Intermittent SIP failures with the Realtime API

API
1

Hi everyone,

We have been experiencing intermittent instability with our SIP integration using the Realtime API, with multiple occurrences of normal temporary failure errors.

The issue is not consistent: sometimes the integration works as expected, and other times it fails with the same error, even when using the same project, the same SIP configuration, and the same endpoint.

So far, we have not been able to identify the root cause on our side. We have already opened a support ticket with OpenAI, but we have not yet received feedback that helps us move forward with the investigation.

Is anyone else seeing similar intermittent issues with SIP and the Realtime API?

In particular, I would appreciate any insight on:

  • Whether you are also seeing recurring normal temporary failure errors;

  • Whether the failures are intermittent for the same project, configuration, or endpoint;

  • Whether the issue appears in specific scenarios

  • Which logs, metrics, or SIP traces helped you narrow down the root cause;

  • Any known mitigation, configuration change, or best practice that improved stability.

Any suggestions or shared experiences would be very helpful.

Thanks in advance.

2

Hi @Inobrega, thanks for sharing this. A few similar intermittent SIP/Realtimes cases have been reported recently, especially around “normal temporary failure” responses that appear inconsistently even with unchanged configs.

The tricky part seems to be that the same project and endpoint can succeed one moment and fail the next, which makes it hard to isolate from the customer side alone.

A similar case has already been escalated internally, and the engineering team is taking a closer look at it. In the meantime, collecting full SIP traces, call IDs, response codes, and timing correlations has been the most useful data for narrowing patterns down.

If others here are seeing the same behavior, adding details publicly will probably help connect the dots faster.

-Mark G.