Cua Launches Windows Support for AI Agent Computer-Use with Synthetic Cursors

Summary

Cua Driver now enables background AI agents—including Claude Code, Codex, Hermes, and any MCP or CLI agent loop—to control Windows desktop applications through computer-use technology. The driver provides agents with three critical capabilities: window pixels, UIA/MSAA accessibility trees, and an action layer for clicks, typing, scrolling, and verification.

Developing Windows support presented significantly greater technical complexity than macOS due to the diversity of Windows application architectures. The driver must seamlessly handle Win32 legacy apps, WPF with routed events, WinUI and UWP backed by XAML and DirectComposition, Electron and Chromium browsers, LibreOffice VCL/SAL apps, GTK applications, and custom-rendered canvases. This architectural fragmentation required Cua to build a sophisticated routing system that intelligently selects the optimal backend for each target and action type.

A key innovation is the synthetic cursor—painted and controlled by Cua Driver and kept visually separate from the user's physical pointer wherever possible. This isolation allows agents to interact with applications without interfering with user workflows or pulling windows into focus. The driver uses PrintWindow and GDI for classic Win32 surfaces, falls back to Windows.Graphics.Capture for DirectComposition-backed windows, and selectively deploys PostMessage, SendInput, or UI Automation based on the target application's characteristics.

Computer-use creates an extended reasoning loop for AI agents: they can edit code, launch applications, observe real window state, interact with UI controls, and verify their own work—enabling more sophisticated task completion on systems where traditional APIs don't exist.

• Synthetic cursors are isolated from the user's physical pointer, preventing agent interactions from disrupting user workflows
• Computer-use enables agents to complete tasks on legacy desktop applications that lack public APIs—critical for Windows business software

Editorial Opinion

Cua Driver's Windows support addresses a real gap in AI agent infrastructure. While computer-vision-based agents can theoretically handle any GUI, having a driver that understands accessibility trees and app state transforms agents from visual simulators into genuine desktop automation tools. For a massive installed base of Windows business software—much of it legacy code without public APIs—this matters. The architectural sophistication required to handle Windows's fragmented app ecosystem signals that computer-use infrastructure, not just model capability, will be a competitive differentiator in agent development.