Realtime support for OpenAI Agents SDK

What problem do you want to solve?

There is currently no OpenTelemetry instrumentation for OpenAI's RealtimeAgent for the openai-agents-sdk. The existing opentelemetry-instrumentation-openai-v2 covers the REST-based Chat Completions API, but the Realtime API has a fundamentally different event-driven lifecycle (persistent sessions, streamed audio deltas, server-side VAD, real-time function calling) that the current instrumentation cannot capture.

Without instrumentation, developers building real-time voice agents have no standardized way to observe session latency, token consumption, tool execution, or error rates — all of which are critical for production reliability.

Describe the solution you'd like

I'd like suggest to extend the opentelemetry-instrumentation-openai-v2 package to also cover the RealtimeAgent use-cases. One way would be to provides a bare-minimum, semantic-convention-compliant listener for the OpenAI Realtime API, which would result in a trace similar to the below example trace:

Describe alternatives you've considered

No response

Additional Context

No response

Would you like to implement a fix?

Yes

Tip

_{React with 👍 to help prioritize this issue. Please use comments to provide useful context, avoiding +1 or me too, to help us triage it. Learn more here.}

Start by reading the opentelemetry-instrumentation-openai-v2 package and its existing REST-based Chat Completions coverage. Compare that instrumentation with the OpenAI RealtimeAgent lifecycle described here, including persistent sessions, streamed audio deltas, server-side VAD, and real-time function calling. Done means a minimal semantic-convention-compliant listener that exposes session latency, token consumption, tool execution, and error rates.