# ghostwright/ghost-os

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

Repository: https://github.com/ghostwright/ghost-os
Canonical: https://ross.abutalabs.com/products/ghost-os
Homepage: https://ghostwright.dev/ghost-os
Language: Swift
License: MIT
License Family: permissive
Topics: accessibility, ai-agents, automation, claude-code, computer-use, llm-tools, macos, mcp, recipes, swift
Last push: 2026-03-23T22:41:44+00:00

## Health v2 (maintenance only)
Score: 61/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 73, release rhythm 74, longevity 13
- inputs: {"age_days": 195, "days_push": 163, "days_rel": 174, "gap_med": 0.0, "n_releases_24m": 9}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1646, forks 158 (observed 2026-08-28T04:05:16.260111+00:00)

## What it is
Ghost OS is a native macOS framework that gives AI agents full computer-use capabilities by exposing the macOS accessibility tree, Chrome DevTools Protocol, an on-device vision model, and synthetic input through an MCP-compatible interface. It includes self-learning recipes that record a user performing a task once and replay it, letting agents click buttons, fill forms, and operate any app without screenshots.

## Use cases
- let an AI agent click buttons and fill forms in native macOS apps
- automate sending emails or organizing files via an LLM agent
- record a workflow once and have the agent replay it forever
- give Claude Code or other MCP clients control of the Mac UI
- control web apps through Chrome DevTools Protocol instead of screenshots
- run on-device visual grounding when accessibility data is unavailable

## When to choose
- you need computer-use automation on macOS without screenshot-based vision loops
- you want an MCP server that exposes native app control to any agent
- you want fast, structured UI interaction via the accessibility tree
- you want self-learning recorded workflows
- you prefer on-device inference on Apple Silicon

## When to avoid
- you need computer-use on Linux or Windows
- your automation is browser-only and a browser automation tool suffices
- you need a screenshot-pixel-based approach for non-native rendering environments
- you cannot grant macOS accessibility and input permissions

## Facets
- artifact type: framework
- maturity: active
- function: agent-framework, mcp, workflow-automation, computer-vision, gui
- domain: operating-systems, accessibility, developer-tools
- platform: cli
- tags: computer-use, accessibility-tree, mcp-server, self-learning-recipes, apple-silicon, cg-event, chrome-devtools-protocol, on-device-llm, ai-agents, automation, macos, swift

## Member repositories
- ghostwright/ghost-os (main) score 61

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:16.260111+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:45:22.973464+00:00, confidence not recorded.
  - readme: https://github.com/ghostwright/ghost-os (fetched 2026-08-28T04:05:16.260111+00:00, sha f933d76ede39)
  - homepage: https://ghostwright.dev/ghost-os (fetched 2026-08-29T11:18:41.449278+00:00, sha 3bbb122749ec)
- Data as of 2026-08-30T08:39:29.467469+00:00.