Hi, I’m building a voice assistant using Vapi and I’m seeing a mismatch between the audio responses and the text shown in the conversation history. From my understanding, the pipeline is: User speech → transcribed to text (STT) Transcription sent to the LLM LLM generates a text response That response is converted to audio (TTS) The audio responses sound correct and high quality, but the text I see in the conversation/messages appears inaccurate or degraded. It seems like the text being stored or displayed is not the raw LLM output, but rather a transcription of the spoken assistant response (or some post-processed version of it). What I would like instead is: Access to the raw model output text Or to have the conversation history reflect the LLM response directly, not a transcribed version of the audio My questions are: Is the text shown in messages/conversation history a transcription of the assistant’s speech rather than the LLM output? How can I configure Vapi to return or store the raw model output instead (e.g. via model-output events or a setting like modelOutputInMessages)? What is the recommended way to ensure the displayed text matches exactly what the model generated?