Repository navigation
v0.1.0, First functional release
First working end-to-end version of ParlAR, validated on real hardware (Fedora, RTX 2050, X11).
Features
- System-wide voice dictation via xdotool/wtype/ydotool injection
- Local transcription with faster-whisper (CUDA or CPU)
- Two modes: utterance-based and real-time streaming (LocalAgreement-2)
- Spanish-first text processing with proper punctuation handling
- Voice commands: new paragraph, delete last sentence, send, stop
- Rule-based or Ollama-powered rewrite modes (formal, concise, email)
Status
Early but functional. No GUI beyond a minimal overlay indicator yet, and the output system is not decoupled. See the README roadmap for what's next.