Survey examines evolution of LLM agent memory mechanisms.

A survey published on arXiv, titled "From Storage to Experience: A Survey on the Evolution of LLM Agent Memory Mechanisms," examines the development of memory mechanisms in Large Language Model (LLM)-based agents ¹. The survey aims to bridge the gap between operating system engineering and cognitive science in the field of LLM agent memory ¹.

The survey proposes an evolutionary framework for LLM agent memory mechanisms, formalizing the development process into three stages: Storage, Reflection, and Experience ¹. The Storage stage focuses on trajectory preservation, while Reflection involves trajectory refinement ¹. The Experience stage is characterized by trajectory abstraction ¹.

The authors identify three core drivers of this evolution: the necessity for long-range consistency, the challenges in dynamic environments, and the goal of continual learning ¹. The survey specifically explores proactive exploration and cross-trajectory abstraction as transformative mechanisms in the Experience stage ¹.

The paper was submitted on May 7, 2026, and accepted by ACL 2026 Findings ¹. The authors include Jinghao Luo, Yuchen Tian, Chuxue Cao, Ziyang Luo, Hongzhan Lin, Kaixin Li, Chuyi Kong, Ruichao Yang, and Jing Ma ¹.

The survey aims to provide design principles and a roadmap for the development of next-generation LLM agents by synthesizing different views on memory mechanisms ¹. The paper is available on arXiv ¹.