First public release of Vocium — a desktop AI voice-input tool whose core is exposed over a single MCP protocol.
Highlights
- Floating pill with five visual states; never steals focus
- Toggle / push-to-talk with a customizable global hotkey
- Multi-provider cloud STT — Groq, OpenAI Whisper, Gemini (BYOK, local-only keys)
- AI polish before injection (basic / conversational / custom), prompt-injection guarded
- Chinese output — force Traditional or Simplified (offline OpenCC)
- VAD silence trimming (optional)
- MCP-native — standalone server;
transcribe_clip,inject_text,polish_textfor external hosts - Tray health panel, bilingual UI (zh-TW / en), multi-microphone selection
- Windows + macOS, from source or packaged installers
Build from source — see the README. Full notes in CHANGELOG.md.