OxyGent: A Modular Open-Source Framework for Multi-Agent Systems
OxyGent, a newly developed open-source framework, seeks to tackle the challenges of scalability, observability, and autonomous evolution in production-ready multi-agent systems (MAS). Created by a team of researchers, it presents a cohesive Oxy abstraction that integrates agents, tools, LLMs, and reasoning flows as modular atomic components, facilitating a Lego-like approach to system assembly and unobtrusive monitoring. Additionally, it incorporates permission-driven dynamic planning, which substitutes inflexible workflows with execution graphs generated at runtime for adaptable visualizations. The OxyBank evolution engine supports automated data backflow, annotation, and collaborative evolution. Empirical assessments and real-world applications validate OxyGent's effectiveness. The research paper can be found on arXiv with the identifier 2604.25602.
Key facts
- OxyGent is an open-source framework for multi-agent systems.
- It introduces a unified Oxy abstraction for agents, tools, LLMs, and reasoning flows.
- Components are pluggable and atomic, enabling scalable composition.
- Permission-driven dynamic planning generates execution graphs at runtime.
- OxyBank is an AI asset management platform for continuous evolution.
- The framework addresses scalability, observability, and autonomous evolution.
- Empirical evaluations and real-world case studies show robustness.
- The paper is on arXiv with ID 2604.25602.
Entities
Institutions
- arXiv