Zhishang Xiang, Yunbo Tang, Zhongqiao Liu, Junkai Chen, Zerui Chen, Qingguo Hu, Yuxuan Zhang, William Grant · Agent Memory State

EvoMemState: Evolving Memory State for Personal Agents

EvoMemState is a state-centric memory framework for LLM agents that goes beyond storing independent facts. It adds Relational States that track how facts remain valid and relate over time, and a Reasoning State, generated by an RL-trained policy, that decides how those facts apply to the current request. The authors report consistent gains over existing memory systems on long-term dialogue and memory-use benchmarks, in retrieval completeness and response accuracy. Code is available on GitHub under the XMUDeepLIT organisation.