How to reduce first response time?
# support
m
In most of our calls, it takes AI a bit more time to respond at start. It takes 3-4 seconds (or more if there is a tool call), but over time responses are much faster, as conversation lasts longer. Is there any way to make the first response faster? Here is an ID for example 019bc268-f237-711b-932c-fdf5e0abe74e
r
I'm experiencing the same issue. did you get this resolved?
m
No, still waiting for a response on how to improve this
c
Hi MILANCHeJ, To speed up the initial response time, try these changes: 1. Config fixes: • Set startSpeakingPlan.waitSeconds to 0.2. • Disable smartEndpointingPlan. • Minimize llmRequestDelaySeconds and responseDelaySeconds to 0. • Lower silenceTimeoutSeconds. • Check numFastTurns settings. 2. Model adjustments: • Use a faster model like gpt-4.1-mini if quality allows. • Enable optimizeStreamingLatency on the voice provider. 3. Transcription/audio path: • Switch to a faster transcriber. • Make sure mic permissions are granted immediately. 4. Operational tips: • Avoid long customerJoinTimeoutSeconds for phone flows. • Pre-warm components for high-volume flows. • Delay heavy tool work if possible. 5. What I found on your call (ID: 019bc268-f237-711b-932c-fdf5e0abe74e): • There was a ~3–4s delay due to endpointing and model overhead. Let me know if you want more details. If you share your current assistant config settings, I can recommend specific values.
m
Thanks Praveen. When you say model overhead, do you mean the prompt was too large?