"Multi-Agent Research Outline" is an interactive e-book dedicated to compiling research papers on large language model (LLM)-based multi-agent systems. This e-book was initiated by the ChatDev team at Tsinghua University, aiming to organize and present cutting-edge research in this field.
Purpose This e-book is used to systematically review research on LLM-driven multi-agent systems, helping readers quickly understand the core directions, key technologies, and latest progress in this field. It is suitable for reference by researchers, developers, and learners interested in multi-agent systems.
Main Capabilities The e-book divides multi-agent systems into two major categories: task-solving-oriented systems and social simulation-oriented systems. Task-solving-oriented systems use autonomous agents to collaborate in solving complex problems, with research focuses including promoting communication between agents, designing effective interaction organization structures, and exploring how agents co-evolve over time. Social simulation-oriented systems focus on modeling and analyzing the social behavior of agents, providing insights for understanding human dynamics and predicting social phenomena.
The e-book is organized into multiple chapters according to the proposed core technologies, covering: promoting agent communication, effectively organizing agents, growing capabilities over time, and simulating social dynamics. In addition, it also includes recent research from the laboratory, including resource-aware multi-agent collaboration, dynamically organizing agents through reinforcement learning, agent-oriented generalizable reward models, enhancing agent communication efficiency, bijective social networks between humans and agents, multi-agent teams for graph analysis, heterogeneous agent networking, multi-agent collaboration in software development, general multi-agent frameworks, cross-task experience co-learning for mutual growth, continuous experience refinement over time, exploring collaboration scaling laws, cross-team multi-agent orchestration, communication for automatic evaluation, finding effective communication protocols, and more.
How to Use Users can access the e-book through a web page and browse the content of each chapter. The e-book supports open-source collaboration, and users can contribute important papers by submitting pull requests containing detailed metadata. All e-book content can be downloaded directly in CSV format.
Applicable Scenarios It is suitable for scenarios such as academic research, technical investigation, and project development reference, especially occasions where it is necessary to understand the latest research trends in LLM multi-agent systems.
Limitations The e-book mainly includes research since 2023. Due to the rapid development of the field, the content will continue to be updated. Currently, no other language versions or detailed reading guides are provided.
