Ghost Os

Ghost OS is open source computer use tooling for AI agents on macOS. Once installed, an agent can see and operate every application on the machine, which closes the gap between agents that can write code in a terminal and the everyday work of clicking buttons, filling forms, sending mail and organizing files. The technical choice that defines it is reading the macOS accessibility tree rather than screenshots. Structured, labeled element data tells the agent what is a button, a field or a link, with no vision inference per action. When the tree falls short the system cascades: the Chrome DevTools Protocol handles web apps where Chrome flattens elements, a local vision model running on Apple Silicon grounds elements by description when structured data is unavailable, and synthetic system level input events are the last resort. Capabilities reach the agent as 29 tools over the Model Context Protocol, grouped around perceiving, acting, navigating, waiting, learning and replaying. The differentiating feature is self learning recipes. A user records a task once while performing it manually, the system captures each action with full element context, and a model synthesizes a parameterized recipe stored as JSON. Recipes are readable before they run, shareable across a team, and cheap to replay on a small model because the reasoning was done once. No prices are published. Ghost OS is open source and installed from its GitHub repository at no cost, and the vendor's pricing link routes to a signup page for Phantom Cloud, its hosted companion offering, rather than to a rate card. Anyone evaluating the cloud service will need to register to see terms.