Cua
Open-source infrastructure for training, evaluating, and benchmarking AI-powered Computer-Use Agents that control full desktop environments (macOS, Linux, Windows) with integrated sandboxes, SDKs, and reproducible evaluation tools.
Cua is built in Python, distributed under the MIT License, 13.4k GitHub stars, latest release sandbox-v0.1.9.
When to use Cua
Cua is listed here as a AI project. The directory calls out Sandboxed desktop environments for agent safety, Support for macOS, Linux, and Windows, SDKs for agent development and integration as capabilities associated with it.
Other recorded traits for Cua include Benchmark suites for evaluation, Supports end-to-end training and validation of agents.
Besides AI, this page also files Cua under Development, Testing, Tools.
Cua compared with
Records in this directory name SIMA, Voyager, OpenInterpreter as products people compare with Cua. That list is editorial metadata, not a claim that Cua replaces each of them.
What the Cua stats reflect
GitHub currently shows 13.4k GitHub stars, 832 forks, 181 open issues, latest tracked release sandbox-v0.1.9. Star and activity counts here are a snapshot used as a proxy for community adoption, not a quality score.
Stats refreshed
- Language
- Python
- Latest Release
- sandbox-v0.1.9
- License
- MIT License
Our Newsletter
Get new AI tools right in your inbox
Get short emails with useful ai projects, releases, and repos worth watching.
Key features of Cua
- Sandboxed desktop environments for agent safety
- Support for macOS, Linux, and Windows
- SDKs for agent development and integration
- Benchmark suites for evaluation
- Supports end-to-end training and validation of agents
Recorded alternatives to Cua
Cua resources
Cua on GitHub
Frequently asked questions
What is Cua?
Open-source infrastructure for training, evaluating, and benchmarking AI-powered Computer-Use Agents that control full desktop environments (macOS, Linux, Windows) with integrated sandboxes, SDKs, and reproducible evaluation tools. This directory highlights Sandboxed desktop environments for agent safety, Support for macOS, Linux, and Windows, SDKs for agent development and integration.
Is Cua free to use?
Cua is published as open source under the MIT License. The directory lists Sandboxed desktop environments for agent safety, Support for macOS, Linux, and Windows, SDKs for agent development and integration among its recorded capabilities.
What language is Cua written in, and what is the latest release?
Cua is written primarily in Python. The latest release tracked on this page is sandbox-v0.1.9.
How widely is Cua used on GitHub?
Cua has about 13.4k GitHub stars. It also has about 832 forks. Those counts are a snapshot of community attention, not a ranking of quality.
What do people compare Cua with?
This directory records SIMA, Voyager, OpenInterpreter as comparison points for Cua.
More AI tools
Developer
The pioneering library allowing the embedding of a developer agent into your application, enhancing interactivity and automation.
Cognee
Memory for AI Agents in 5 lines of code. Cognee provides a lightweight and efficient memory management solution tailored for AI agent development.
Trae-agent
Trae Agent is an LLM-based agent designed to assist with a wide range of software engineering tasks, leveraging the power of large language models for effective development solutions.
Csm
A cutting-edge Conversational Speech Generation Model designed for creating realistic and engaging speech outputs.
Olmocr
A toolkit designed for converting and linearizing PDFs to create datasets optimized for large language model (LLM) training and evaluation.
CodeGeeX
CodeGeeX: An Open Multilingual Code Generation Model featured at KDD 2023, enabling generation and understanding of code across multiple languages.