Interruption handling model region/plan restrictions

Getting Started
,
1

I’m trying to call wss://agent-gateway.livekit.cloud/v1/bargein for adaptive interruption handling on the Build plan
can you tell me which region/plan does serve it?

Currently I am getting
RuntimeError: interruption gateway error 2008: service might be unavailable in this region

2

Can you tell me more about your setup:

Are your agents being hosted in the LiveKit cloud or you are self hosting them?

I’m trying to call wss://agent-gateway.livekit.cloud/v1/bargein for adaptive interruption handling

What do you mean by you are trying to call that? How are you calling it?

RuntimeError: interruption gateway error 2008: service might be unavailable in this region

What region are you in?

3

While @CWilson confirms theplan/regiongating, you can unblock today with the local turn-detector model (no gateway dependency):

  from livekit.plugins.turn_detector.multilingual import MultilingualModel
  from livekit.plugins import silero

  session = AgentSession(
      turn_handling=TurnHandlingOptions(turn_detection=MultilingualModel()),
      vad=silero.VAD.load(),
  )

Different model from Bargain (contextual on transcripts, not audio-based barge-in), but gets you turn detection working today without the gateway.

4

@CWilson Thanks for the quick response.

  1. Hosting: Not on LiveKit Cloud. I’m running locally for testing — a python script on my machine, no lk agent deploy.

  2. Calling directly over WebSocket to wss://agent-gateway.livekit.cloud/v1/bargein, doing the same thing livekit-agents’ InterruptionWebSocketStream does:

  • JWT signed with my LIVEKIT_API_KEY / LIVEKIT_API_SECRET and InferenceGrants(perform=True)
  • session.create with sample_rate=16000, num_channels=1, encoding=s16le
  • Stream 100 ms s16le PCM chunks

The connection opens fine, then the gateway returns {“type”:“error”,“code”:2008,“message”:“service
might be unavailable in this region”}. I get the same error running the official livekit-agents worker
with TurnHandlingOptions(interruption={“mode”:“adaptive”}).

  1. Region: I’m in Armenia (Yerevan). The Build-plan dashboard doesn’t show a region field, so I don’t
    know which one my project is pinned to.

Questions:

  1. Is /bargein available on the Build plan?
  2. If yes, which regions serve it, and can my project be moved?
  3. If no, which plan includes it?

Thanks!

5

The Barge-in model is only available for LiveKit cloud-hosted agents. You should be able to test it locally for a limited time, but there is no way to increase that limit once it’s used up. Looking here, it is included in the build plan, Conversational Intelligence, but again for agents hosted in the LiveKit cloud.

Once your limit is reached for self hosted agents it will fall back to VAD transparently. You will see an error message in the agent logs when this happens.

6

@Grigor_Shahverdyan, Now that @CWilson has clarified the gating, three paths from here:

  • Deploy to Cloud Agents for full /bargein access on Build plan (lk agent deploy).
  • Stay self-hosted, accept the transparent VAD fallback once your trial limit hits, per CWilson’s note. Zero code change.
  • Stay self-hosted with MultilingualModel (snippet I posted above). Transcript-context turn detection, no gateway dependency, no trial limit. Different model class from Bargein, but a meaningful step up from raw VAD.

If Cloud Agents is on your roadmap anyway, that’s the closest to your original intent.