Meet Shadow: the other half of the story

[mcheemaa]

Ghost OS gives AI agents eyes and hands on your Mac. We have been building the other half.

Shadow is a personal intelligence engine for macOS. It captures every signal your computer produces while you work across 14 modalities, turns raw behavior into structured episodes, and proactively surfaces what matters. All on-device, all open source.

Ghost OS acts. Shadow observes, learns, and remembers.

What is built and working today:

• 14-modality capture (screen, audio, keystrokes, accessibility tree, clipboard, files, git, terminal, and more)
• Episode generation from your work with LLM summaries
• Proactive heartbeat that pushes suggestions without being asked
• On-device LLM inference (Qwen 7B/32B via MLX)
• Vision grounding with ShowUI-2B and built-in LoRA fine-tuning
• 25-tool agent runtime with safety gates and undo support
• Semantic search (CLIP visual + Tantivy full-text)
• Meeting transcription and summarization
• 3-tier retention (hot/warm/cold) with automatic storage management
• Crash-proof recording, sleep/wake recovery, multi-display hot-plug

We are open-sourcing Shadow because the capture layer is the hardest problem to solve and we have solved it. The next layer belongs to the community: memory graphs, MCP servers, personal models, agents trained on real human behavior.

We will be pushing releases and updates soon to make it even easier to build from source, contribute, and extend. If you care about personal AI, computer use, or building on top of real behavioral data, check it out.

github.com/ghostwright/shadow

Your computer was paying attention the whole time.