Hey HN,

  I built OpenTypeless because macOS Dictation and Windows Voice Typing both lack AI polishing — you
  get raw, messy transcription with no way to fix it.

  How it works:
  - Hold a global hotkey → speak → release
  - STT provider transcribes (Whisper, Groq, Deepgram, etc.)
  - LLM polishes the text (GPT, Claude, Gemini, Ollama...)
  - Result is typed directly into whatever app is focused

  A few things I think HN would appreciate:
  - Fully BYOK — no backend, your API keys stay local
  - Ollama support for completely offline use
  - Translation mode: speak in Chinese, output in English
  - Cross-platform: Windows, Mac, Linux (Tauri)
  - MIT license

  It started as a personal tool for reducing typing fatigue. Now it supports 6+ STT providers and a
  dozen LLMs.

  GitHub: https://github.com/tover0314-w/opentypeless

  Happy to answer any questions about the architecture or design decisions.