Repository navigation
Releases
Boulito 0.2.2
Download
Boulito-0.2.2.dmg (below, in Assets) · Mac with Apple Silicon, macOS 14 or later.
- Open the
.dmgand drag Boulito into Applications. - Open Boulito. The first time, macOS blocks it because Boulito is not signed with a paid Apple Developer account: close the message, then go to System Settings → Privacy & Security and click Open Anyway.
- In the Setup window, click Download next to Speech recognition (2.5 GB, once), then pick an AI model.
Already on 0.2.1: Boulito offers this update by itself (notification, then Update in the menu or the Setup window), and permissions are kept.
SHA-256: c3cd7181b5ca3a279182beee98d400cae26b64575cb0285f1154341f83787025
What's new
- One permission request at a time: the Setup buttons show macOS's own window first, and open System Settings only afterwards; Boulito no longer asks for Input Monitoring at launch.
- The update notice is never lost on first launch: Boulito waits until you have answered the notification request (allowed or not) before announcing a new version.
- The Open button next to Notifications opens the settings of the right app.
Boulito 0.2.1
Download
Boulito-0.2.1.dmg (below, in Assets) · Mac with Apple Silicon, macOS 14 or later.
- Open the
.dmgand drag Boulito into Applications. - Open Boulito. The first time, macOS blocks it because Boulito is free and not signed with a paid Apple Developer account: close the message, then go to System Settings → Privacy & Security and click Open Anyway.
- In the Setup window, click Download next to Speech recognition (2.5 GB, once), then pick an AI model.
Coming from 0.2.0: download this version by hand once and grant the permissions again. From now on, updates install themselves from the app.
SHA-256: 4f6ba0703974b0c38d36567d0ff78cb2891c7f1e80beda84eb9755828cc1189e
What's new
- Updates from the app: Boulito asks GitHub once a day whether a new version is out (you can turn it off in the Setup window) and shows a notification. Update in the menu or the Setup window downloads it, checks it and restarts. A new version is installed only if its SHA-256 matches the one GitHub publishes and it is signed by Boulito's release certificate.
- Permissions are kept across updates: the app is now signed with the same certificate on every version, so macOS no longer asks for the microphone, Input Monitoring or Accessibility again after an update.
- The talk key explains what to do when speech recognition is not installed yet, and opens the Setup window.
Boulito 0.2.0
Download
Boulito-0.2.0.dmg (below, in Assets) · Mac with Apple Silicon, macOS 14 or later.
- Open the
.dmgand drag Boulito into Applications. - Open Boulito. The first time, macOS blocks it because Boulito is free and not signed with a paid Apple Developer account: close the message, then go to System Settings → Privacy & Security and click Open Anyway. Needed once per version.
- In the Setup window, click Download next to Speech recognition (2.5 GB, once). Boulito listens as soon as it is installed; then pick an AI model.
The app contains its own Python: no Terminal, no Homebrew. Settings, models and logs are kept in ~/Library/Application Support/Boulito; Erase everything… in the Setup window removes them. The source version (install.sh) still works as before.
SHA-256: 8fa6c8aaaef2ab834ff8f00ace93d49b6c6477a31b9aa816dd3257c56981c83e
What's new
- Download Boulito as a Mac app (.dmg), with speech recognition downloaded from the Setup window and an Erase everything… button.
- Volume to a value: "turn it down to 30", "set the volume to 50".
- Follow-ups: "again", "a bit more", "lower it again", "same"… repeat the previous command, through fast rules or the local AI. In conversation mode (no name needed for a few seconds), the AI can only run simple commands and stays silent otherwise.
- Open listening: the name said alone during conversation mode now waits for the command instead of being dropped, and the name is recognized anywhere in the sentence.
- Lower the sound while listening works again with echo cancellation (it stays optional).
- More misheard commands are recognized without the AI.