Skip to content
Get packs

MagenticLite

An experimental agent that uses a web browser and your local files to finish multi-step tasks.microsoft/magentic-ui

GitHub Download v0.2.1
Stars
10.1k
Forks
1k
Licence
MIT
Language
Python
Last update
1 months ago
Type
AI model / agent

What it is

MagenticLite is an experimental agent from Microsoft AI Frontiers that can work in a web browser and on your local file system to complete tasks such as web research, form filling, and organizing files. It is designed to run on smaller models rather than frontier-scale ones, pairing an orchestrator model with a browser-use model. You can steer it, approve actions, or take over at any point.

  • Filling out expense forms in a browser
  • Finding prices for recipe ingredients across websites
  • Finding and booking a restaurant
  • Organizing files on your local drive

Great for

  • People who want to automate repetitive browser tasks like form filling or price lookups
  • Users experimenting with agents that combine browser and local file actions
  • Developers interested in running agents on small or self-hosted models
  • Anyone who wants to approve critical actions before an agent takes them

Not ideal for

  • Users who need a fully reliable, production-grade automation tool, since it is described as experimental
  • People without a model endpoint, since the app asks you to connect one and the README points to a separate hosting guide
  • Users on platforms other than macOS or Windows via WSL, since the quick start covers only those, and other platforms are described in a separate installation guide that was not reviewed
Step-by-step guide: how to use MagenticLiteInstall, first run, examples, tips and FAQs
MagenticLite alternativesBrowser Use vs MagenticLite

How to get it

  1. 1

    Install with a package manager

    You'll need: Python 3.12 (via uv) · uv package manager · macOS or Windows with WSL

    Create a project directory

    mkdir magentic-lite && cd magentic-lite

    Create and activate a virtual environment

    uv venv --python=3.12 --seed .venv && source .venv/bin/activate

    Install the latest 0.2.x release from PyPI

    uv pip install "magentic_ui>=0.2.0"

    Run the app

    magentic-ui --port 8081
  2. 2

    Run it

    • Run magentic-ui --port 8081 in your terminal
    • Open http://127.0.0.1:8081/ in a browser
    • Follow the in-app onboarding to connect a model endpoint
    • If you have no model endpoint, see the Model Hosting Guide in the docs
  3. 3

    Get the most out of it

    • Read the Configuration doc for sandbox, agent mode, and tool approval settings, including a full config.yaml example
    • Follow the Model Hosting Guide to set up a model endpoint before first use
    • Check the Troubleshooting doc if you hit common errors
    • Review the Limitations doc to know which tasks it handles poorly
    • Build from source using the Build from source guide if you want to modify it

Downloads

Source code (.zip)main branch · straight from GitHubRelease v0.2.11 files incl. magentic_ui-0.2.1-py3-none-any.whlOfficial website & docswww.microsoft.com/en-us/research/blog/magenticlite-magenticbrain-fara1-5-an-agentic-experience-optimized-for-small-models/

Strengths

  • Runs browser sessions inside a lightweight VM sandbox (Quicksand) to limit access to the rest of your machine
  • Asks for approval before critical actions and lets you take over at any point
  • Open source under the MIT License

Trade-offs

  • Described by the README as experimental, so behavior may change between releases
  • The README links a Limitations document listing tasks it does not handle well, which was not included in the reviewed text
  • Requires a model endpoint to be connected before it can do useful work

How it compares

ProjectStarsTypeLevelUpdatedBest for
magentic-ui (this)10.1kAI model / agentintermediate1 months agoPeople who want to automate repetitive browser tasks like form filling or price lookups
browser-use120kLibraryintermediate4 days agoPython developers who want to prototype an AI agent that operates websites
tambo11.2kLibraryintermediate5 days agoReact developers who want an AI chat to show charts, forms, or boards instead of only text
hindsight57.9kAI model / agentintermediate4 days agoDevelopers building AI agents that need memory across sessions
text-to-cad24.9kAI model / agentintermediate4 days agoProduct designers who want to sketch physical parts by describing them in plain language
SuperAGI17.7kAI model / agentadvanced1.8 years agoPython developers experimenting with autonomous AI agents

Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.