Cua AI has released Cua, an open-source project designed to provide AI agents with the tools necessary to operate computers. The framework includes desktop automation, isolated cloud desktops, and specialized decision models to support the development and evaluation of computer-use agents.
Cua Launches Open-Source Framework for Computer-Use Agents
The ecosystem consists of several core components: Cua Driver, Cua Bench, and Lume. Cua Driver acts as an open-source background driver that allows agents to click, type, and inspect accessibility trees on macOS, Windows, and Linux via MCP, CLI, or SDKs. Crucially, it can operate without taking focus away from the user's cursor. Lume provides a virtualization stack for managing macOS and Linux VMs on Apple Silicon, while Cua Bench offers tools to build tasks and evaluate agent performance.
The project also introduces CUA-S1, a family of small, specialized "System 1" models. These models are designed for fast, bounded decisions, such as determining if a specific value belongs in a UI field, rather than performing complex high-level reasoning. The CUA-S1-FORMS model specifically focuses on scoring decisions from structured interface elements and document values.
Cua aims to scale computer-use workflows by allowing users to run training, evaluation, and data-generation workloads in parallel across diverse environments, including Ubuntu containers, Windows 11, and Android VMs. The core software is released under the MIT license.
Sources
- Show HN: CUA-S1 – A System One Model for Computer Use （Hacker News Frontpage, 2026-09-19）
- Cua