v0.26.1 v1.0.0
high risk
facebook/astryxSep 6, 2026 → Sep 6, 20260 days between releases
What's New
Major release introduces new commands and speech backends while removing deprecated flags.
Breaking Changes
- Removed deprecated --mode realtime and --mode local flags.
- Removed socket and raw-websocket transport modes.
- Removed redundant extras: facebook-mms, language-detection, websocket.
- Removed --mac-optimal-settings preset.
- Default hosted LLM changed to gpt-5.6-terra.
New Features
- Qwen3-ASR speech recognition backend.
- OpenAI-compatible STT and TTS endpoints.
- Stateful OpenAI Realtime transcription.
- Experimental vLLM Realtime transcription.
- Optional OmniVoice and Supertonic TTS backends.
Improvements
- More reliable realtime conversations with ordered output.
- Updated Apple Silicon setup with MLX dependencies and optimized Qwen3 models.
- Clearer setup with opt-in transcript logging and improved diagnostics.
Upgrade Notes
Replace deprecated --mode flags with new serve, talk, or local commands. Remove socket/raw-websocket transport modes and redundant extras. Update default hosted LLM configuration to gpt-5.6-terra.
The Two Releases
v0.26.1
v0.26.1
Sep 6, 2026· 2 months ago
Docker image release fixing accounting revision activation.
Full release notesv1.0.0
v1.0.0
Sep 6, 2026· 2 months ago
First major release introducing a single Realtime engine, explicit server and microphone-client commands, expanded speech backends including Qwen3-ASR and OpenAI-compatible endpoints, and improved conversation lifecycle handling.
Full release notes