VAPI, what are you doing ? Without any change my costs have exploded 0,16$ from 36s + the call end with end-of-call-report Server URL Error: timeout of 20000ms exceeded + latency when requesting my KB CallId : 15ed0654-6cd6-40af-b83b-070db03d9e37 Everything was ok last week How can I trust you to go to production ? I spend money to test and can't sell this solution to my customer. Any explanation ?

Hey, we will be having an Office Hour in 20 minutes. Can you hop in and get your query resolved during it?

What was the outcome of this? I made some changes and noticed some cost increase, but havent pinned it down.

No news...

Hey Zxdeam, Here's a breakdown of the call costs: Total cost: $0.1595, which includes: \- Speech-to-text (Deepgram Nova-2): $0.0069 for \~0.7 minutes of transcription \- LLM (GPT-4): $0.0763 for processing 14,767 prompt tokens and 163 completion tokens \- Text-to-speech (Eleven Labs): $0.0359 for generating 718 characters of speech \- Voice API processing: $0.0307 for \~0.6 minutes of audio Additional analysis costs using Claude 3.5 Sonnet: \- Summary generation: $0.0022 (300 prompt / 84 completion tokens) \- Structured data extraction: $0.0019 (412 prompt / 46 completion tokens) \- Success evaluation: $0.0056 (1,844 prompt / 4 completion tokens)

@Shubham Bajaj @Sahil I'm also curious about a similar issue for this call log id a91bf8ca-936c-4a5a-a4df-1f3bad524279 It mentions 119363 prompt tokens but I only spoke 3 times and also saw a logged error which mentioned I requested only 20000 tokens. Why is the kb being requested so many times incurring extra costs? To me it looks like there are extra calls to the LLM being made. The one file in my KB shows up many many times in the call log. (look up "This comprehensive guide provides information" and it will show 27 times in the logs when it's only there once in my document). Can you help me understand what's happening?

It is to add context to the model. What you can do is increase the KB threshold to 0.6 or 0.7. When there is a similarity of 60-70%, only then will it attach the content.

@Sahil ok but I'm still not clear why the KB file is showing up 27 times when I've only spoken 3 times and really just asked one question which should have required the KB. How can I justify .50cts/min as a cost?