Memory in the Age of AI Agents
Yuyang Hu, Shichun Liu, Yanwei Yue, Guibin Zhang, Boyang Liu, Fangyi Zhu, Jiahang Lin, Honglin Guo, Shihan Dou, Zhiheng Xi, Senjie Jin, Jiejun Tan, Yanbin Yin, Jiongnan Liu, Zeyu Zhang, Zhongxiang Sun, Yutao Zhu, Hao Sun, Boci Peng, Zhenrong Cheng, Xuanbo Fan, Jiaxin Guo
2025-12-16
Summary
This paper is a comprehensive overview of how 'memory' is being used and developed in artificial intelligence agents powered by large language models, often called foundation models.
What's the problem?
The field of agent memory is growing quickly, but it's become disorganized and confusing. Different researchers are using the term 'memory' in different ways, building systems with varying goals, and evaluating them using different methods. The old ways of categorizing memory, like short-term versus long-term, don't really capture the complexity of what's happening now, making it hard to understand the overall progress and compare different approaches.
What's the solution?
The authors created a detailed survey of current agent memory research. They clarified what 'agent memory' actually means, distinguishing it from similar concepts like simply giving information to a language model or using retrieval-based methods. They then organized the different types of agent memory by looking at *how* the memory is stored (token-level, parametric, or latent), *what* kind of information it holds (facts, experiences, or working information), and *how* the memory changes over time (how it's created, updated, and used). Finally, they provided a list of tools and benchmarks to help developers and pointed out areas where future research is needed.
Why it matters?
This work is important because it provides a common understanding and framework for thinking about agent memory. By organizing the field and identifying key research areas, it will help researchers build more effective and intelligent AI agents, and it lays the groundwork for treating memory as a fundamental building block in future AI systems.
Abstract
Memory has emerged, and will continue to remain, a core capability of foundation model-based agents. As research on agent memory rapidly expands and attracts unprecedented attention, the field has also become increasingly fragmented. Existing works that fall under the umbrella of agent memory often differ substantially in their motivations, implementations, and evaluation protocols, while the proliferation of loosely defined memory terminologies has further obscured conceptual clarity. Traditional taxonomies such as long/short-term memory have proven insufficient to capture the diversity of contemporary agent memory systems. This work aims to provide an up-to-date landscape of current agent memory research. We begin by clearly delineating the scope of agent memory and distinguishing it from related concepts such as LLM memory, retrieval augmented generation (RAG), and context engineering. We then examine agent memory through the unified lenses of forms, functions, and dynamics. From the perspective of forms, we identify three dominant realizations of agent memory, namely token-level, parametric, and latent memory. From the perspective of functions, we propose a finer-grained taxonomy that distinguishes factual, experiential, and working memory. From the perspective of dynamics, we analyze how memory is formed, evolved, and retrieved over time. To support practical development, we compile a comprehensive summary of memory benchmarks and open-source frameworks. Beyond consolidation, we articulate a forward-looking perspective on emerging research frontiers, including memory automation, reinforcement learning integration, multimodal memory, multi-agent memory, and trustworthiness issues. We hope this survey serves not only as a reference for existing work, but also as a conceptual foundation for rethinking memory as a first-class primitive in the design of future agentic intelligence.