Developers need to build AI agents that can interact with graphical user interfaces across multiple operating systems, but provisioning secure, scalable, and cross‑platform execution environments is complex and time‑consuming. Managing sandboxed containers, integrating diverse large language and vision models, and handling infrastructure for benchmarking and deployment adds significant overhead.

Cua delivers an open‑source framework and a managed cloud container platform that abstracts away the infrastructure required for computer‑use agents. The Computer SDK provides a pyautogui‑like API to control Windows, Linux, and macOS sandboxes consistently, while the Agent SDK standardizes model schemas and supports benchmarking on datasets such as OSWorld‑Verified and SheetBench‑V2. Developers can run agents in isolated sandboxes locally or in Cua Cloud, choose from over 100 vision‑language models (including Claude, GPT‑4, Gemini, Ollama), and orchestrate multi‑turn workflows through a unified interface. The platform also offers a web‑based Playground for code‑free execution and a pay‑as‑you‑go credit model for compute and VLM inference, enabling rapid prototyping, scaling, and secure deployment.

Open‑source framework for building, deploying, and evaluating computer‑use agents.

Cloud container platform that provides sandboxed Windows, Linux, and macOS environments with a consistent pyautogui‑like API.

Computer SDK for automating UI actions (click, type, run code) across OSes.

Agent SDK with a unified schema, support for 100+ vision‑language models, and built‑in benchmarking (OSWorld‑Verified, SheetBench‑V2, ScreenSpot).

Managed cloud sandboxes and a web Playground UI for running agents without writing code.

Flexible integration with multiple LLM providers (Claude, GPT‑4, Gemini, Ollama, MLX) via composed agents.

Free tier with 10 starter credits and a Pro tier priced per credit for cloud compute and VLM inference; Enterprise custom plans available.

Community resources: documentation, tutorials, examples, and a Discord support channel.

AI developers, RPA engineers, and organizations that need to automate GUI‑based workflows or build agents that interact with computers across Windows, Linux, and macOS without managing underlying infrastructure.

The platform monetizes through a credit‑based pay‑as‑you‑go model: a free tier with limited credits, a Pro tier starting at $10 for additional cloud compute and VLM inference credits, and custom Enterprise contracts for larger teams.

Cua has amassed over 9 000 GitHub stars within four months of release, indicating strong community adoption. The company announced the general availability of Windows sandboxes and provides a free tier to encourage early experimentation. Seed funding has been secured, supporting continued development and scaling of the platform.