feat(vad): bundle optimized silero vad and deprecate the plugin

examples/avatar_agents/audio_wave/agent_worker.py

@@ -18,7 +18,6 @@
 from livekit.agents.voice.avatar import DataStreamAudioOutput
 from livekit.agents.voice.io import PlaybackFinishedEvent, PlaybackStartedEvent
 from livekit.agents.voice.room_io import ATTRIBUTE_PUBLISH_ON_BEHALF
-from livekit.plugins import silero
 
 load_dotenv()
 
@@ -75,7 +74,7 @@ async def entrypoint(ctx: JobContext):
         stt=inference.STT("deepgram/nova-3"),
         llm=inference.LLM("google/gemini-2.5-flash"),
         tts=inference.TTS("cartesia/sonic-3"),
-        vad=silero.VAD.load(),
+        vad=inference.VAD(model="silero"),
         resume_false_interruption=False,
     )
 

examples/avatar_agents/keyframe/agent_worker.py

@@ -15,7 +15,7 @@
     inference,
 )
 from livekit.agents.inference import AudioTurnDetector
-from livekit.plugins import keyframe, silero
+from livekit.plugins import keyframe
 from livekit.plugins.keyframe import Emotion
 
 load_dotenv()
@@ -51,7 +51,7 @@ async def entrypoint(ctx: JobContext):
         llm=inference.LLM("google/gemini-2.5-flash"),
         tts=inference.TTS("cartesia/sonic-3"),
         resume_false_interruption=False,
-        vad=silero.VAD.load(),
+        vad=inference.VAD(model="silero"),
         turn_detection=AudioTurnDetector(),
     )
 

examples/drive-thru/agent.py

@@ -35,7 +35,6 @@
     inference,
 )
 from livekit.agents.inference import AudioTurnDetector
-from livekit.plugins import silero
 
 load_dotenv()
 
@@ -488,7 +487,7 @@ async def drive_thru_agent(ctx: JobContext) -> None:
         llm=inference.LLM("openai/gpt-5-mini"),
         tts=inference.TTS("cartesia/sonic-3", voice="f786b574-daa5-4673-aa0c-cbe3e8534c02"),
         turn_detection=AudioTurnDetector(),
-        vad=silero.VAD.load(),
+        vad=inference.VAD(model="silero"),
         max_tool_steps=10,
     )
 

examples/frontdesk/agent.py

@@ -39,7 +39,6 @@
     tool_use_judge,
 )
 from livekit.agents.inference import AudioTurnDetector
-from livekit.plugins import silero
 
 load_dotenv()
 
@@ -267,7 +266,7 @@ async def frontdesk_agent(ctx: JobContext):
         llm=inference.LLM("google/gemini-2.5-flash"),
         tts=inference.TTS("cartesia/sonic-3", voice="39b376fc-488e-4d0c-8b37-e00b72059fdd"),
         turn_detection=AudioTurnDetector(),
-        vad=silero.VAD.load(),
+        vad=inference.VAD(model="silero"),
         max_tool_steps=1,
     )
 

examples/healthcare/agent.py

@@ -32,7 +32,6 @@
     WarmTransferTask,
 )
 from livekit.agents.llm import ToolError, function_tool
-from livekit.plugins import silero
 
 logger = logging.getLogger("HealthcareAgent")
 
@@ -754,7 +753,7 @@ async def entrypoint(ctx: JobContext):
         stt=inference.STT("deepgram/nova-3", language="multi"),
         llm=inference.LLM("openai/gpt-4.1-mini"),
         tts=inference.TTS("inworld/inworld-tts-1"),
-        vad=silero.VAD.load(),
+        vad=inference.VAD(model="silero"),
         preemptive_generation=True,
     )
 

examples/inference/agent.py

@@ -12,7 +12,6 @@
     cli,
     inference,
 )
-from livekit.plugins import silero
 from livekit.rtc import RpcInvocationData
 
 logger = logging.getLogger("inference")
@@ -59,7 +58,7 @@ async def entrypoint(ctx: JobContext) -> None:
         stt=inference.STT(model=DEFAULT_STT),
         llm=inference.LLM(model=DEFAULT_LLM),
         tts=inference.TTS(model=DEFAULT_TTS),
-        vad=silero.VAD.load(),
+        vad=inference.VAD(model="silero"),
     )
 
     def parse_value(payload: str, fallback: str) -> str:

examples/other/elevenlab_scribe_v2.py

@@ -2,8 +2,8 @@
 
 from dotenv import load_dotenv
 
-from livekit.agents import Agent, AgentServer, AgentSession, JobContext, JobProcess, cli, inference
-from livekit.plugins import elevenlabs, silero
+from livekit.agents import Agent, AgentServer, AgentSession, JobContext, cli, inference
+from livekit.plugins import elevenlabs
 
 logger = logging.getLogger("realtime-scribe-v2")
 logger.setLevel(logging.INFO)
@@ -14,13 +14,6 @@
 server = AgentServer()
 
 
-def prewarm(proc: JobProcess) -> None:
-    proc.userdata["vad"] = silero.VAD.load()
-
-
-server.setup_fnc = prewarm
-
-
 @server.rtc_session()
 async def entrypoint(ctx: JobContext) -> None:
     # Using ElevenLabs STT plugin directly for realtime mode support
@@ -37,7 +30,7 @@ async def entrypoint(ctx: JobContext) -> None:
 
     session: AgentSession = AgentSession(
         allow_interruptions=True,
-        vad=ctx.proc.userdata["vad"],
+        vad=inference.VAD(model="silero"),
         stt=stt,
         llm=inference.LLM("openai/gpt-4.1-mini"),
         tts=inference.TTS("cartesia/sonic-3"),

examples/other/kokoro_tts.py

@@ -2,9 +2,17 @@
 
 from dotenv import load_dotenv
 
-from livekit.agents import Agent, AgentServer, AgentSession, JobContext, JobProcess, cli, metrics
+from livekit.agents import (
+    Agent,
+    AgentServer,
+    AgentSession,
+    JobContext,
+    cli,
+    inference,
+    metrics,
+)
 from livekit.agents.voice import MetricsCollectedEvent
-from livekit.plugins import deepgram, openai, silero
+from livekit.plugins import deepgram, openai
 
 logger = logging.getLogger("kokoro-tts-agent")
 
@@ -27,13 +35,6 @@ def __init__(self) -> None:
 server = AgentServer()
 
 
-def prewarm(proc: JobProcess):
-    proc.userdata["vad"] = silero.VAD.load()
-
-
-server.setup_fnc = prewarm
-
-
 @server.rtc_session()
 async def entrypoint(ctx: JobContext):
     # each log entry will include these fields
@@ -42,7 +43,7 @@ async def entrypoint(ctx: JobContext):
         "user_id": "your user_id",
     }
     session = AgentSession(
-        vad=ctx.proc.userdata["vad"],
+        vad=inference.VAD(model="silero"),
         # any combination of STT, LLM, TTS, or realtime API can be used
         llm=openai.LLM(model="gpt-4.1-mini"),
         stt=deepgram.STT(model="nova-3", language="multi"),

examples/other/transcription/multi-user-transcriber.py

@@ -10,15 +10,13 @@
     AgentSession,
     AutoSubscribe,
     JobContext,
-    JobProcess,
     StopResponse,
     cli,
     inference,
     llm,
     room_io,
     utils,
 )
-from livekit.plugins import silero
 
 load_dotenv()
 
@@ -92,7 +90,7 @@ async def _start_session(self, participant: rtc.RemoteParticipant) -> AgentSessi
             return self._sessions[participant.identity]
 
         session = AgentSession(
-            vad=self.ctx.proc.userdata["vad"],
+            vad=inference.VAD(model="silero"),
         )
         await session.start(
             agent=Transcriber(
@@ -136,11 +134,5 @@ async def cleanup():
     ctx.add_shutdown_callback(cleanup)
 
 
-def prewarm(proc: JobProcess):
-    proc.userdata["vad"] = silero.VAD.load()
-
-
-server.setup_fnc = prewarm
-
 if __name__ == "__main__":
     cli.run_app(server)

examples/other/transcription/transcriber.py

@@ -11,11 +11,12 @@
     MetricsCollectedEvent,
     StopResponse,
     cli,
+    inference,
     llm,
     metrics,
     room_io,
 )
-from livekit.plugins import openai, silero
+from livekit.plugins import openai
 
 load_dotenv()
 
@@ -50,7 +51,7 @@ async def entrypoint(ctx: JobContext):
 
     session = AgentSession(
         # vad is needed for non-streaming STT implementations
-        vad=silero.VAD.load(min_silence_duration=0.3),
+        vad=inference.VAD(model="silero", min_silence_duration=0.3),
     )
 
     @session.on("metrics_collected")

examples/other/transcription/translator.py

@@ -12,12 +12,13 @@
     MetricsCollectedEvent,
     StopResponse,
     cli,
+    inference,
     llm,
     metrics,
     room_io,
     utils,
 )
-from livekit.plugins import openai, silero
+from livekit.plugins import openai
 
 load_dotenv()
 
@@ -76,7 +77,7 @@ async def entrypoint(ctx: JobContext):
 
     session = AgentSession(
         # vad is only needed for non-streaming STT implementations
-        vad=silero.VAD.load(),
+        vad=inference.VAD(model="silero"),
     )
 
     @session.on("metrics_collected")

examples/primitives/echo-agent.py

@@ -9,9 +9,9 @@
     AutoSubscribe,
     JobContext,
     cli,
+    inference,
 )
 from livekit.agents.vad import VADEventType
-from livekit.plugins import silero
 
 load_dotenv()
 logger = logging.getLogger("echo-agent")
@@ -35,7 +35,8 @@ async def entrypoint(ctx: JobContext):
         participant=participant,
         track_source=rtc.TrackSource.SOURCE_MICROPHONE,
     )
-    vad = silero.VAD.load(
+    vad = inference.VAD(
+        model="silero",
         min_speech_duration=0.2,
         min_silence_duration=0.6,
     )

examples/survey/agent.py

@@ -22,7 +22,6 @@
 from livekit.agents.beta.workflows import GetEmailTask, TaskGroup
 from livekit.agents.inference import AudioTurnDetector
 from livekit.agents.llm import function_tool
-from livekit.plugins import silero
 
 logger = logging.getLogger("SurveyAgent")
 
@@ -354,7 +353,7 @@ async def entrypoint(ctx: JobContext):
         llm=inference.LLM("google/gemini-2.5-flash"),
         stt=inference.STT("deepgram/nova-3", language="multi"),
         tts=inference.TTS("inworld/inworld-tts-1"),
-        vad=silero.VAD.load(),
+        vad=inference.VAD(model="silero"),
         turn_detection=AudioTurnDetector(),
         preemptive_generation=True,
     )

examples/telephony/amd.py

@@ -11,12 +11,10 @@
     AgentServer,
     AgentSession,
     JobContext,
-    JobProcess,
     cli,
     inference,
 )
 from livekit.agents.inference import AudioTurnDetector
-from livekit.plugins import silero
 
 logger = logging.getLogger("basic-agent")
 
@@ -38,13 +36,6 @@ def __init__(self) -> None:
 server = AgentServer()
 
 
-def prewarm(proc: JobProcess):
-    proc.userdata["vad"] = silero.VAD.load()
-
-
-server.setup_fnc = prewarm
-
-
 @server.rtc_session()
 async def entrypoint(ctx: JobContext):
     ctx.log_context_fields = {
@@ -55,7 +46,7 @@ async def entrypoint(ctx: JobContext):
         llm=inference.LLM("openai/gpt-4.1-mini"),
         tts=inference.TTS("cartesia/sonic-3", voice="9626c31c-bec5-4cca-baa8-f8ba9e84c8bc"),
         turn_detection=AudioTurnDetector(),
-        vad=ctx.proc.userdata["vad"],
+        vad=inference.VAD(model="silero"),
         preemptive_generation=True,
     )
 

examples/telephony/bank-ivr/ivr_navigator_agent.py

@@ -9,15 +9,13 @@
     AgentServer,
     AgentSession,
     JobContext,
-    JobProcess,
     MetricsCollectedEvent,
     RunContext,
     cli,
     inference,
     metrics,
 )
 from livekit.agents.llm.tool_context import function_tool
-from livekit.plugins import silero
 
 logger = logging.getLogger("phone-tree-agent")
 
@@ -76,13 +74,6 @@ async def record_task_result_and_hang_up(self, context: RunContext, content: str
         context.session.shutdown(drain=True)
 
 
-def prewarm(proc: JobProcess) -> None:
-    proc.userdata["vad"] = silero.VAD.load()
-
-
-server.setup_fnc = prewarm
-
-
 @server.rtc_session(agent_name=PHONE_TREE_AGENT_DISPATCH_NAME)
 async def dtmf_session(ctx: JobContext) -> None:
     await ctx.connect()
@@ -91,7 +82,7 @@ async def dtmf_session(ctx: JobContext) -> None:
     }
 
     session: AgentSession = AgentSession(
-        vad=ctx.proc.userdata["vad"],
+        vad=inference.VAD(model="silero"),
         llm=inference.LLM("openai/gpt-4.1"),
         stt=inference.STT("deepgram/nova-3"),
         tts=inference.TTS("rime/arcana"),