Agent KB: Leveraging Cross-Domain Experience for Agentic Problem Solving
Xiangru Tang, Tianrui Qin, Tianhao Peng, Ziyang Zhou, Daniel Shao, Tingting Du, Xinming Wei, Peng Xia, Fang Wu, He Zhu, Ge Zhang, Jiaheng Liu, Xingyao Wang, Sirui Hong, Chenglin Wu, Hao Cheng, Chi Wang, Wangchunshu Zhou
2025-07-09
Summary
This paper talks about Agent KB, a system that helps AI agents solve problems better by learning from past experiences gathered by different agents across various tasks and domains through a shared knowledge base.
What's the problem?
The problem is that AI agents usually work alone and cannot learn from each other's past successes or mistakes, which means they often have to solve similar problems repeatedly without benefiting from earlier solutions.
What's the solution?
The researchers created Agent KB with a hierarchical structure that stores knowledge at different levels, such as detailed execution steps, summarized workflows, and high-level strategies. Agents use a Reason-Retrieve-Refine process to understand a new problem, find relevant past experiences, and adapt those solutions to the current task.
Why it matters?
This matters because it allows AI agents to reuse knowledge and strategies from different sources, making them smarter and more efficient at solving complex problems, especially when dealing with new or different tasks.
Abstract
Agent KB, a hierarchical experience framework, enhances problem-solving success rates across different agents by enabling cross-agent knowledge transfer through a Reason-Retrieve-Refine pipeline.