Clicky is a local-first macOS menu bar assistant that lives next to your cursor. It uses push-to-talk voice input, captures screenshots only when you ask it to, answers through a local Ollama model, speaks with the system voice, and can point at UI elements with [POINT:x,y:label:screenN] tags.
This fork removes the old Cloudflare Worker and cloud AI stack. The app now runs with:
- Ollama for local multimodal chat
- WhisperKit for on-device speech-to-text
- Apple Speech as the local fallback transcription path
AVSpeechSynthesizerfor local text-to-speech
- Apple silicon Mac
- macOS 14.2+
- Xcode 16+
- Ollama for macOS installed
gemma4:e4b is the preferred default model, and the Clicky panel can install and load it directly.
- Install Ollama for macOS.
- Open
leanring-buddy.xcodeprojin Xcode. - Select the
leanring-buddyscheme, set your signing team, and run the app from Xcode. - Grant Microphone, Accessibility, Screen Recording, and Screen Content permissions.
- Use the Clicky panel to start Ollama if needed, install the recommended model if none are present, and load the selected model.
For the local setup checklist, see LOCAL_SETUP.md.
For the new runtime architecture, see LOCAL_ARCHITECTURE.md.
- Hold
Control+Optionto talk. - Clicky captures microphone audio locally.
- WhisperKit transcribes the buffered push-to-talk audio on release.
- Clicky captures screenshots when the selected Ollama model supports vision.
- Ollama streams the response back locally.
- The visible response is spoken with the macOS system voice.
- If the response ends with a
[POINT:...]tag, the overlay animates the blue cursor to that location.
- Non-vision Ollama models are still selectable. When you choose one, Clicky falls back to text-only help for that turn and suppresses pointing.
- The app never stores or displays model reasoning or thought traces in conversation history.
- Do not run
xcodebuildfrom the terminal for this project. Run from Xcode so the macOS privacy permissions stay intact.