A curated list of must-read papers and resources on LLM Agents maintained by ZJUNLP, covering single agent capabilities (memory, planning, tool use), multi-agent collaboration, benchmarks, and mainstream development frameworks.
About LLMAgentPapers#
LLMAgentPapers is an Awesome List resource repository created and maintained by the Natural Language Processing Group at Zhejiang University (ZJUNLP), aiming to systematically organize academic literature, frameworks, and evaluation tools in the LLM Agent field.
Core Value#
- Addressing Information Overload: The LLM Agent field is evolving rapidly with a surge of papers; this repository provides a structured classification index
- Systematic Understanding: Classified by agent capabilities and interaction patterns to help developers quickly build a knowledge framework
- Engineering Reference: Aggregates mainstream Agent frameworks and tool lists for direct engineering selection reference
Paper Classification System#
Overview (Survey)#
LLM Agent field surveys and overall frameworks, e.g., A Survey on Large Language Model based Autonomous Agents
Agent (Single Agent)#
- Personality: Role-playing and personality modeling
- Memory: MemoryBank, HippoRAG, Buffer of Thoughts
- Planning: ReAct, Tree of Thoughts, Reflexion, KnowAgent
- Tool use: WebGPT, Toolformer, ToolLLM, Gorilla
- RL training: Reinforcement learning-based agent training
Multiple Agents (Multi-Agent)#
- Task-Oriented Communication
- Collaborative Exchanges
- Adversarial Interactions
- Casual/Open Conversations
Notable projects: AutoGen, MetaGPT, CAMEL, AgentVerse
Application#
Specific industry or scenario applications such as social simulation, gaming, software development
Resources & Tools#
Benchmarks#
| Name | Description |
|---|---|
| AgentBench | Multi-dimensional Agent capability evaluation |
| WebArena | Real web environment operation evaluation |
| OSWorld | Operating system-level operation evaluation |
| TravelPlanner | Planning capability test benchmark |
Frameworks & Tools#
- Agent with tool: AutoGPT, LangChain, Transformer Agents, WorkGPT, AutoChain, Langroid, GPT Researcher, BMTools, ToolBench, AgentGPT, xlang, Agently, Lagent, ToolEmu, Storm, Comet Opik
- Multi-Agent: CAMEL, GPTeam, AgentVerse, MetaGPT, SocraticAI, AutoGen, Agents
- Others: AutoAgents, GPT Engineer
How to Use#
Quick Access#
Visit the GitHub repository directly to read the README.md and click on categories of interest.
Local Clone#
git clone https://github.com/zjunlp/LLMAgentPapers.git
cd LLMAgentPapers
Contributing#
The repository is open for community contributions. If you find important works missing, submit an Issue or PR following the repository guidelines.
Related Ecosystem#
This repository is part of ZJUNLP's series of Paper Lists:
- Prompt4ReasoningPapers: Focus on reasoning capabilities
- KnowledgeEditingPapers: Focus on model knowledge editing