The transcription output (can be also seen from "Transcripts" tab from call logs) drops significant amount of words/sentences even though they were actually present in the "Deepgram final transcript" in the Logs and LLM layer received them too. This is significantly impacting our operations. We have started seeing this today. CallId: 019b8ff8-33ea-7ffb-b1af-8a8b27c2a2e8

Action Step: Go to your Vapi dashboard, navigate to

Observe > Call Logs

, and compare the "Transcripts" tab output with the Deepgram final transcript for the affected call. If you see discrepancies, copy the Call ID and timestamp, then report the issue to Vapi support with a brief description and the evidence. Example report format:

Call ID: 019b8ff8-33ea-7ffb-b1af-8a8b27c2a2e8
Timestamp: [time of issue]
Issue: Words/sentences missing in final transcript but present in Deepgram logs
Expected: All spoken content should appear in the final transcript

This will help the support team quickly investigate and resolve the issue. For more details, see the [Debugging voice agents documentation](https://docs.vapi.ai/debugging). Sources: - [Debugging voice agents documentation](https://docs.vapi.ai/debugging) - [How to Report Issues Effectively](https://docs.vapi.ai/issue-reporting)

Call ID: 019b8ff8-33ea-7ffb-b1af-8a8b27c2a2e8 Timestamp: Jan 05, 13:02:49.910 Issue: Words/sentences missing in final transcript but present in Deepgram logs Expected: All spoken content should appear in the final transcript

Next Step: Your issue report is correctly formatted. Please submit it to Vapi support via the [Support page](https://docs.vapi.ai/support) or your usual support channel. Tip: Attach a screen recording or screenshot of the discrepancy if possible to speed up investigation. Once submitted, monitor your email or the support portal for follow-up from the Vapi team. Sources: - [Debugging voice agents documentation](https://docs.vapi.ai/debugging) - [How to Report Issues Effectively](https://docs.vapi.ai/issue-reporting)

Hi jay, Thanks — I reviewed the call and related docs. To help us investigate quickly, please do the following and paste the results here: 1. In the dashboard [Observe > Call Logs](https://dashboard.vapi.ai/), open Call ID 019b8ff8-33ea-7ffb-b1af-8a8b27c2a2e8 and capture: - The Transcripts tab output (screenshot or copy) - The Deepgram final transcript from the Logs (screenshot or copy) - The exact timestamp(s) where words are missing (you already gave 2026-01-05 13:02:49.910 — please confirm UTC/local) 2. From the same call in the dashboard, copy the call object (GET /call/{id}) or screenshot showing: - endedReason - phoneCallProviderId - artifact (pcapUrl if available) 3. Attach a short screen recording or screenshots that show the discrepancy (Transcripts vs Deepgram logs) and the LLM request/response around the same timestamp. Why this helps: per our debugging guide, we need the dashboard transcript + Deepgram logs + timestamps to determine whether the missing text occurred during transcription ingestion, post-processing, or later in Vapi’s pipeline (see https://docs.vapi.ai/debugging).