It is an open-source tool on Microsoft AutoGen that can help researchers and developers create agentic applications that can handle complicated, multi-step tasks independently.

Agentic AI is the way of the future. Microsoft Research created the generalist multi-agent system Magentic-One, which can solve open-ended problems in a number of domains.

Magentic-One, an open-source AI tool on Microsoft AutoGen, aims to help academics and developers create agentic apps that can handle complicated, multi-step activities on their own.

In order to accomplish various subtasks, a lead orchestrator agent collaborates with other specialized agents in Magentic-One's modular multi-agent architecture. These agents include ComputerTerminal for running code, WebSurfer for navigating websites, FileSurfer for managing files, and Coder for programming activities.

Because of this separation of duties, Magentic-One can manage files, code, and browse the web, which makes it appropriate for a variety of software engineering, data analysis, and scientific research applications.

The purpose of Magentic-One is to serve as a scalable and adaptable substitute for single-agent systems. It permits agents to be added or removed without changing the fundamental architecture of the system.

By enabling agents to be added or removed without affecting other agents or the architecture as a whole, Magentic-One's plug-and-play design further supports simple adaptation and extensibility, the company claims, in contrast to single-agent systems, which often suffer from constrained and rigid workflows.

Microsoft AI framework enables the integration of multiple large language models (LLMs) and short language models (SLMs) to meet specific cost and performance requirements, thereby enabling this flexibility. For some jobs, the system is being evaluated with OpenAI's o1-preview and GPT-4o, although it is yet model-agnostic.

Microsoft released AutoGenBench, a tool for evaluating agentic performance on benchmarks such as GAIA, AssistantBench, and WebArena, in order to examine the efficacy of Magentic-One. According to Microsoft data from October 2024, Magnetic-One has demonstrated competitive outcomes versus state-of-the-art methodologies in several benchmarks, which include multi-step planning and tool usage.