# Microsoft Research

Magentic-One is a general-purpose multi-agent system developed by Microsoft Research, designed to tackle open-ended internet and file tasks. This system represents an important step in the evolution of artificial intelligence towards agent systems that can handle complex, multi-step tasks encountered in various work and life situations. Magentic-One utilizes a main agent called the Orchestrator, responsible for planning, tracking progress, and replanning when necessary, while guiding other specialized agents to perform tasks such as operating a web browser, navigating local files, or writing and executing Python code. Magentic-One demonstrates performance comparable to state-of-the-art technologies across various challenging agent benchmarks without requiring modifications to its core capabilities or architecture.

OmniParser is a method developed by the Microsoft Research team for parsing user interface screenshots. It significantly enhances the capability of vision-based language models (like GPT-4V) to generate accurate interface interactions by recognizing interactive icons and understanding the semantics of various elements in screenshots. This technology utilizes finely tuned detection and description models to parse interactive areas in screenshots and extract functional semantics, outperforming baseline models in multiple benchmark tests. OmniParser can be utilized as a plugin with other visual language models to improve their performance.

GraphRAG (Graphs + Retrieval Augmented Generation) is a technology that enhances the understanding of textual datasets by combining text extraction, network analysis, and prompts and summaries from large language models (LLM). Set to be open-sourced on GitHub as part of Microsoft's research projects, this technology aims to enhance the processing and analysis capabilities of textual data through advanced algorithms.