Feature request: typed user turns mid-session on /api/chat
Author: apostololegCreated May 9, 2026Updated May 9, 2026
Labelsquestion
Due diligence
- I have done my due diligence in trying to find the answer myself.
Topic
The paper
Question
Today text_prompt looks connect-time only for persona; live client traffic is audio (0x01 + Opus).
Please support user text after connect (same WebSocket), e.g. an extra binary frame kind for UTF-8 user turns, so sessions can alternate speech and typing without reconnecting or piping text through TTS.
Short scope note (upstream Moshi vs PersonaPlex-only vs won’t do) would help integrators a lot.
Source: NVIDIA/personaplex