The Agents SDK combined with the Realtime API is an incredibly powerful toolkit. When integrated with the Twilio adapter, it enables highly capable voice and phone agents. Among its features, Traces serve a particularly meaningful purpose — they enhance transparency, accountability, and facilitate debugging.
However, the current Trace Dashboard in the OpenAI API Platform is insufficient for many real-world use cases. In most scenarios, agentic workflows built with this SDK are offered as part of a platform or service to third-party users. This means that end users — the actual customers of the developer’s platform — cannot access their traces through OpenAI’s dashboard.
This limitation has already been highlighted in the following issues:
- We need a way to show traces to third parties #56
- Add Trace Export or Self-Hosted Tracing Option #590
Fortunately, custom tracing processors provide a way for developers to store traces in their own persistent storage and display them through a custom UI. This works seamlessly with standard agents — until the Realtime API comes into play.
Currently, there is no way to persist or retrieve traces when using the Realtime API. Custom tracing processors are not triggered, and there is no alternative method to fetch or export trace data. Similar concerns have been raised by other developers:
To deliver a complete and transparent experience for end users, developers need a way to collect and store Realtime API traces in their own persistent storage — similar to how it works with standard agent tracing.