OmniScribe Documentation

Private voice dictation that works in every app. 100% local. No telemetry.

OmniScribe is the easiest way to add voice dictation to your Linux workflow. It runs entirely on your machine — your voice never leaves your device. Pro tier works fully offline.

How it works

OmniScribe listens to your microphone, transcribes speech to text locally using Whisper models, and injects the text into whichever text field you're focused on. No cloud. No round-trip. Words appear as fast as you speak.

1
Click any text field

Browser, terminal, IDE, Slack — anything that accepts text input.

2
Press your hotkey

Default is F9. Configurable in ~/.config/omniscribe/config.toml.

3
Speak

OmniScribe transcribes in real time. Press hotkey again to stop and inject the text.

Polish modes

OmniScribe can clean and format your dictation before injecting it. Choose a mode with a different hotkey:

ModeHotkeyWhat it does
VoiceF9Cleans grammar, removes filler words, adds punctuation
NotesF10Structures into Markdown bullet points
EmailF11Formats as a professional email

BYO-LLM

Bring your own LLM keys for the polish modes. Supports OpenAI, Anthropic, and local models. You control which model processes your text, how much it costs, and where the data goes. We never see your LLM requests.

Requirements

  • Linux (X11 or Wayland)
  • ydotool for text injection (sudo apt install ydotool)
  • ~1 GB disk space for the STT model
  • 4 GB RAM recommended
  • Pro tier: no internet required after initial activation

Ready to start?

Go to Quickstart →