magentic-ui automates real work such as web research, form filling, and file management using small models efficiently, operating across the browser and local file system while keeping users in control through approval workflows. It does not depend on frontier-scale compute.
Project overview
Built specifically for small models, magentic-ui runs browser sessions inside a lightweight VM sandbox and lets users steer, approve, or take over agent actions, addressing both efficiency and safety for real-world task automation.
Project type
AI Agent · Workflow
Use cases
Documents & Office · Automation
Deployment
Refer to project documentation
License
MIT
Best for
General users and developers who want to automate browser and file tasks with small models while retaining human-in-the-loop control.
Key capabilities
Automates web-based tasks like web research, filling expense forms, and finding prices.
Automates organizing local files.
Allows users to steer, approve, or take over agent actions; checks in before taking critical actions.
Runs browser sessions inside a lightweight VM sandbox (Quicksand) to prevent the agent from reaching the rest of the machine without permission.
Limitations and risks
Quick start is supported on macOS or Windows (WSL) only.
Requires a user-provided connectable model endpoint to operate.
The project does not require frontier-scale models; confirm target workloads align with small-model constraints.
Getting started
Create a directory, set up a Python 3.12 virtual environment using uv, install magentic_ui (>=0.2.0) via uv pip, then run magentic-ui on port 8081. Commands: mkdir magentic-lite && cd magentic-lite; uv venv --python=3.12 --seed .venv && source .venv/bin/activate; uv pip install "magentic_ui>=0.2.0"; magentic-ui --port 8081.
Evidence and sources
README: MagenticLite is the next generation of Magentic-UI — an agentic application from Microsoft AI Frontiers, redesigned to do more with less. It pairs an on-device-friendly orchestrat…
README: Works across the browser and your local file system. Web research, form filling, file management — in one workflow.
README: Keeps you in the loop and in control. Steer, approve, or take over at any point. MagenticLite stops and checks in before taking critical actions.
README: Safe by design. Browser sessions run inside a lightweight VM sandbox (Quicksand) so the agent can't reach the rest of your machine without your say-so.
README: Open <http://127.0.0.1:8081/> and follow the in-app onboarding to connect a model endpoint.
AI Search
Find projects, verify facts, compare options, or turn a complex need into an actionable plan
Try a searchA click only fills the search box; you stay in control