Accepts arbitrary constructor arguments for API compatibility. All are ignored.
LLM Plugins
plugins.OpenAILLM
Stub for the LiveKit OpenAI LLM plugin. The constructor accepts arbitrary options for API compatibility.
All constructor arguments are accepted for API compatibility. To set the LLM model
on the SignalWire agent, pass the model string to the
AgentSession constructor’s llm option instead.
TTS Plugins
plugins.CartesiaTTS
Stub for the LiveKit Cartesia TTS plugin.
Accepts arbitrary constructor arguments for API compatibility. All are ignored.
plugins.ElevenLabsTTS
Stub for the LiveKit ElevenLabs TTS plugin.
Accepts arbitrary constructor arguments for API compatibility. All are ignored.
VAD Plugins
plugins.SileroVAD
Stub for the LiveKit Silero VAD plugin.
load
Static factory method that mirrors SileroVAD.load() from the LiveKit SDK.
Returns a new SileroVAD instance. Logs an informational no-op message on
first call.
Inference Stubs
The inference namespace mirrors the livekit.agents.inference module. These
classes are available as named exports from the inference namespace.
inference.STT
Stub for livekit.agents.inference.STT.
model
Model identifier. Accepted for API compatibility.
Accepts an optional second argument for additional options. All are ignored.
inference.LLM
Stub for livekit.agents.inference.LLM. This class is a no-op — the constructor
accepts but does not store its arguments. To set the LLM model on the SignalWire
agent, pass a plain string (e.g., 'gpt-4o') to AgentSession({ llm })
instead of an inference.LLM instance.
model
Model identifier. A provider prefix (e.g., "openai/") is stripped automatically
by the session when building the underlying SignalWire agent.
Accepts an optional second argument for additional options. All are ignored.
inference.TTS
Stub for livekit.agents.inference.TTS.
model
Model identifier. Accepted for API compatibility.
Accepts an optional second argument for additional options. All are ignored.