Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Rethinking Memory Mechanisms of Foundation Agents in the Second Half: A Survey

Wei-Chieh Huang, Weizhi Zhang, Yueqing Liang, Yuanchen Bei, Yankai Chen, Tao Feng, Xinyu Pan, Zhen Tan, Yu Wang, Tianxin Wei, Shanglin Wu, Ruiyao Xu, Liangwei Yang, Rui Yang, Wooseong Yang, Chin-Yuan Yeh, Hanrong Zhang, Haozhen Zhang, Siqi Zhu, Henry Peng Zou, Wanjia Zhao, Song Wang, Wujiang Xu, Zixuan Ke, Zheng Hui, Dawei Li, Yaozu Wu, Langzhou He, Chen Wang, Xiongxiao Xu, Baixiang Huang, Juntao Tan, Shelby Heinecke, Huan Wang, Caiming Xiong, Ahmed A. Metwally, Jun Yan, Chen-Yu Lee, Hanqing Zeng, Yinglong Xia, Xiaokai Wei, Ali Payani, Yu Wang, Haitong Ma, Wenya Wang, Chenguang Wang, Yu Zhang, Xin Wang, Yongfeng Zhang, Jiaxuan You, Hanghang Tong, Xiao Luo, Xue Liu, Yizhou Sun, Wei Wang, Julian McAuley, James Zou, Jiawei Han, Philip S. Yu, Kai Shu

TL;DR

This survey provides a unified view of foundation agent memory along three dimensions: memory substrate, cognitive mechanism, and memory subject (agent- and user-centric), and examines how memory is instantiated and operated under different agent topologies and highlights learning policies over memory operations.

Abstract

The research of artificial intelligence is undergoing a paradigm shift from prioritizing model innovations over benchmark scores towards emphasizing problem definition and rigorous real-world evaluation. As the field enters the "second half," the central challenge becomes real utility in long-horizon, dynamic, and user-dependent environments, where agents face context explosion and must continuously accumulate, manage, and selectively reuse large volumes of information across extended interactions. Memory, with hundreds of papers released this year, therefore emerges as the critical solution to fill the utility gap. In this survey, we provide a unified view of foundation agent memory along three dimensions: memory substrate (internal and external), cognitive mechanism (episodic, semantic, sensory, working, and procedural), and memory subject (agent- and user-centric). We then analyze how memory is instantiated and operated under different agent topologies and highlight learning policies over memory operations. Finally, we review evaluation benchmarks and metrics for assessing memory utility, and outline various open challenges and future directions.

Rethinking Memory Mechanisms of Foundation Agents in the Second Half: A Survey

TL;DR

This survey provides a unified view of foundation agent memory along three dimensions: memory substrate, cognitive mechanism, and memory subject (agent- and user-centric), and examines how memory is instantiated and operated under different agent topologies and highlights learning policies over memory operations.

Abstract

The research of artificial intelligence is undergoing a paradigm shift from prioritizing model innovations over benchmark scores towards emphasizing problem definition and rigorous real-world evaluation. As the field enters the "second half," the central challenge becomes real utility in long-horizon, dynamic, and user-dependent environments, where agents face context explosion and must continuously accumulate, manage, and selectively reuse large volumes of information across extended interactions. Memory, with hundreds of papers released this year, therefore emerges as the critical solution to fill the utility gap. In this survey, we provide a unified view of foundation agent memory along three dimensions: memory substrate (internal and external), cognitive mechanism (episodic, semantic, sensory, working, and procedural), and memory subject (agent- and user-centric). We then analyze how memory is instantiated and operated under different agent topologies and highlight learning policies over memory operations. Finally, we review evaluation benchmarks and metrics for assessing memory utility, and outline various open challenges and future directions.
Paper Structure (84 sections, 9 figures, 5 tables)

This paper contains 84 sections, 9 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Large Language Models and Foundation Agents
  4. Memory
  5. Taxonomy of Memory in Foundation Agents
  6. Memory Substrates
  7. External Memory
  8. Vector Index.
  9. Text-record.
  10. Structural Store.
  11. Hierarchical Store.
  12. Internal Memory
  13. Weights.
  14. Latent-State.
  15. KV Cache.
  16. ...and 69 more sections

Figures (9)

  • Figure 1: Roadmap of Foundation Agent Memory. A timeline illustrating the trend of foundation agent memory frameworks, categorized by memory substrates and subjects (user or agent-centric).
  • Figure 2: The taxonomy of Foundation Agent Memory. The Memory Substrate (what form is represented) for foundation agents includes the internal and external memory. In the Memory Cognitive Mechanism (how memory functions) perspective, memories are categorized into episodic, semantic, sensory, working, and procedural memory. Based on the Memory Subject (who is supported), the memory is characterized into user-centric and agent-centric perspectives.
  • Figure 3: Cumulative publication trends of memory-related research in LLM agents (2023 Q1 – 2025 Q4). The plots illustrate the distribution of 218 collected papers across three key dimensions: memory substrate (left), memory cognitive mechanism (middle), and memory subject (right). The shaded region highlights the rapid acceleration of research output observed in 2025.
  • Figure 4: A taxonomy of the Foundation Agent Memory System
  • Figure 5: Connections between memory cognitive mechanism and memory subject. Each cluster corresponds to number of paper of memory cognitive mechanism (sensory, working, semantic, episodic, procedural) for agent- or user-centric memory work, and the area size is proportional to the paper number. The scatter distribution represents the publication time for the paper, from 2023 to 2025.
  • ...and 4 more figures