Magentic-UI

Magentic-UI is Microsoft's open-source agentic web interface, announced at Build 2025 and pitched as a "human-centred web agent." It is a computer-use agent system built on the Magentic-One architecture, with four agents — Orchestrator, WebSurfer, Coder, and FileSurfer — that browse and navigate through virtual machines, access web browsers, execute Python and shell code, and analyze files. It can perform tasks like restaurant reservations or online purchases.

Its differentiator is transparency and user collaboration: all agent actions are shown step-by-step in a visible task panel, users can modify plans in real time, pause execution, give natural-language feedback, or resume direct control of the browser. It asks for explicit user consent before any irreversible action, such as placing an order. It is available on GitHub and via Azure AI Foundry Labs. Caveat: as a research release, it is positioned for experimentation rather than production deployment, and computer-use agents remain error-prone on complex real-world sites.

Available open-source on GitHub and through Azure AI Foundry Labs. Runs agents in virtual machines for web tasks.