An intelligent AI assistant that lives in your macOS status bar and can see and control your desktop.
- 🎯 Global Hotkey — Press
⌘+Shift+Spaceto summon/dismiss the assistant - 🖥️ Screen Awareness — Captures screenshots and reads UI via Accessibility API
- 🖱️ Computer Control — Simulates mouse clicks, keyboard input, scrolling, and dragging
- 🧠 Multi-LLM Support — Claude, OpenAI, Gemini, or custom (OpenAI-compatible) endpoints
- 📋 Task Planning — Auto-generates step-by-step plans for complex tasks
- ↩️ Undo Support — Undo the last AI-performed action
⚠️ Safety Confirmations — Prompts for approval before sensitive operations- 💬 Chat Interface — Beautiful glassmorphism UI with image attachment support
- 📎 Image Attachment — Attach screenshots or images to your messages
- 🔄 Background Operation — Runs as a status bar app, no Dock icon
- macOS 14.0 (Sonoma) or later
- Xcode 15.0+
- An API key for at least one LLM provider
cd Zz
# Generate Xcode project (requires xcodegen)
brew install xcodegen
xcodegen generate
# Build
xcodebuild build -project Zz.xcodeproj -scheme Zz \
-configuration Debug -derivedDataPath .build \
CODE_SIGN_IDENTITY="-" CODE_SIGNING_REQUIRED=NO
# Run
open .build/Build/Products/Debug/Zz.app- The app appears in your status bar (sparkles icon ✨)
- Press ⌘+Shift+Space to open the chat panel
- Click the ⚙️ gear icon to open Settings
- Enter your API key for your preferred provider
- Grant Accessibility and Screen Recording permissions when prompted
| Permission | Why | How |
|---|---|---|
| Accessibility | Read UI elements, simulate mouse/keyboard | System Settings → Privacy & Security → Accessibility |
| Screen Recording | Capture screenshots for AI vision | System Settings → Privacy & Security → Screen Recording |
Zz/
├── Models/ # Data models (ChatMessage, TaskPlan, ToolAction, Settings)
├── Views/ # SwiftUI views (ChatView, FloatingPanel, etc.)
├── Services/ # System services (HotkeyManager, ScreenCapture, InputSimulator)
├── Agent/ # AI brain (LLM providers, AgentLoop, SystemPrompts)
└── Utilities/ # Helpers (ImageUtils, KeyCodes)
| Provider | Computer Use | Notes |
|---|---|---|
| Claude (Anthropic) | ✅ Native | Best support via computer_use tool |
| OpenAI (GPT-4o) | ✅ Via function calling | Vision + function calling |
| Gemini (Google) | ✅ Via function calling | Vision + function calling |
| Custom | ✅ Via OpenAI-compatible API | Any compatible endpoint |
- You type a request → "Open Safari and go to google.com"
- AI generates a plan → Step 1: Take screenshot → Step 2: Click Spotlight → ...
- AI executes each step → Screenshots → clicks → types → verifies
- You see progress → Real-time plan updates with step status
- Task completes → AI reports what was done
- Sensitive operations (quit app, delete, destructive commands) trigger a confirmation dialog
- Input control banner warns when AI is controlling mouse/keyboard
- Undo support lets you reverse the last action
- Iteration limit prevents runaway API costs (default: 30 iterations)
MIT
