Cua

Cua is a computer use agent platform that enables developers to build, deploy, and scale AI agents that can control full operating systems — seeing screens, clicking buttons, typing, and running code. It provides managed cloud VM sandboxes across Linux, Windows, and macOS, paired with a unified inference API supporting 9+ vision-language models from providers like Anthropic, OpenAI, Google, and ByteDance. Built on an open-source foundation (MIT License), Cua is designed for teams that need reliable, scalable computer-use automation without managing infrastructure.

• Cloud Sandboxes: Spin up Linux, Windows, or macOS VMs instantly via the CLI or API — billed per minute with full desktop access and persistent storage.
• Unified VLM Inference: Access Claude, Gemini, GPT, UI-TARS, and other vision-language models through a single API key and credit system.
• Cuazar Playground UI: Use the web-based playground to run agents and automate tasks without writing any code — just describe the task and watch it execute.
• Agent SDK (Python): Build multi-turn computer-use agents with structured outputs, custom tools, and screenshot-based reasoning using the open-source Python SDK.
• MCP Integration: Run computer-use agents inside VMs via the Model Context Protocol, enabling safe and isolated automation workflows.
• Benchmarking & Evals: Generate UI datasets, capture agent trajectories, and run benchmarks like OSWorld with a single line of code using the HUD integration.
• Cross-Platform Support: Choose between Linux (small/medium/large), Windows (small/medium/large), and macOS (preview) sandbox environments based on your automation needs.
• Open Source Core: The underlying agent framework is MIT-licensed and available on GitHub, with active community support via Discord.