microsoft/magentic-ui is the repository for MagenticLite, an experimental agent that operates across the browser and the local file system. The latest release is v0.2.1, written in Python and published under the MIT license.
Microsoft AI Frontiers presents the app as the next-generation Magentic-UI. It pairs MagenticBrain, an orchestrator model suited to on-device environments, with Fara, a model specialized for browser tasks. The agent performs work such as web research, filling out forms, file management, and restaurant search and booking, while the user can steer the progress or approve steps and then take over the task directly. Browser sessions run in a Quicksand lightweight VM sandbox, designed so that the agent cannot reach the rest of the computer without permission. The repository has 10095 GitHub stars.
That said, the project itself states that it is at an experimental stage. The isolation described in the README covers only the sandbox that holds the browser session, so it is hard to confirm whether file management tasks are handled within the same boundary. There are no release notes describing the v0.2.1 changes, and no figures showing task success rates either. The evidence for judging performance and stability is still insufficient. It is a repository worth looking at if the goal is to see the shape of an agent that moves between the browser and files, but whether it is at a level usable in practice right away is too early to judge.