MetaAgent: Automatically Building Multi-Agent System based on Finite State Machine
Keywords: LLM Agent, Multi-Agent System
TL;DR: We introduce a new framework that can automatically produce multi-agent system base on finite state machine
Abstract: Large Language Models (LLMs) can solve various practical tasks via a multi-agent system. However, existing human-designed multi-agent systems can only adapt to a limited number of pre-defined scenarios. Current auto-designed methods also have several drawbacks, including no tool support, reliance on in-bag training, and inflexible communication structure. Therefore, we propose \textbf{MetaAgent}, a novel framework to automatically generate a multi-agent system based on a finite state machine. Given a task description, MetaAgent will design a multi-agent system and polish it through self-generated test queries. When the multi-agent system is deployed, the finite state machine, which supports the traceback and is more suitable for tool-using, will control the process to handle every case in the task domain. To evaluate our framework, we conduct experiments on both practical tasks and basic NLP tasks, the results indicate that the generated multi-agent system surpasses other auto-designed methods and can achieve a comparable performance with the human-designed multi-agent system which is polished for those specific tasks.
Primary Area: foundation or frontier models, including LLMs
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 12996
Loading