We are seeing an increase of conversations halting...
# supportc
Chris
04/17/2025, 10:31 AMWe are seeing an increase of conversations halting after each function call - this results in a spike in silence-timed-out
It's been happening in multiple different accounts.
We've tried changing the LLM model.
Call IDs:
7bdb4473-779e-4945-93b3-297b0659413f
9fcc9941-3bbb-48c9-b0f0-17efcfd0ded3
bec879e8-5191-421f-93bd-e5bb58aa2fae
a
Askar
04/17/2025, 10:58 AMSame issue here
k
Kings_big💫
04/17/2025, 11:11 AMSorry for the inconvenience, the Devs are looking into it now thanks
Kings_big💫
04/17/2025, 11:11 AMTo help track down this issue, could you share one or two:
- When exactly this happened (the timestamp)
- What response you expected to get
- What response you actually got instead
This would really help us figure out what went wrong!
c
Chris
04/17/2025, 11:16 AMThank you for your response.
It's happening on all calls at the moment.
We are using OpenAI for the LLM - we are testing now with other providers, but we have tried changing OpenAI models and same issue.
I can see it in the Vapi logs, the tools are getting called and the response comes, but there is no further response from the LLM back to the caller.
The voice then triggers the 'are you still there?' response - and after the user prompts something else, the conversation does continue. However, it halts again after each tool call.
Chris
04/17/2025, 11:24 AMWe've just done some testing with Google Gemini also and same issue - c7c347c8-7fc2-4c86-a6e4-8aa1b8e45881
Chris
04/17/2025, 11:33 AMThis is from one example, you can see how the tool was called, then the voice idleMessages plan responds:
10:55:24:133[LOG]
Model called tool: check_availability({...})
10:55:24:136[LOG]
Model request cost (attempt #1, $0.00251205, 16747 prompt, 0 completion)
10:55:24:143[LOG]
Voice cached: Just a moment
10:55:24:179[INFO]
Turn latency: 1132ms (transcriber: 0ms, endpointing: 101ms, kb: N/A, model: 0ms, voice: 0ms)
10:55:24:180[CHECKPOINT]
Assistant speech started
10:55:25:333[CHECKPOINT]
Assistant speech stopped
10:55:37:286[LOG]
Voice cached: hello?
10:55:40:078[CHECKPOINT]
User speech possibly starting
10:55:40:674[LOG]
Transcriber output: Yes.
d
Demux
04/17/2025, 2:40 PMEvery tool call shows the same issue. The webhook results are received by Vapi but no verbal response is being generated. You can ask for the result and only then the conversation continues.
c
Chris
04/17/2025, 5:09 PMWorking now for me 👍