Fully local recognition
NVIDIA Parakeet v3 downloads once and is cached on your PC. Works offline forever after.
Local voice typing for Windows & Linux
Hold a key. Speak.
It types for you.
ShadowWhispr turns your voice into text anywhere on Windows and Linux. Speech recognition runs locally with NVIDIA Parakeet v3 — no audio leaves your PC unless you ask it to.
Windows may show “Unknown publisher”
The official ShadowWhispr installer does not contain viruses or malware. This warning appears because the project does not use a paid code-signing certificate—not because Windows detected a threat. Download only from the official GitHub release linked here; every release also includes a SHA-256 checksum.
Two small setup steps on Linux
Prefer the Arch package: build it from packaging/arch/PKGBUILD in the repo (makepkg -si), or just unpack the tarball and run ./shadowwhispr. Either way, run sudo usermod -aG input $USER once and log back in so ShadowWhispr can see your hotkey and type for you. On GNOME, add the AppIndicator extension for the tray icon (Ubuntu ships it already).
Three seconds of setup, zero thought after
- 1Hold your hotkey
Any combo you like — even F13–F24.
- 2Speak
Transcribed on your own GPU, offline.
- 3Release
Text lands wherever your cursor is.
Built to disappear into your workflow
Types anywhere
Chat, docs, email, code — text appears in whatever input is focused. No copy-pasting.
Optional AI cleanup
Polish rough dictation with your existing Claude, ChatGPT/Codex, or Antigravity login. Off by default.
Your hotkey, your rules
Click the hotkey field and press any key or modifier combination to change it.
Privacy isn’t a toggle. It’s the default.
- Audio goes to a temp WAV file, deleted right after transcription.
- Raw speech stays local unless AI cleanup is enabled.
- Credentials live in each provider’s official CLI — ShadowWhispr never reads or stores them.
Requirements
Windows 10 / 11
Linux (Arch package, X11 & Wayland)
NVIDIA GPU required
No runtime to install
No Python needed