v1.6.7 v0.2.12

medium risk

cheaper-inference/OmniRouteAug 5, 2026 → Aug 5, 20260 days between releases

What's New

Adds WebRTC support, Smart Turn v3.2, and direct audio input for LLMs

Breaking Changes

  • Renamed raw PCM WebSocket mode from `--mode websocket` to `--mode raw-websocket`

New Features

  • WebRTC transport for OpenAI Realtime API via `[webrtc]` extra
  • Direct audio input for audio-capable LLMs via `--stt none`
  • Optional remote-LLM proxy functionality
  • Mic and speaker device selectors in browser demo
  • Qwen3-TTS GGML features including quantization and local file loading

Improvements

  • Smarter endpointing with Smart Turn v3.2 using quantized CPU model
  • Improved OpenAI Realtime protocol behavior
  • Significantly improved browser demo

Upgrade Notes

Update the `--mode` flag to `raw-websocket` and install the `webrtc` extra to use the new transport.

The Two Releases

v1.6.7

FluidVoice v1.6.7

Aug 5, 2026· 3 months ago

Performance is significantly improved for Apple Silicon devices, with Parakeet transcription running up to 2x faster. The release also reduces startup latency and fixes several microphone selection and media playback issues.

Full release notes
v0.2.12

v0.2.12

Aug 5, 2026· 3 months ago

The final planned release in the 0.2.x line focusing on smarter turn-taking, WebRTC support, direct audio input for LLMs, and an improved browser demo.

Full release notes