I have an agent set up and seemingly working OK. Except for the fact that sometimes - actually very often, probably 40% of the time - it doesn't respond when people answer its questions. It's not a noise issue. And I've tried with and without background noise reduction. I can say "YES" very clearly and loudly and it says nothing. I'll repeat and nothing. I'll then say "CAN'T YOU HEAR ME!? I SAID YES!!!" and finally it might respond, apologising. And it's happening to multiple people on our team who are testing it, with different voices, responses, phones, environments, so it's not just me. Is there a checklist of things that might be causing this kind of behaviour? It only seems to happen on actual calls it makes to a phone, not when testing in the browser. I'm using Google Gemini 2.0 Flash for model and transcriber - this is the only setup where it doesn't just randomly hang up on me.

I understand how frustrating it can be when your agent doesn't consistently respond, especially when you've already ruled out noise issues and have tested across different users and environments. Since this only happens during actual phone calls and not in browser testing, it does suggest the issue might be related to the telephony integration or how the audio is being processed over phone lines. While I don't have a specific checklist available in the information I have, your detailed description of the problem and the steps you've already taken are really helpful for narrowing down potential causes. If you notice any patterns—such as certain types of responses being missed more often, or issues occurring with specific phone carriers or devices—those details could be useful for further troubleshooting.

Send your call id

2a5201c2-14c3-406c-b9e8-09a3ea3700d8

Hey, Could you please schedule a brief call at your convenience so we can discuss this matter in detail? Kindly use the following link to book a suitable time: .

Nothiing available until the end of next month!

The assistant's delayed response is due to end pointing, which determines the duration to wait before recognizing the user has finished speaking. - End Pointing Plan Configuration: You need to adjust the Smart End Pointing plan provider to livekit and update the wait function within the Start Speaking Plan for English Transcription. - Wait Function Optimization: Iterative testing with potential users is necessary to determine the optimal wait function values, as suggested values may not be suitable for all scenarios.

startSpeakingPlan": {
  "waitSeconds": 0.4,
  "smartEndpointingPlan": {
          "provider": "livekit",
          "waitFunction": "2000 / (1 + exp(-10 * (x - 0.5)))"
        }
}

Also I will suggest you to update the assistant transcriber to deepgram instead for better results.