Skip to content

Latest commit

 

History

56 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 

Repository files navigation

OpenVox app icon

OpenVox

Tap. Speak. In any app.

Build status Latest release

OpenVox Home window: words dictated, time saved, speaking pace, day streak, a words-per-day chart, and recent dictations

Download for macOS

Features

  • Any app. The text lands at your cursor, in Mail, in a terminal, or on a web page.
  • Hold or tap. Hold the key to talk, or tap once and keep your hands free. Escape drops a dictation before it lands.
  • History. Search it, copy from it, and keep it for 7 days, for 30 days, or forever.

Speech modes

Mode Model What it does
Standard Moonshine Streaming Transcribes when you release the key.
Live Nemotron Speech Streaming EN 0.6B Types while you speak, then finalizes.

The app ships without the models. Your mode downloads only what it needs.

Both picks come from a benchmark of seven on-device models on an 8 GB M1. See why these two.

Effort

Standard mode runs one of three Moonshine sizes. Change the level in Settings at any time.

Effort Memory held For
Best ~700 MB The hardest audio.
Balanced ~250 MB Most dictation. The default.
Low ~75 MB The lightest machines. Misspells more.

Live mode runs one model and ignores this setting.

Privacy

Your audio and your dictations stay on your Mac. Setup connects to Hugging Face once, to download the model. History stays in local only.

Requirements

  • macOS 14 Sonoma or later
  • Apple silicon
  • uv or Python 3.10 or later. Setup builds the speech runtime with it.
  • Internet for the first model download

Install

  1. Download OpenVox-*-macOS.dmg from Releases.
  2. Open the disk image and drag OpenVox.app onto the Applications alias.
  3. Open it and follow the setup.
  4. Grant Microphone and Accessibility access.

Builds are ad-hoc signed, so macOS blocks the first launch. It reports the app as damaged and offers only Move to Trash. The app is not damaged. macOS adds a quarantine flag to every download, and it refuses an unsigned app that carries the flag. Remove the flag after you drag the app to Applications:

xattr -dr com.apple.quarantine /Applications/OpenVox.app

Then open the app as usual.

An ad-hoc signature also changes with every build, and macOS then drops the Accessibility grant. Repair it after each update: open System Settings > Privacy & Security > Accessibility, remove OpenVox with the minus button, then add the new app. The toggle alone does not repair the grant.

Build from source

git clone https://github.com/0xAnto/openvox.git
cd openvox/mac
./scripts/build-app.sh
open /Applications/OpenVox.app

build-app.sh installs the app in /Applications and signs it there. A local build carries no quarantine flag, so this path avoids the release-download warning above. Set CI=1 to keep the bundle in the source folder instead.

The script replaces /Applications/OpenVox.app. A local build and a release build use the same bundle identifier, so they also share the settings, the history, and the downloaded runtime in ~/Library/Application Support/OpenVox. Keep a release copy under a different name before you build, or build with CI=1.

Each ad-hoc build gets a new signature, so macOS drops the Accessibility grant. The script runs tccutil reset Accessibility for you. You still grant Accessibility again in System Settings after each build.

Run the self-test to skip the permissions and the downloads:

swift build
.build/debug/OpenVox --selftest

The bare binary has no bundle and no Info.plist, so it gets no microphone prompt and no Accessibility identity. Use --selftest only. Build the app for every other test.

Releases

  1. Merge a pull request into main.
  2. The workflow builds the app, packs the disk image, and signs it when the signing secrets exist.
  3. The workflow publishes OpenVox-v1.0.<run number>-macOS.dmg on Releases. GitHub shows its SHA-256 there.

About

Private, fast, on-device voice dictation for macOS

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Contributors

Languages