v3.36.1 v1.0.0
ScaleX-IO/uGDSSep 5, 2026 → Sep 6, 20260 days between releases
What's New
Major release introduces unified Realtime engine and new speech backends
Breaking Changes
- Renamed `speech-to-speech` to `speech-to-speech serve`
- Deprecated `--mode realtime` and `--mode local` flags
- Removed `--local_mac_optimal_settings` flag
- Removed `socket` and `raw-websocket` modes
New Features
- Added explicit commands `serve`, `talk`, and `local`
- Added Qwen3-ASR speech recognition support
- Added WebRTC transport support
- Added OmniVoice and Supertonic TTS backends
- Added opt-in transcript logging
Improvements
- Improved realtime conversation reliability with ordered output
- Updated Apple Silicon setup with 4-bit Qwen3-4B-Instruct LLM
- Enhanced setup and diagnostics documentation
Upgrade Notes
Update command syntax by replacing `speech-to-speech` with `speech-to-speech serve` and migrating deprecated flags to new explicit commands like `serve` and `local`. Remove obsolete socket or raw-websocket modes in favor of the new Realtime WebSocket or WebRTC API.
The Two Releases
v3.36.1
Sep 5, 2026· 2 months ago
This patch release focuses on performance optimizations and bug fixes. It improves the efficiency of random number generation and resolves issues related to command execution and proxy server connectivity with Xray.
Full release notesv1.0.0
Sep 6, 2026· 2 months ago
The first major release introduces a unified Realtime engine with explicit commands (`serve`, `talk`, `local`), expanded speech backend support including Qwen3-ASR and OpenAI-compatible endpoints, and significant improvements to conversation reliability and Apple Silicon setup.
Full release notes