Autonomous artificial intelligence (AI) agents are designed to accomplish specific objectives by self-guided instructions, automatic memorization, planning, and action. With the advent and prospering of large language models (LLMs), there is an expanding frontier in using LLMs as core controllers for these autonomous entities. However, a unified view that ties together the diverse studies in this field has been lacking. This repository houses a comprehensive and systematic survey that fills this gap, focusing on LLM-based autonomous AI agents in their construction, applications, and evaluation strategies. In particular, we explore the essential components of an AI agent, including a profile module, a memory module, a planning module, and an action module. We further investigate the potential applications in natural and social sciences and introduce methods to evaluate their effectiveness. Challenges and future directions of this field are also discussed. Our paper and this repository aim to serve as a resource for researchers and practitioners alike, providing insights, related references, and continuous updates on this exciting and rapidly evolving field.
- [8/23/2023] The first version of the paper is released on arXiv: A Survey on Large Language Model based Autonomous Agents
- Construction of LLM-based Autonomous Agent
- Applications of LLM-based Autonomous Agent
- Evaluation on LLM-based Autonomous Agent
- More Comprehensive Summarization
| Model | Profile | Memory | Planning | Action | LS | Time | Paper | Code | |
| Operation | Structure | ||||||||
| WebGPT | - | - | - | - | w/ tools | Human feedback | 12/2021 | Paper | - |
| SayCan | - | - | - | w/o feedback | w/ tools | Environment feedback | 04/2022 | Paper | Code |
| MRKL | - | - | - | w/o feedback | w/ tools | - | 05/2022 | Paper | - |
| Inner Monologue | - | - | - | w/ feedback | w/ tools | Environment feedback | 07/2022 | Paper | Code |
| Social Simulacra | GPT-Generated | - | - | - | w/o tools | - | 08/2022 | Paper | - |
| ReAct | - | - | - | w/ feedback | w/ tools | Environment feedback | 10/2022 | Paper | Code |
| REPLUG | - | Read/Write/Reflection | Unified | - | w/o tools | - | 01/2023 | Paper | - |
| DEPS | - | - | - | w/ feedback | w/ tools | Environment feedback | 02/2023 | Paper | - |
| Toolformer | - | - | - | w/o feedback | w/ tools | Example | 02/2023 | Paper | Code |
| Reflexion | - | Read/Write/Reflection | Hybrid | w/ feedback | w/o tools | Environment feedback | 03/2023 | Paper | Code |
| CAMEL | Handcrafting & GPT-Generated | - | - | w/ feedback | w/o tools | - | 03/2023 | Paper | Code |
| API-Bank | - | - | - | w/ feedback | w/ tools | Example | 04/2023 | Paper | Code |
| ViperGPT | - | - | - | - | w/ tools | - | 03/2023 | Paper | Code |
| HuggingGPT | - | Read/Write | Unified | w/o feedback | w/ tools | - | 03/2023 | Paper | Code |
| Generative Agents | Handcrafting | Read/Write/Reflection | Hybrid | w/o feedback | w/o tools | - | 04/2023 | Paper | Code |
| LLM+P | - | - | - | w/o feedback | w/ tools | - | 04/2023 | Paper | - |
| ChemCrow | - | - | - | w feedback | w/ tools | - | 04/2023 | Paper | Code |
| OpenAGI | - | - | - | w feedback | w/ tools | Example | 04/2023 | - | Code |
| AutoGPT | - | Read/Write | Hybrid | w feedback | w tools | Environment feedback | 04/2023 | - | Code |
| SCM | - | Read/Write | Hybrid | - | w/o tools | - | 04/2023 | Paper | Code |
| Socially Alignment | - | Read/Write | Hybrid | - | w/o tools | Example | 05/2023 | Paper | Code |
| GITM | - | Read/Write/Reflection | Hybrid | w/ feedback | w/o tools | Environment feedback | 05/2023 | Paper | Code |
| Voyager | - | Read/Write/Reflection | Unified | w/ feedback | w/o tools | Environment feedback | 05/2023 | Paper | Code |
| Introspective Tips | - | Read/Write/Reflection | Unified | w/ feedback | w/o tools | Example & Environment feedback | 05/2023 | Paper | - |
| RET-LLM | - | Read/Write/Reflection | Unified | - | w/o tools | Example | 05/2023 | Paper | - |
| ChatDB | - | Read/Write/Reflection | Unified | w/ feedback | w/ tools | - | 06/2023 | Paper | - |
| S3 | Dataset alignment | Read/Write/Reflection | Hybrid | w/o feedback | w/o tools | - | 07/2023 | Paper | - |
| ChatDev | Handcrafting | Read/Write/Reflection | Unified | w/ feedback | w/o tools | - | 07/2023 | Paper | Code |
| ToolLLM | - | - | - | w/ feedback | w/ tools | Example | 07/2023 | Paper | Code |
| MemoryBank | - | Read/Write/Reflection | Hybrid | - | w/o tools | - | 07/2023 | Paper | Code |
| MetaGPT | Handcrafting | Read/Write/Reflection | Hybrid | w/ feedback | w/ tools | - | 08/2023 | Paper | Code |
- More papers can be found at More comprehensive Summarization.
| Title | Social Science | Natural Science | Engineering | Time | Paper | Code |
| Drori et al. | - | Science Education | - | 12/2021 | Paper | - |
| SayCan | - | - | Robotics & Embodied AI | 04/2022 | Paper | Code |
| Social Simulacra | Social Simulation | - | - | 08/2022 | Paper | - |
| TE | Psychology | - | - | 08/2022 | Paper | Code |
| Out of One | Political Science and Economy | - | - | 09/2022 | Paper | - |
| Blind Judgement | Jurisprudence | - | - | 01/2023 | Paper | - |
| Horton | Political Science and Economy | - | - | 01/2023 | Paper | - |
| DECKARD | - | - | Robotics & Embodied AI | 01/2023 | Paper | Code |
| Planner-Actor-Reporter | - | - | Robotics & Embodied AI | 02/2023 | Paper | - |
| Generative Agents | Social Simulation | - | - | 04/2023 | Paper | Code |
| SCG | - | - | CS&SE | 04/2023 | Paper | - |
| IGLU | - | - | Civil Engineering | 04/2023 | Paper | - |
| IELLM | - | - | Industrial Automation | 04/2023 | Paper | - |
| ChemCrow | - | Document and Data Management; Documentation, Data Managent; Science Education |
- | 04/2023 | Paper | - |
| Boiko et al. | - | Document and Data Management; Documentation, Data Managent; Science Education |
- | 04/2023 | Paper | - |
| GPT4IA | - | - | Industrial Automation | 04/2023 | Paper | Code |
| Akata et al. | Psychology | - | - | 05/2023 | Paper | - |
| Ziems et al. | Psychology; Political Science and Economy; Research Assistant |
- | - | 05/2023 | Paper | - |
| AgentVerse | Social Simulation | - | - | 05/2023 | - | Code |
| SmolModels | - | - | CS&SE | 05/2023 | - | Code |
| Voyager | - | - | Robotics & Embodied AI | 05/2023 | Paper | Code |
| NLSOM | - | Science Education | - | 05/2023 | Paper | - |
| LLM4RL | - | - | Robotics & Embodied AI | 06/2023 | Paper | - |
| GPT Engineer | - | - | CS&SE | 06/2023 | - | Code |
| Grossman et al. | - | Experiment Assistant; Science Education |
- | 06/2023 | Paper | - |
| Chatlaw | Jurisprudence | - | - | 06/2023 | Paper | Code |
| TaPA | - | - | Robotics & Embodied AI | 07/2023 | Paper | - |
| Ma et al. | Psychology | - | - | 07/2023 | Paper | - |
| Math Agents | - | Science Education | - | 07/2023 | Paper | - |
| SocialAI School | Social Simulation | - | - | 07/2023 | Paper | - |
| Unified Agent | - | - | Robotics & Embodied AI | 07/2023 | Paper | - |
| Wiliams et al. | Social Simulation | - | - | 07/2023 | Paper | - |
| Li et al. | Social Simulation | - | - | 07/2023 | Paper | - |
| S3 | Social Simulation | - | - | 07/2023 | Paper | - |
| Dialogue Shaping | - | - | Robotics & Embodied AI | 07/2023 | Paper | - |
| ToolLLM | - | - | CS&SE | 07/2023 | Paper | Code |
| Chao et al. | Social Simulation | - | - | 08/2023 | Paper | - |
| AgentSims | Social Simulation | - | - | 08/2023 | Paper | Code |
| ChatMOF | - | Document and Data Management; Science Education |
- | 08/2023 | Paper | - |
| MetaGPT | - | - | CS&SE | 08/2023 | Paper | Code |
| Codehelp | - | Science Education | CS&SE | 08/2023 | Paper | - |
| AutoGen | - | Science Education | - | 08/2023 | Paper | - |
- More papers can be found at More comprehensive Summarization.
| Model | Subjective | Objective | Benchmark | Time | Paper | Code |
| WebShop | - | Environment Simulation; Multi-task Evaluation |
✓ | 07/2022 | Paper | Code |
| Social Simulacra | Human Annotation | Social Evaluation | - | 08/2022 | Paper | - |
| TE | - | Social Evaluation | - | 08/2022 | Paper | Code |
| LIBRO | - | Software Testing | - | 09/2022 | Paper | - |
| ReAct | - | Environment Simulation | ✓ | 10/2022 | Paper | Code |
| Out of One, Many | Turing Test | Social Evaluation; Multi-task Evaluation |
- | 02/2023 | Paper | - |
| DEPS | - | Environment Simulation | ✓ | 02/2023 | Paper | - |
| Jalil et al. | - | Software Testing | - | 02/2023 | Paper | Code |
| Reflexion | - | Environment Simulation; Multi-task Evaluation |
- | 03/2023 | Paper | Code |
| IGLU | - | Environment Simulation | ✓ | 04/2023 | Paper | - |
| Generative Agents | Human Annoation; Turing Test |
- | - | 04/2023 | Paper | Code |
| ToolBench | Human Annoation | Multi-task Evalution | ✓ | 04/2023 | Paper | Code |
| GITM | - | Environment Simulation | ✓ | 05/2023 | Paper | Code |
| Two-Failures | - | Multi-task Evalution | - | 05/2023 | Paper | - |
| Voyager | - | Environment Simulation | ✓ | 05/2023 | Paper | Code |
| SocKET | - | Social Evaluation; Multi-task Evaluation |
✓ | 05/2023 | Paper | - |
| Mobile-Env | - | Environment Simulation; Multi-task Evaluation |
✓ | 05/2023 | Paper | Code |
| Clembench | - | Environment Simulation; Multi-task Evaluation |
✓ | 05/2023 | Paper | Code |
| Dialop | - | Social Evaluation | ✓ | 06/2023 | Paper | Code |
| Feldt et al. | - | Software Testing | - | 06/2023 | Paper | - |
| CO-LLM | Human Annoation | Environment Simulation | - | 07/2023 | Paper | Code |
| Tachikuma | Human Annoation | Environment Simulation | ✓ | 07/2023 | Paper | - |
| WebArena | - | Environment Simulation | ✓ | 07/2023 | Paper | Code |
| RocoBench | - | Environment Simulation; Social Evaluation; Multi-task Evaluation |
✓ | 07/2023 | Paper | Code |
| AgentSims | - | Social Evaluation | - | 08/2023 | Paper | Code |
| AgentBench | - | Multi-task Evaluation | ✓ | 08/2023 | Paper | Code |
| BOLAA | - | Environment Simulation; Multi-task Evaluation; Software Testing |
✓ | 08/2023 | Paper | Code |
| Gentopia | - | Isolated Reasoning; Multi-task Evaluation |
✓ | 08/2023 | Paper | Code |
| EmotionBench | Human Annotation | - | ✓ | 08/2023 | Paper | Code |
| PBLLM | - | Software Testing | ✓ | 08/2023 | Paper | - |
- More papers can be found at More comprehensive Summarization.
We are maintaining an interactive table that contains more comprehensive papers related to LLM-based Agents. This table includes details such as tags, authors, publication date, and more, allowing you to sort, filter, and find the papers of interest to you.
- Lei Wang@Paitesanshi
- Chen Ma@Uily
- Xueyang Feng@XueyangFeng
If you find this survey useful, please cite our paper:
@misc{wang2023survey,
title={A Survey on Large Language Model based Autonomous Agents},
author={Lei Wang and Chen Ma and Xueyang Feng and Zeyu Zhang and Hao Yang and Jingsen Zhang and Zhiyuan Chen and Jiakai Tang and Xu Chen and Yankai Lin and Wayne Xin Zhao and Zhewei Wei and Ji-Rong Wen},
year={2023},
eprint={2308.11432},
archivePrefix={arXiv},
primaryClass={cs.AI}
}
We kindly invite contributions to this survey and interactive table. If you have a paper or are aware of relevant research that should be incorporated, please contribute via pull requests, issues, email, or other suitable methods.
We thank the following people for their valuable suggestions and contributions to this survey:
- Yifan Song@Yifan-Song793
- Qichen Zhao@Andrewzh112
- Ikko E. Ashimine@eltociear
If you have any questions or suggestions, please contact us via:
- Email: [email protected]