I'm constantly having problems with not getting latency data from VAPI. This is extremely annoying with my production calls as it makes all our dashboards completely incorrect. For example: 019eb860-1755-7001-8e3f-40c68baf6d43 This call says 0ms latency for the LLM which of course we know isn't the case. I don't understand what I can do to make VAPI give me the actual latency measurements consistently.

Hi Hugo, The zero values you're seeing on tool-call turns are expected behavior. Vapi doesn't capture modelLatency for turns where the assistant invokes a tool, so those entries will always report 0. This applies to all assistants using tools, not just yours. For per-turn latency on tool-call turns, the most complete source is the raw JSONL log available via artifact.logUrl in your end-of-call report, that file contains the underlying checkpoint events at a more granular level than the turnLatencies summary. Regards, Chiranjeet, Vapi Support

Why is this something that isn't captured?

Hi Hugo, Capturing latency on tool-call turns is something our team has flagged as a feature to be built. We don't have a timeline to share yet, but it's on the roadmap. In the meantime, the raw JSONL log via artifact.logUrl in your end-of-call report has the underlying checkpoint events at a more granular level if that helps bridge the gap. Regards, Chiranjeet, Vapi Support