magentic-ui: Automate Browser and Local File Tasks with AI

Overview

MagenticLite is an experimental agent for carrying out tasks across a web browser and a local file system. It is designed to let smaller, more efficient models handle practical workflows such as researching information online and then organizing or using local files.

The agent keeps a person involved: users can steer or take over, and it checks before critical actions. Browser sessions run in a lightweight VM sandbox, limiting access to the rest of the machine without permission.

Key Features

• Coordinates browser activity and local file operations in one workflow.
• Pairs an orchestrator model, MagenticBrain, with the specialized browser-use model Fara.
• Supports supervised operation, including steering, approval, and takeover.
• Checks in before critical actions.
• Runs browser sessions in a Quicksand VM sandbox.
• Provides a web interface for onboarding and connecting a model endpoint.

Use Cases

• Researchers can gather information from websites and work with related local files in the same task.
• Employees can use an agent to fill expense forms while reviewing or approving consequential steps.
• People planning meals can ask it to look up ingredient prices across websites.
• Users organizing a local collection of files can combine file management with browser-based research.

Getting Started

On macOS or Windows with WSL, install and start it with:

uv venv --python=3.12 --seed .venv
source .venv/bin/activate
uv pip install "magentic_ui>=0.2.0"
magentic-ui --port 8081

Then open http://127.0.0.1:8081/ and follow the onboarding to connect a model endpoint. See the README and installation guide for details and other platforms.

Alternatives

• index: Index focuses on multi-step web tasks and structured results, and is archived; MagenticLite also combines browser automation with local file access.
• browserable: Browserable is a self-hostable JavaScript library with a REST API, while MagenticLite is a Python agent for supervised workflows that include local files.
• UI-TARS-desktop: UI-TARS-desktop uses vision-language models to operate desktop interfaces, while MagenticLite focuses on supervised browser and local-file workflows.

Considerations

• This is an experimental project, so check the limitations and troubleshooting guide against your intended workflows before relying on it.
• You need Python 3.12 or newer and a model endpoint. The repository does not specify a particular GPU or database requirement in the supplied project facts.
• The browser sandbox and approval flow are part of its safety approach, but the project describes them as safeguards, not a guarantee that every task is risk-free. Review the transparency note and configuration guidance.
• The MIT license permits broad use, subject to its terms. Microsoft names and trademarks are not granted by the project license.
• Project activity is recent, with releases and merged pull requests, and contributions are not concentrated in one person. However, the measured period had no closed issues, so issue handling may warrant checking for teams that depend on timely support.