Skip to content
Get packs

Guide · Open source

How to use magentic-ui: install MagenticLite and run it

Updated · written with AI from the project's own documentation and checked against our catalogue data

magentic-ui is the open-source project that contains MagenticLite, an experimental agent from Microsoft AI Frontiers. It can work in a web browser and on your local file system to handle tasks such as web research, form filling and organizing files. This guide walks through the install on macOS or Windows with WSL, the first run, and what to set up before the agent can do anything useful.

MagenticLitemicrosoft/magentic-ui · ★ 10.1k · MIT

Quick start

  1. 1Create a project folder. Make a directory called magentic-lite and move into it.
  2. 2Create and activate a virtual environment. Run uv venv with Python 3.12 and activate .venv.
  3. 3Install magentic_ui. Install the latest 0.2.x release from PyPI with uv pip install "magentic_ui>=0.2.0".
  4. 4Start the app. Run magentic-ui --port 8081.
  5. 5Open the web interface. Go to http://127.0.0.1:8081/ in your browser.
  6. 6Connect a model endpoint. Follow the in-app onboarding. If you have no endpoint, use the Model Hosting Guide.

What MagenticLite does and who it suits

MagenticLite is built to run on smaller models rather than frontier-scale ones. It pairs an orchestrator model, MagenticBrain, with a browser-use model, Fara, so you can automate real work without depending on heavy compute.

You can steer it, approve actions, or take over at any point. It stops and checks in before critical actions, which makes it a reasonable fit for people who want to review what an agent does.

  • Automating repetitive browser tasks such as form filling or price lookups
  • Experimenting with agents that combine browser and local file actions
  • Running agents on small or self-hosted models
  • Approving critical actions before the agent takes them

Check the prerequisites before you install

The quick start covers macOS or Windows via WSL. The prerequisites are Python 3.12, installed through uv, and the uv package manager itself. Other platforms are described in a separate installation guide that this review did not cover.

You also need a model endpoint. The app asks you to connect one during onboarding, and the README points to a separate Model Hosting Guide if you do not have one yet.

  • Python 3.12 (installed via uv)
  • uv package manager
  • macOS, or Windows with WSL

Install magentic-ui from PyPI

Create a project folder, then a virtual environment with Python 3.12, and activate it. Finally, install the latest 0.2.x release from PyPI with uv.

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"

Start the app and open it in your browser

Run the magentic-ui command with a port. The README uses 8081. Then open the local address in a browser and follow the in-app onboarding to connect a model endpoint.

The README notes that the previous Magentic-UI 0.1 release, which was optimized for frontier models, lives on a separate branch. The guide above uses the 0.2.x install path.

  • Open http://127.0.0.1:8081/ after the app starts
  • Complete the in-app onboarding to connect a model endpoint
magentic-ui --port 8081

Understand the safety controls

Browser sessions run inside a lightweight VM sandbox called Quicksand, so the agent cannot reach the rest of your machine without your say-so. Sandbox, agent mode and tool approval are covered in the Configuration docs.

Use approval and take-over whenever a task touches money, accounts or files you cannot easily restore. The README says the agent checks in before critical actions.

Where to go for more detail

The repository documentation covers each stage. Use the Installation page for platform details, the Build from source page to run from a clone, the Configuration page for config.yaml, and the Troubleshooting page for common errors.

  • Installation: supported platforms and install commands
  • Build from source: uv sync and pnpm dev workflows
  • Model hosting guide: standing up a model endpoint
  • Configuration: sandbox, agent mode, tool approval, config.yaml
  • Troubleshooting: common errors and fixes
  • Limitations: tasks the project does not handle well

What people use it for

  • Filling expense forms in a browser
  • Finding prices for recipe ingredients
  • Finding and booking a restaurant
  • Organizing files on your local file system

Frequently asked questions

How do I install magentic-ui?

Create a folder, make a Python 3.12 virtual environment with uv, activate it, and run uv pip install "magentic_ui>=0.2.0". The README gives these exact commands for macOS or Windows with WSL.

What do I need to run magentic-ui?

You need Python 3.12 installed through uv, the uv package manager, and macOS or Windows with WSL. You also need a model endpoint, which the app asks you to connect during onboarding.

Is magentic-ui free to use?

The code is licensed under the MIT License, as stated in the README. The README does not list prices, so check your model endpoint provider for any costs of its own.

Does magentic-ui work on Linux?

The quick start covers only macOS and Windows via WSL. Other platforms are described in a separate installation guide that was not reviewed for this page.

What is the difference between magentic-ui and Browser Use?

Browser Use is one of the comparable projects listed for this item. magentic-ui focuses on combining browser tasks with local file actions and keeping you in control through approvals. Check each project's documentation to compare features in detail.

Is MagenticLite production ready?

No. The project is described as experimental, so it is not a fully reliable, production-grade automation tool. Use it for experiments and review its Limitations page before relying on it.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

magentic-ui suits people who want to experiment with browser and file automation, approve critical actions, or run agents on small or self-hosted models. It is not the right choice if you need a fully reliable, production-grade automation tool, or if you have no model endpoint to connect.

Install options & downloads