Hey Vapi team, This is an urgent recurring issue, I see you've responded to other people after me, so I'm sending this message again, mind you, we've been on the platform for a year and a half, we've done over 30k calls on the platform and our whole business runs on Vapi. We've been on Vapi for about a year and a half and run a real estate sales training platform at salesgo.ai. For the last two weeks we've been seeing a consistent, high-volume bug across nearly all of our calls and need to escalate this urgently. The core issue: user speech is not being picked up by the transcriber during live calls, even though the audio is clearly captured in the recording. The assistant gets no input, so it doesn't respond, leaving users waiting 10-15 seconds until they say something like "Hello?" or "Are you still there?", at which point that phrase gets picked up and the AI responds to that instead of anything the user actually said. This is breaking the core experience of our product and is happening on scale. Yesterday alone we had 120 calls and this occurred on almost every single one. I couldn't find a formal bug reporting channel, so I'm reaching out directly. Please prioritize this -- we're a startup and our users are experiencing this in real time. Here are specific timestamped examples separate calls: https://docs.google.com/document/d/1yJBoNXlbWexUN9w280DC30WV_72R9DdXxf0arevF4Ts/edit?usp=sharing These are just a couple examples of the issue, but this happens on almost every call. Please keep me updated on this, our business relies on this and this puts us in the position where we might have to migrate off of Vapi if this isn't resolved.

That honestly sounds brutal, especially at the volume you’re handling. Since the audio is still present in the recordings, the issue is likely happening in the live transcription/streaming flow during the call, which is why the assistant misses the user’s first response and only reacts later. I’d start by checking the transcriber behavior, latency, endpointing/VAD settings, and whether this is tied to a specific provider or call flow. I can help you troubleshoot this directly and narrow down exactly where the breakdown is happening. Are you seeing this across every assistant/transcriber setup, or mainly with one provider/model? @Jack Stein

Yeah, this was happening after every call. We ended up going through all of it, latency, endpointing, VAD settings, provider config, and the issue traced back to how we were organizing the transcriber on the database side. Once we fixed that, things started behaving correctly. Still want to run a bit more testing before I fully close it out, but I think we're in good shape.