macOS-native local AI trained exclusively on Apple platform knowledge.
A small 3B model with deep macOS expertise rather than shallow general knowledge. If it's safe, it can do it.
See VISION.md for full platform vision.
- File System - Organization, metadata, UTIs
- System Preferences - Bluetooth, WiFi, Display, Sound, Accessibility
- Troubleshooting - Common macOS issues and solutions
- Automation - AppleScript, Shortcuts, Automator
- CLI Tools - Terminal commands, shell operations
- Privacy/Security - Permissions, Keychain, FileVault
- Hardware - Apple Silicon, peripherals, displays
| Product | Focus | Status |
|---|---|---|
| SaneSync | File organization | In development |
| SaneBar | Menu bar hub | Planned |
| SaneHelp | Troubleshooting | Planned |
| SaneAuto | Automation | Planned |
- Apple Developer Documentation
- System Preference panes
- UTI type definitions
- CLI tool man pages
- Troubleshooting flows
- Model: Llama 3.2 3B or Qwen 2.5 Coder 3B (~2GB)
- Training: MLX LoRA on Apple Silicon
- Runtime: Fully local, no cloud
# Training (once data is ready)
./scripts/train.sh
# Test model
./scripts/test.shSaneAI/
├── VISION.md # Platform vision
├── training_data/ # JSONL training examples
├── models/ # Trained adapters and fused models
├── scripts/ # Training and testing scripts
└── .claude/ # Claude Code configuration
MIT — see LICENSE