OpenGCUA
Open stack from Tongyi Lab's MAI-UI team for generalist computer-use agents that combine GUI clicks, accessibility trees, terminal commands, APIs and code: an OSWorld harness, a macOS real-device harness, a cross-device SDK and published task runs of eight frontier models.
Updated
✨ Try AgentScout's free AI image generator →Good for: coding in the terminal, browsing the web, controlling a computer, building AI agents
Use cases
About OpenGCUA
OpenGCUA is listed in our browser/computer use category and is aimed at builders. Open stack from Tongyi Lab's MAI-UI team for generalist computer-use agents that combine GUI clicks, accessibility trees, terminal commands, APIs and code: an OSWorld harness, a macOS real-device harness, a cross-device SDK and published task runs of eight frontier models. Pricing, as the maker states it: Free and open source (Apache-2.0); research preview. Maturity: Early (research preview). We last checked the official link on 2026-10-20.
What is OpenGCUA?
OpenGCUA (Open Generalist CUA) is an ongoing open project from the MAI-UI team at Tongyi Lab to build the full stack for generalist computer-use agents: systems that complete computer tasks by deciding for themselves whether to look at a screenshot or an accessibility tree, and whether to click, call an API, run a shell command or write a small program that combines them.
The first research preview (October 2026) releases an OSWorld integration example with reference screenshot-action and code-action agents, evaluation trajectories for eight frontier models, a preliminary macOS real-device harness (a menu-bar app that runs the agent on your own Mac) and a Cross-Device SDK with one programming interface for computers, browsers and phones. On the OSWorld-v2 offline set the team reports the design raised the mean partial score of the eight models from 45.3 to 64.9 and cut steps per task by 53% on average. A model release is listed as next on the roadmap.
Sources: the official README and our directory listing, checked 2026-10-20. Official site: tongyi-mai.github.io/OpenGCUA.
Who is OpenGCUA for?
Researchers and developers building computer-use agents who want an open harness, SDK and benchmark trajectories to start from, and Mac users who want to try a real-device agent.
OpenGCUA at a glance
- Open source: Yes (Apache License 2.0)
- GitHub: Tongyi-MAI/OpenGCUA · 315 stars · 23 forks (as of Oct 20, 2026)
Key features
- Generalist design: GUI, accessibility tree, shell, API and code actions in one agent
- OSWorld integration example with screenshot-action and code-action agents
- Evaluation trajectories of 8 frontier models released
- macOS real-device harness (macOS 14+) that can work in background windows
- Cross-Device SDK (qwen-intelligence-sdk) for computers, browsers and phones
- Interactive lab and tech report on the project site
OpenGCUA FAQ
What is OpenGCUA?
OpenGCUA (Open Generalist CUA) is an ongoing open project from the MAI-UI team at Tongyi Lab to build the full stack for generalist computer-use agents: systems that complete computer tasks by deciding for themselves whether to look at a screenshot or an accessibility tree, and whether to click, call an API, run a shell command or write a small program that combines them.
Who is OpenGCUA for?
Researchers and developers building computer-use agents who want an open harness, SDK and benchmark trajectories to start from, and Mac users who want to try a real-device agent.
Is OpenGCUA free?
OpenGCUA's pricing, as the maker states it: Free and open source (Apache-2.0); research preview. Check the official site for current plans.
Is OpenGCUA open source?
Yes. OpenGCUA is released under the Apache License 2.0, according to its GitHub repository.
What does OpenGCUA need to run?
The macOS real-device harness needs macOS 14 or later; Python 3.12 is recommended, and prebuilt SDK wheels do not need Rust. The project calls the harness and SDK preliminary.
Is there an OpenGCUA model?
Not yet. The roadmap lists a GCUA model for computer and browser use as the next release; this preview opens the harness, SDK, OSWorld example and trajectories.
What are alternatives to OpenGCUA?
Similar tools include Browser Harness, ego lite, OpenAdapt, CloakBrowser, Scrapling and Scout (Outcrawl).
OpenGCUA ratings and reviews
OpenGCUA alternatives
Other browser/computer use tools in our directory. See all browser/computer use tools →
Browser Harness
Agent skill from the Browser Use team that lets Claude Code, Codex or Cursor drive your own logged-in Chrome to click, type, read pages and download files.
ego lite
Free macOS browser where Claude Code or Codex run automation tasks in their own isolated Spaces next to your tabs, reusing your logged-in sessions through an agent skill.
OpenAdapt
Execution layer for computer-use agents that turns a demonstrated task into a local program that repeats the clicks without model calls and stops on mismatches.
CloakBrowser
Chromium build with source-level fingerprint patches, used as a drop-in Playwright or Puppeteer replacement for agent browsing.
Scrapling
Python web-scraping framework with fast HTTP requests, stealth browsers, adaptive parsing and concurrent crawling, plus an MCP server for agents.
Scout (Outcrawl)
Open-source browser agent from Outcrawl that opens a real browser and clicks, types and reads pages to finish a task you describe.
Related lists
🏆 See also: Best AI Browser Agents · All best-of lists
🧭 Task guides: Agent frameworks and SDKs for terminal coding · Computer and desktop control tools for web browsing agents
Badge for makers
Built OpenGCUA? Add this badge to your README or site. It links back to this page. About badges
