Inference still serves models without an EU endpoint when the EU restriction is on

My project is in the EU (Frankfurt) with the Inference region restriction enabled in the dashboard.

To test it, I requested two models that, according to the documentation, have no EU endpoint: deepseek-ai/deepseek-v4.1-flash and fishaudio/s2-pro. Both responded. With the restriction active, I expected both requests to be rejected.

Is the restriction enforced for all LiveKit Inference requests? How can I confirm that STT, LLM and TTS data are processed only in the EU? I need this before handling real customer calls.

How exactly did you make the requests?

The easiest way to test this would be to:

  1. Create an agent with the required models
  2. Deploy that agent to EU
  3. Use agent console to test

Note that just using the “test call” in agent builder is insufficient.