This is an early release preview. You may encounter bugs.
Cua logo
Unclaimed

Platform interface deployment

Cua

Open-source sandboxes, cross-OS desktop drivers, SDKs, MCP server and benchmarks for computer-use agents.

A 85/100 GitHub score ? This grade is derived from GitHub signals, not user votes. Open for the full breakdown.
Discussion

01 / Summary

About Cua.

Cua gives AI agents computers to operate. It provides the machines and the automation tools, and you bring your own agent and model. The project covers an agent moving between code, APIs, and graphical interfaces within one task, which it calls Computer-Use 2.0.

Cua Driver is the core component: a background driver that lets agents click, type, scroll, read accessibility trees, and capture window state in native desktop apps and browsers on macOS, Windows, and Linux, without moving the user's pointer or taking focus where the platform allows it. One binary runs as an MCP stdio server, a long-running daemon, or a one-shot shell command, with typed SDKs alongside. Agents such as Claude Code, Codex, and OpenClaw connect to it.

Around the driver sit the rest of the stack. The Cua SDK and CLI boot local VMs and containers (gVisor, Docker, QEMU, Apple VZ) or machines in Cua Cloud, with the same API in Python, TypeScript, Swift, and Kotlin. Lume manages macOS and Linux VMs on Apple Silicon through Apple's Virtualization framework. Cua Spaces is a macOS desktop app that gives agents full desktops, on your Mac, on machines you own, or in your own AWS, Google Cloud, or Modal account. Cua Bench builds computer-use tasks, evaluates agents, and exports trajectories for training, and CUA-S1 is a research family of small models for fast, bounded computer-use decisions such as filling forms.

Hosted Cua Cloud adds warm pools of pre-booted Linux, Windows, macOS, and Android machines that scale to zero, aimed at evaluation runs, reinforcement-learning loops, and data generation.

  • Cua Driver: background clicks, typing, scrolling, accessibility-tree reads, and window capture on macOS, Windows, and Linux
  • MCP and CLI surface: one binary runs as an MCP stdio server, a daemon, or a one-shot command, so agents and people drive the same harness
  • Sandboxes: Linux containers and VMs, Windows 11 and Server 2025, macOS, and Android machines through one API
  • Multi-language SDK: the same sandbox API in Python, TypeScript, Swift, and Kotlin, embedded or through a shared daemon
  • Lume: creates and manages macOS and Linux VMs on Apple Silicon
  • Cua Spaces: a macOS app giving agents full desktops on your Mac, your own machines, or your own cloud account
  • Teleport: moves a signed-in app such as Chrome or Slack into a Space, with sessions kept in an encrypted local keyvault
  • Multiplayer desktops: you and your agents share one desktop, each with your own cursor
  • Cua Bench: builds computer-use tasks, evaluates agents, and exports trajectories for training
  • CUA-S1: small specialized models for computer-use decisions, starting with forms, with weights on Hugging Face
  • Cloud warm pools: pre-booted machines claimed on demand for evals, RL loops, and batch rollouts, scaling to zero when idle

02 / Discussion CREDIBILITY-GATED

Discussion

Reading is open to everyone. Posting and voting need a verified identity or a GitHub grade of B or higher.

  • No discussions yet.

04 / Build

Build with Cua.

Browse the catalog for frameworks, tools, and harnesses, each scored on real GitHub credibility.

Get Cua →

Browse the catalog