# Realtime model over sip: Advice for logging agent audio packets

**URL:** <https://community.openai.com/t/realtime-model-over-sip-advice-for-logging-agent-audio-packets/1375861>\
**Category:** API\
**Tags:** realtime, api-realtime, sip\
**Created:** [March 6, 2026, 10:30am UTC](https://community.openai.com/t/realtime-model-over-sip-advice-for-logging-agent-audio-packets/1375861 "2026-03-06T10:30:21Z")\
**Posts on this page:** 4\
**Page:** 1

<div class="post-metadata">

**Author:** ![Noah\_D](https://sea2.discourse-cdn.com/openai1/user_avatar/community.openai.com/noah_d/32/727673_2.png) [@Noah\_D](https://community.openai.com/u/Noah_D)\
**Post date:** [March 6, 2026, 10:30am UTC](https://community.openai.com/t/realtime-model-over-sip-advice-for-logging-agent-audio-packets/1375861/1 "2026-03-06T10:30:21Z")

</div>

Hey all,  
First time posting.  
I’ve set up my realtime agent and it works fine with sip. However, I have a requirement that the calls need to be recorded, or at least have some playback capability.  
I thought the agent was going to be logging audio packets and subsequently i set up a redis streams message broker service to handle these. As it turns out, i can only seem to get the agent audio packets on interrupted agent utterances. I can get user ones just fine currently.

---

<div class="post-metadata">

**Author:** ![jochenschultz](https://sea2.discourse-cdn.com/openai1/user_avatar/community.openai.com/jochenschultz/32/772274_2.png) [@jochenschultz](https://community.openai.com/u/jochenschultz)\
**Post date:** [March 7, 2026, 3:38am UTC](https://community.openai.com/t/realtime-model-over-sip-advice-for-logging-agent-audio-packets/1375861/2 "2026-03-07T03:38:36Z")

</div>

There are multiple options.. you can let the sip provider (twillio?) do the recording I suppose. As far as I remember they have two parallel channels so that should be possible.

Or you add a SBC/Media Proxy between OpenAI and the provider and do the recording over there.

---

<div class="post-metadata">

**Author:** ![Noah\_D](https://sea2.discourse-cdn.com/openai1/user_avatar/community.openai.com/noah_d/32/727673_2.png) [@Noah\_D](https://community.openai.com/u/Noah_D)\
**Post date:** [March 9, 2026, 10:22am UTC](https://community.openai.com/t/realtime-model-over-sip-advice-for-logging-agent-audio-packets/1375861/3 "2026-03-09T10:22:08Z")

</div>

Ahh, cheers for the response! I was hoping there was a way of transcoding audio packets so i get two birds with one stone with my log saving application. Its seeming like more of a headache than getting the media proxy up between them.

---

<div class="post-metadata">

**Author:** ![jochenschultz](https://sea2.discourse-cdn.com/openai1/user_avatar/community.openai.com/jochenschultz/32/772274_2.png) [@jochenschultz](https://community.openai.com/u/jochenschultz)\
**Post date:** [March 9, 2026, 3:38pm UTC](https://community.openai.com/t/realtime-model-over-sip-advice-for-logging-agent-audio-packets/1375861/4 "2026-03-09T15:38:35Z")

</div>

Hmm well, there is whisper and many other STT models.
