Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

26 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

NAVA

Local-first AI voice dictation for Linux (X11). Hold a key, speak, and polished text is injected into whatever app has focus. Clean-room — no code, assets, or branding from any other product.

Project site, install guide, and downloads.

Quickstart (Debian / Ubuntu / Mint)

Download the latest .deb from Releases, then:

wget https://github.com/arasheslamii/nava/releases/download/v0.0.1/nava_0.0.1_all.deb
sudo apt install ./nava_0.0.1_all.deb     # installs system deps + a venv in /opt/nava
nava setup                                 # pick model / hotkey / cues; seeds your dictionary
nava start                                 # background daemon: hold Right-Ctrl, speak, release
nava enable                                # optional: auto-start on login

X11 only. Wayland isn't supported yet — switch to an X11 session to use NAVA for now.

Other distros / from source / AppImage: see docs/INSTALL.md.

Commands

Command What it does
nava dictate foreground: hold Right-Ctrl → speak → release → inject
nava start / stop / status background daemon (systemd --user)
nava enable auto-start on login
nava paste-last re-inject the most recent transcript (bind to a DE shortcut)
nava setup / config TUI wizard / config editor
nava doctor diagnostics (X11, xdotool, clipboard, mic, hotkey, ASR)
echo hi | nava-inject low-level text injection

How it works

  • Injection (X11): XTEST typing → clipboard-paste → notify, with automatic per-app paste policy and Unicode-safe routing.
  • ASR: faster-whisper small.en int8 on CPU (~0.45 RTF on a 2015 i7), Silero VAD + anti-hallucination params (silence → empty). medium.en available for more accuracy.
  • Formatting: filler removal + custom-dictionary proper-noun correction (Tier-1, ~0 ms).
  • UX: terminal-first (Rich + questionary), headless daemon, no GUI.

Privacy

Local by default, no telemetry, audio never written to disk. Only the latest transcript is cached (for paste-last); disable with [history] keep_last = false.

License

GPL-3.0


See PLAN.md, ROADMAP.md, DECISIONS.md for the design history.

About

The voice dictation app Linux was missing. Hold-to-talk, speak, and clean text appears anywhere — powered by Whisper, 100% local and offline. No cloud, no telemetry.

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages