Whisper Local
Offline push-to-talk AI dictation that types speech at the cursor on Windows and macOS.
Tested on macOS
Use it
macOS arm64
You need first
- Runtimes:
python>=3.11 - A GPU is optional
Install whisper-local via uv with Python 3.13
uv tool install --python 3.13 whisper-local
Where the README and reality differ
- README: pipx install whisper-local / uv tool install whisper-local on Python 3.11+
What worked: uv tool install --python 3.13 whisper-local (uv auto-downloads CPython 3.13.18); default Homebrew Python 3.15 fails resolving ctranslate2==4.7.1 with no cp315 wheels
Check that it works
~/.local/bin/whisper-local --versionThe output should contain “whisper-local 0.23.0”.
What it does
Good for
- dictate text in any application via push-to-talk hotkey
- run hotkeys or shell commands by voice trigger
- translate spoken language to English
An alternative to Wispr Flow, Dragon and Otter.
How far pickfit got
macOS arm64
- Installs
- Starts
- Not used yet
- Works offline
Windows: the project says it runs there; pickfit hasn’t tried it.
How pickfit tested it
Each run starts from a clean machine and follows the steps above. How pickfit verifies
macOS arm64
Installs and starts
- Replayed from scratch
- Works offline
- Checked on 1 Nov 2026
- Version 0.23.0
Differences the test machine hit
- README: Install the macOS build (cask or .dmg/.zip) and open the .app
In the test: No Homebrew cask and no macOS .dmg/.zip exist (release v0.23.0 has only whisper-local.exe + PyPI wheel/sdist); on macOS the app is the Python package installed via uv/pipx, not a ~/Applications .app - README: whisper-local runs the tray app / first-launch welcome window
In the test: Full app launch needs a microphone and Accessibility permission; this VM has neither, so plain launch exits 1 with sounddevice.PortAudioError: Error querying device -1 then a secondary AttributeError in shutdown, and --settings/--cheat-sheet stay alive with no window - README: whisper-local --doctor Exit 0 = clean
In the test: whisper-local --doctor exits 1 when it finds issues (here: Audio device probe X and Accessibility permission X); README says Exit 0 = clean, so exit 1 is the expected diagnostic signal, not a tool crash