Even with

modelOutputInMessagesEnabled = true

the next request going to the model uses the transcription value for assistant instead of the actual model output. Call ID:

019bafbd-33e8-7ee0-bd6b-593a46343ecf

Log with id:

log-26

notice the spelling of the last name Log with id:

log-61

notice how the conversation history now has a different version of the last name spelling (from transcriber)

Hi there, Thank you for your message. Our team is currently out of the office. We operate Monday through Friday, from 9:00 AM to 8:00 PM Pacific Standard Time (PST). We’ll get back to you as soon as possible during our normal business hours. If your message is urgent, please mark it accordingly or include “URGENT” in the subject line, and we’ll do our best to respond promptly. Warm regards, Vapi Customer Support Team

Thanks for sharing the logs, this is a common VAPI issue and I see exactly why it’s causing confusion. Even with modelOutputInMessagesEnabled = true, the transcriber text can still overwrite the assistant message in history, which is why the name spelling changes on the next turn. The key is to decouple transcription from conversation memory and make sure only the raw model output is stored as the assistant message. Are you using VAPI’s default memory handling, or have you customized how messages are saved between turns? @Pat

> The key is to decouple transcription from conversation memory and make sure only the raw model output is stored as the assistant message. Can you point me to the documentation on this please?

Hey Pat, reported your issue to the product team, there seem to be an issue. While this takes time to be fixed, I would like to reccomened few things to improve your call quality- 1. Try another voice provider, or reduce the generationConfig.speed to 1. The voice doesn't play the words correctly; the speed argument might fix that, but if don't try using a different provider. 2. use better transcriber, like deepgram nova 3 for English.