Reproduce “Memory in the Age of AI Agents” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yuyang Hu, Shichun Liu, Yanwei Yue, Guibin Zhang, Boyang Liu, Fangyi Zhu, Jiahang Lin, Honglin Guo and 39 more
This survey provides a comprehensive analysis of memory systems in AI agents, introducing a novel "Forms–Functions–Dynamics" taxonomy to categorize and understand current research. It clarifies conceptual boundaries with related areas and identifies promising future research directions, establishing a unified framework for the field.
Key insights
- Agent memory differs from LLM memory, RAG, and context engineering by focusing on an external, evolving cognitive state for persistent, adaptive decision-making.
- Memory forms are classified by *where it resides* and *how it's represented* (token-level, parametric, latent), with token-level memory further structured into flat, planar, and hierarchical organizations.
- Memory functions are categorized by *purpose* (factual, experiential, working memory), while dynamics cover the lifecycle of memory via *formation*, *evolution*, and *retrieval* processes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.