#814 · Primary category: AI Agents & Automation

ghost-os

accessibility ai-agents automation claude-code computer-use llm-tools macos mcp recipes swift

Full computer-use for AI agents. Self-learning workflows. Native macOS. No screenshots required.

Project last updated:03/23/26

GitHub Stars

1.6K

Forks

158

Contributors

2

License

MIT

Why we included this project

Most agent frameworks are good at one thing: talking to code and terminals. Ghost OS instead gives a model direct control over the Mac, letting it click, type, drag, and manage windows in native apps like Slack or Messages rather than working inside a browser tab, and any MCP-compatible client such as Claude Code or Cursor can drive it. The interesting part is the self-learning recipe flow: show it a task once, and it records the steps, has a frontier model turn that into a reusable recipe, and a small local model replays it later. Recipes are plain JSON, so you can read exactly what will run before it runs, and the design is local-first: data stays on your machine. A vision-model fallback covers web apps where the accessibility tree is too thin, which is how it avoids screenshots.

Articles for this project

No articles for this project yet.

To suggest a topic or contribute an article, contact us.

Related projects in this category