MagenticLite is an experimental computer-use agent from Microsoft AI Frontiers — the next generation of the Magentic-UI project — designed to do real work across the browser and your local file system using small, efficient models instead of frontier-scale ones. The open-source repository has gathered over 10,000 stars on GitHub and was actively developed through 2026.
MagenticLite pairs two specialized models: MagenticBrain, an on-device-friendly orchestrator that plans and coordinates tasks, and Fara, a browser-use model that operates web pages. Together they handle jobs like web research, form filling, price hunting, and file management end to end. Human oversight is a design pillar: you can steer, approve, or take over at any point, and the agent checks in before critical actions. For safety, browser sessions run inside a lightweight VM sandbox (Quicksand) so web activity cannot reach the rest of your machine without permission. The project descends from Microsoft Research’s Magentic-UI and AutoGen work on multi-agent orchestration, rethought around a simple bet: carefully specialized small models, given the right tools and supervision, can complete most everyday computer tasks at a fraction of the cost of the largest models.
Free. MagenticLite / Magentic-UI is an open-source research project from Microsoft, released under the MIT license. You run it yourself and supply model access through supported providers or local models.
Researchers and builders exploring practical computer-use agents — especially those interested in capable automation that runs on modest models and hardware, with strong human oversight and sandboxing.