@lively-winter-88668 I’m using LiveKit Agents (livekit-agents==1.3.2) + OpenAI Realtime API via: from livekit.plugins import openai model = openai.realtime.RealtimeModel(...) session = AgentSession(llm=model) Everything works — audio in/out, tool calls, VAD, etc. But I’m not receiving any usage metrics: @session.on(“metrics_collected”) def on_metrics(ev): print(“METRICS:“, ev.metrics) This callback never fires. I saw in the docs: “STTMetrics… are only available when an STT component is configured (Realtime APIs do not emit it).” And GitHub issue #2222 says: “Realtime APIs do not emit metrics.” “There is no metric-emitting code in the Realtime implementation.” My questions: Is it expected that metrics_collected does NOT fire at all when using OpenAI Realtime API? (Not for STT/TTS — I mean specifically LLMMetrics for the LLM output.) Does AgentSession currently forward the OpenAI event response.done.usage? Or is this usage data not exposed through the LiveKit API? Is there any supported way to track OpenAI Realtime token usage (input/output) through LiveKit Agents? Or do we need to directly use OpenAI’s WebSocket for cost tracking? Is metrics support planned for Realtime API? Goal: We need token usage for billing & cost tracking of Realtime sessions (greetings, conversations, etc). Thanks — just want to confirm current behavior vs roadmap.

It should be working for realtime: https://github.com/livekit/agents/blob/main/livekit-plugins/livekit-plugins-openai/livekit/plugins/openai/realtime/realtime_model_beta.py#L1469