Vapi is the Voice AI platform for developers.

Vapi AI

callId: `857ee7d8-bd90-4e19-9f2a-244d72c1f770`, `ab27a924-32e8-4989-9bbc-9219ee2a2321`
assistantId: `af74c1cf-8e4c-4f04-b226-0a1db2602cb8`

When using the .send() (https://docs.vapi.ai/sdk/web#send) functionality via the Web SDK, the assistant is not recognizing and responding to the input, despite the call transcript showing the text coming in from the user and consuming the input tokens. I am passing the `user` messages to the assistant per the docs:

`vapi.send({
  type: "add-message",
  message: {
    role: "user",
    content: "What kind of features do you offer?",
  },
});`

This appears to be an OpenAI Realtime specific issue within VAPI - all of the non-realtime OpenAI models correctly respond to `user` messages.

Is there possibly something wrong with my assistant configuration or is this a known issue with the VAPI Web SDK?

Were you able to figure out what's going on here?

I've forwarded it to the executives waiting for feedback from them, we appreciate your patience

Hey <@693639816235057243>, this looks like a bug given how the normal approach functions compared to real-time OpenAI.


Could you create a bug report for it? https://roadmap.vapi.ai/bug-reports

<@319034765808369666> <@1006897684965244938> Thanks guys I'll create a ticket and I'll run through https://github.com/VapiAI/client-sdk-web and see if I can't get you guys a PR early next week to take a peek at. This is pretty high priority for my product so I'd like to make it as easy as possible for you guys to check a bug off the list. I'm sure someone else out there has ran into this issue anyways

<@693639816235057243> likely won't be able to offer much help with this since not many people use OpenAi Realtime. Resolving this issue may take longer than usual. So, would suggest you to take a look at another approach via normal cluster.

<@319034765808369666> Ok that's interesting I've been using their realtime model exclusively since that's been giving me the best results as far as latency and conversation quality (most human-like). Can you recommend another configuration that has similar/better performance? I am prioritizing the highest quality here over cost - and this is for web calls strictly

<@319034765808369666> <@1006897684965244938> Any feedback here?

It is gpt-4o, 11labs model v2, smartEndpointing = Vapi and transcriber deepgram nova-2

If I used the openai-realtime-preview model for voice, why is transcribe so slow over webrtc with openai 4o-mini selected?

This ticket has been marked as solved, and as a result, you will no longer receive any further responses. Kindly create a new support ticket to escalate your inquiry.&nbsp;