iAgents: Autonomous Agents for Collaborative Task under Information Asymmetry

Wei Liu1, Chenxi Wang1, Yifei Wang1, Zihao Xie1, Rennai Qiu1, Yufan Dang1, Zhuoyun Du1, Weize Chen1, Cheng Yang2, Chen Qian1
1Tsinghua University, 2Peng Cheng Laboratory, China
Paper arXiv Demo Code Data

iAgents is a platform for assigning each human user a personal agent and enabling them to collaborate on tasks (Speed up by 3x using GPT4o for agent's slow thinking).

Abstract

iAgents is a platform designed to create a world weaved by humans and agents, where each human has a personal agent that can work on their behalf to cooperate with other humans' agents. It is a new paradigm for Large Language Model-powered Multi-Agent Systems. iAgents proactively interact with human users to exchange information, while autonomously communicating with other agents to eliminate information asymmetry and collaborate effectively to accomplish tasks (see our paper).

iAgents features an instant messaging web UI that users can utilize as a conventional chat application, with each user automatically equipped with a personal agent. Messages beginning with '@' are automatically transformed into collaborative task commands, prompting the agents of both chat participants to engage and resolve the task through autonomous communication.

iAgents also features agent cultivation. You can cultivate your agent by clicking on the Agent Admin Panel button and talking with your Agent. You can also use your human feedback data to automatically optimize it. The agent profile will be improved during the cultivation. iAgents intergrates many AI tools such as Jina Reader, LLama Index, for better organizing your information. It supports various of LLM backends including OpenAI, Gemini, Qwen, Deepseek and GLM.

Gallery

Gallery image 1
iAgents can be easily deployed on your laptop
Gallery image 2
iAgents can cultivate your agent by talking with it
Gallery image 3
informative benchmark for evaluating information asymmetry ability of agents
Gallery image 4
iAgents bridge the gap between humans and agents
Gallery image 5
iAgents can be easily integrated with other AI tools
Gallery image 6
iAgents overall pipeline

BibTeX

@inproceedings{liuautonomous,
  title={Autonomous Agents for Collaborative Task under Information Asymmetry},
  author={Liu, Wei and Wang, Chenxi and Wang, YiFei and Xie, Zihao and Qiu, Rennai and Dang, Yufan and Du, Zhuoyun and Chen, Weize and Yang, Cheng and Qian, Chen},
  booktitle={The Thirty-eighth Annual Conference on Neural Information Processing Systems}
}