background-pattern

Comprehensive Outline of Large Language Model-based Multi-Agent Research

This project presents an interactive eBook that compiles an extensive collection of research papers on large language model (LLM)-based multi-agent systems. Organized into multiple chapters and continuously updated with significant research, it strives to provide a comprehensive outline for both researchers and enthusiasts in the field. We welcome ongoing contributions to expand and enhance this resource.

Initiated by the ChatDev Group at Tsinghua University.

Cover
background-pattern

Multi-Agent Directions

Multi-agent systems are currently classified into two categories based on whether the agents are designed to achieve specific task goals under external human instructions: task-solving-oriented systems and social-simulation-oriented systems.

  • Task Solving
  • Social Simulation
Comprehensive Resources

Task solving-oriented multi-agent systems employ autonomous agents working collaboratively to tackle complex problems. Cutting-edge research in this direction revolves around three primary areas: facilitating communication among agents, designing effective organizational structures for interaction, and exploring how agents co-evolve over time.

Dataset cover
Community Driven

Social simulation-oriented multi-agent systems concentrate on modeling and analyzing the social behaviors of agents, offering valuable insights into human dynamics and enhances the ability to analyze or predict social phenomena.

Dataset cover
Dive into Each Chapter

This ebook contains research papers on the multi-agent layer and above, organized into multiple chapters based on proposed core technologies. Let's dive into each section.

Systems cover

§1: Communication

facilitating agent communication

Read
Benchmark cover

§2: Organization

organizing agents effectively

Read
Dataset cover

§3: Evolution

growing capabilities over time

Read
Systems cover

§4: Simulation

simulating societal dynamics

Read

Learn More

In addition to the aforementioned resources, we also feature recent research from our lab. If you find our work of interest, we invite you to read, extend, or collaborate.

iAgents

Bijective Social Networks of Humans and Agents

PDF IconPaper GitHub IconCode

IoA

Networking Heterogeneous Agents

PDF IconPaper GitHub IconCode

ChatDev

Multi-Agent Collaboration for Software Development

PDF IconPaper GitHub IconCode

AgentVerse

General-Purpose Multi-Agent Framework

PDF IconPaper GitHub IconCode

Co-Learning

Cross-Task Experience Co-Leaning for Mutual Growth

PDF IconPaper GitHub IconCode

Co-Evolving

Continuous Experience Refinement over Time

PDF IconPaper GitHub IconCode

MacNet

Exploring Collaborative Scaling Law

PDF IconPaper GitHub IconCode

CTC

Cross-Team Multi-Agent Orchestration

PDF IconPaper GitHub IconCode

ChatEval

Communication for Automated Evaluation

PDF IconPaper GitHub IconCode

AutoForm

Finding Effective Communication Protocals

PDF IconPaper GitHub IconCode

Frequently Asked Questions

This ebook gathers leading research on LLM-powered multi-agent systems since 2023, categorized by key perspectives in the field. As this area rapidly evolves, updates will be ongoing.

We encourage open-source collaboration on this project. You can contribute by submitting a pull request with detailed metadata for notable papers in the table.

You can download all ebook content in CSV format directly from here.

Initiated by the ChatDev Group, Tsinghua University
Contact us via qianc62@gmail.com
Total PV