• By Alex David
  • Tue, 20 May 2025 06:16 PM (IST)
  • Source:JND

At the Build 2025 conference, Microsoft showcased a new “agenetic” web interface called Magenetic-UI meant to assist users in engaging with processes online. As an open-source interface for the web, Magenetic-UI also powers AI agents with form algorithms that allow them to perform basic tasks such as filling out forms, clicking buttons, or traversing websites on the user's behalf. With this release, Microsoft aims to capture the AI-driven automation market—enabling developers, researchers, and businesses to design smarter apps and tools. Given the burgeoning interest in task-orientated AI agents like OpenAI’s GPT agents, Magenetic-UI will help a broader audience not only enhance the capabilities of open-source AI but also shift concepts of web-based browsing and workflows entirely.

Microsoft Launches Magenetic-UI

Microsoft has finally launched the long-awaited Magenetic-UI at its Build 2025 conference. Dubbed the “human-centred web agent,” Magenetic-UI now marks the class of an agenetic parallel processor designed for assisting users in completing online tasks. Unlike traditional autonomous agents, Magenetic-UI works transparently alongside users — showing all its actions in a separate panel, with the capability for users to modify its plans in real time.

For anyone looking to delve into a more engaging experience with AI, Magenetic-UI is now accessible on GitHub and via the Azure AI Foundry Labs, where it has been designed as an open-source project.

Key Capabilities and Features

Magenetic works inside an agent network with a computer as the terminal. It can browse, interact with applications and code, and interact with various files on a virtual machine. Some of the most essential features include:

- Browsing the web to perform tasks like ordering food or booking services

- Writing and executing Python and shell code via Docker

- Analysing and converting files

- Pausing actions, taking feedback in natural language, and accepting plan modifications

- Requesting user approval before irreversible steps like final purchases

- Remembering past tasks to improve future performance

The agenetic System Behind Magenetic-UI

Magenetic-UI is powered by four AI agents that work in parallel:

- Orchestrator – Develops the overall task plan with the user and assigns subtasks

- WebSurfer – Uses a browser to click, scroll, type, and navigate across web pages

- Coder – Executes Python and shell code inside a Docker container

- FireSurfer – Handles file interactions and conversions, also within a secure container

All these agents put together enable the user to take full advantage of the AI in browsing or performing other tasks, while the user remains in full control.

Availability and Access

- Source code: Open-source under the MIT License, available on GitHub

- Cloud version: Hosted on Azure AI Foundry Labs for easy access and testing

Magenetic-UI showcases the ongoing efforts by Microsoft in enhancing AI technology and provides a glimpse into the future with user-agent collaborative interactions.