Microsoft has released Magentic-UI, a research prototype designed to automate web-based tasks.
This tool uses a multi-agent system that can browse websites, execute code, and analyze files, providing users with a transparent, interactive interface.
Magentic-UI is powered by AutoGen's Magentic-One system and includes specialized agents like Orchestrator, WebSurfer, Coder, FileSurfer, and UserProxy.
Key features include a co-planning interface for collaborative plan creation and Action Guards for user approval of sensitive actions.
The system also supports parallel task execution and plan learning for improved automation.
Magentic-UI, built with Docker, is available on macOS, Linux, and Windows (with WSL2), and allows users to study human-agent interaction.