Memory in the Age of AI Agents: A Survey

Figure: Overview of agent memory organized by the unified taxonomy of forms, functions, and dynamics. The diagram positions memory artifacts by their dominant form and primary function. It further maps representative systems into this taxonomy to provide a consolidated landscape.

Memory serves as the cornerstone of foundation model-based agents, underpinning their ability to perform long-horizon reasoning, adapt continually, and interact effectively with complex environments.

Despite the explosion of research in this field, the landscape remains highly fragmented, with loosely defined terminologies and inconsistent taxonomies. This repository aims to bridge this gap. We distinguish Agent Memory from related concepts like RAG and Context Engineering, and provide a comprehensive overview through three unified lenses:

• Forms (What Carries Memory?): Categorizing memory by its storage medium—Token-level (explicit & discrete), Parametric (implicit weights), and Latent (hidden states) .
• Functions (Why Agents Need Memory?): Moving beyond simple temporal divisions to a functional taxonomy: Factual (knowledge), Experiential (insights & skills), and Working Memory (active context management) .
• Dynamics (How Memory Evolves?): Dissecting the operational lifecycle into Formation (extraction), Evolution (consolidation & forgetting), and Retrieval (access strategies) .

Through this structure, we hope to provide a conceptual foundation for rethinking memory as a first-class primitive in future agentic intelligence.

Conceptual comparison of Agent Memory with LLM Memory, RAG, and Context Engineering. The diagram illustrates shared technical implementations while highlighting fundamental distinctions: unlike the architectural optimizations of LLM Memory, the static knowledge access of RAG, or the transient resource management of Context Engineering, Agent Memory is uniquely characterized by its focus on maintaining a persistent and self-evolving cognitive state that integrates factual knowledge and experience.

• [2025/12] Hindsight is 20/20: Building Agent Memory that Retains, Recalls, and Reflects. [paper
• [2025/11] O-Mem: Omni Memory System for Personalized, Long Horizon, Self-Evolving Agents. [paper]
• [2025/11] In Prospect and Retrospect: Reflective Memory Management for Long-term Personalized Dialogue Agents. [paper]
• [2025/11] Memoro: Using Large Language Models to Realize a Concise Interface for Real-Time Memory Augmentation. [paper]
• [2025/11] RCR-Router: Efficient Role-Aware Context Routing for Multi-Agent LLM Systems with Structured Memory. [paper]
• [2025/11] Enabling Personalized Long-term Interactions in LLM-based Agents through Persistent Memory and User Profiles. [paper]
• [2025/10] Livia: An Emotion-Aware AR Companion Powered by Modular AI Agents and Progressive Memory Compression. [paper]
• [2025/10] D-SMART: Enhancing LLM Dialogue Consistency via Dynamic Structured Memory And Reasoning Tree. [paper]
• [2025/10] WebWeaver: Structuring Web-Scale Evidence with Dynamic Outlines for Open-Ended Deep Research. [paper]
• [2025/10] CAM: A Constructivist View of Agentic Memory for LLM-Based Reading Comprehension. [paper]
• [2025/10] MovieChat: From Dense Token to Sparse Memory for Long Video Understanding. [paper]
• [2025/10] Pre-Storage Reasoning for Episodic Memory: Shifting Inference Burden to Memory for Personalized Dialogue. [paper]
• [2025/10] LightMem: Lightweight and Efficient Memory-Augmented Generation. [paper]
• [2025/09] Mem-α: Learning Memory Construction via Reinforcement Learning. [paper]
• [2025/09] SGMem: Sentence Graph Memory for Long-Term Conversational Agents. [paper]
• [2025/09] Nemori: Self-Organizing Agent Memory Inspired by Cognitive Science. [paper]
• [2025/09] MOOM: Maintenance, Organization and Optimization of Memory in Ultra-Long Role-Playing Dialogues. [paper]
• [2025/09] Multiple Memory Systems for Enhancing the Long-term Memory of Agent. [paper]
• [2025/09] Semantic Anchoring in Agentic Memory: Leveraging Linguistic Structures for Persistent Conversational Context. [paper]
• [2025/09] Recommender AI Agent: Integrating Large Language Models for Interactive Recommendations. [paper]
• [2025/09] ComoRAG: A Cognitive-Inspired Memory-Organized RAG for Stateful Long Narrative Reasoning. [paper]
• [2025/08] Seeing, Listening, Remembering, and Reasoning: A Multimodal Agent with Long-Term Memory. [paper]
• [2025/08] RoleLLM: Benchmarking, Eliciting, and Enhancing Role-Playing Abilities of Large Language Models. [paper]
• [2025/08] Memory-R1: Enhancing Large Language Model Agents to Manage and Utilize Memories via Reinforcement Learning. [paper]
• [2025/08] Intrinsic Memory Agents: Heterogeneous Multi-Agent LLM Systems through Structured Contextual Memory. [paper]
• [2025/07] MIRIX: Multi-Agent Memory System for LLM-Based Agents. [paper]
• [2025/07] Hierarchical Memory for High-Efficiency Long-Term Reasoning in LLM Agents. [paper]
• [2025/06] G-Memory: Tracing Hierarchical Memory for Multi-Agent Systems. [paper]
• [2025/06] Embodied Agents Meet Personalization: Exploring Memory Utilization for Personalized Assistance. [paper]
• [2025/05] MemGuide: Intent-Driven Memory Selection for Goal-Oriented Multi-Session LLM Agents. [paper]
• [2025/05] Pre-training Limited Memory Language Models with Internal and External Knowledge. [paper]
• [2025/05] SeCom: On Memory Construction and Retrieval for Personalized Conversational Agents. [paper]
• [2025/05] Embodied VideoAgent: Persistent Memory from Egocentric Videos and Embodied Sensors Enables Dynamic Scene Understanding. [paper]
• [2025/05] Human-inspired Episodic Memory for Infinite Context LLMs. [paper]
• [2025/04] Mem0: Building production-ready ai agents with scalable long-term memory. [paper]
• [2025/02] Zep: A Temporal Knowledge Graph Architecture for Agent Memory. [paper]
• [2025/02] A-MEM: Agentic Memory for LLM Agents. [paper]
• [2025/02] Unveiling Privacy Risks in LLM Agent Memory. [paper]
• [2025/02] Mem2Ego: Empowering Vision-Language Models with Global-to-Ego Memory for Long-Horizon Embodied Navigation. [paper]
• [2024/12] AI PERSONA: Towards Life-long Personalization of LLMs. [paper]
• [2024/11] OASIS: Open Agent Social Interaction Simulations with One Million Agents. [paper]
• [2024/10] Memolet: Reifying the Reuse of User-AI Conversational Memories. [paper]
• [2024/10] From Isolated Conversations to Hierarchical Schemas: Dynamic Tree Memory Representation for LLMs. [paper]
• [2024/10] Enhancing Long Context Performance in LLMs Through Inner Loop Query Mechanism. [paper]
• [2024/09] Crafting Personalized Agents through Retrieval-Augmented Generation on Editable Memory Graphs. [paper]
• [2024/07] Arigraph: Learning knowledge graph world models with episodic memory for llm agents. [paper]
• [2024/07] ChatHaruhi: Reviving Anime Character in Reality via Large Language Model. [paper]
• [2024/07] Toward Conversational Agents with Context and Time Sensitive Long-term Memory. [paper]
• [2024/06] Enhancing Long-Term Memory using Hierarchical Aggregate Tree for Retrieval Augmented Generation. [paper]
• [2024/06] Towards Lifelong Dialogue Agents via Timeline-based Memory Management. [paper]
• [2024/05] HippoRAG: Neurobiologically Inspired Long-Term Memory for Large Language Models. [paper]
• [2024/05] Memory Sharing for Large Language Model based Agents. [paper]
• [2024/05] Knowledge Graph Tuning: Real-time Large Language Model Personalization based on Human Feedback. [paper]
• [2024/04] From Local to Global: A Graph RAG Approach to Query-Focused Summarization. [paper]
• [2023/10] MemGPT: Towards LLMs as Operating Systems. [paper]
• [2023/10] GameGPT: Multi-agent Collaborative Framework for Game Development. [paper]
• [2023/10] CALYPSO: LLMs as Dungeon Masters' Assistants. [paper]
• [2023/10] MemGPT: Towards LLMs as Operating Systems. [paper]
• [2023/10] Lyfe Agents: Generative agents for low-cost real-time social interactions. [paper]
• [2023/08] MetaGPT: Meta Programming for A Multi-Agent Collaborative Framework. [paper]
• [2023/08] MemoChat: Tuning LLMs to Use Memos for Consistent Long-Range Open-Domain Conversation. [paper]
• [2023/08] Prompted LLMs as Chatbot Modules for Long Open-domain Conversation. [paper]
• [2023/08] Recursively summarizing enables long-term dialogue memory in large language models. [paper]
• [2023/07] S${}^3$: Social-network Simulation System with Large Language Model-Empowered Agents. [paper]
• [2023/05] RecurrentGPT: Interactive Generation of (Arbitrarily) Long Text. [paper]
• [2023/05] Memorybank: Enhancing large language models with long-term memory. [paper]
• [2023/05] RET-LLM: Towards a general read-write memory for large language models. [paper]
• [2023/04] Generative agents: Interactive simulacra of human behavior. [paper]
• [2023/04] HuaTuo: Tuning LLaMA Model with Chinese Medical Knowledge. [paper]
• [2023/04] SCM: Enhancing Large Language Model with Self-Controlled Memory Framework. [paper]

• [2025/10] Pretraining with hierarchical memories: separating long-tail and common knowledge. [paper]
• [2025/09] MLP Memory: Language Modeling with Retriever-pretrained External Memory. [paper]
• [2025/07] Self-Updatable Large Language Models by Integrating Context into Model Parameters. [paper]
• [2025/06] WISE: Rethinking the Knowledge Memory for Lifelong Model Editing of Large Language Models. [paper]
• [2025/06] CharacterGLM: Customizing Social Characters with Large Language Models. [paper]
• [2025/04] ELDER: Enhancing Lifelong Model Editing with Mixture-of-LoRA. [paper]
• [2025/02] Online Adaptation of Language Models with a Memory of Amortized Contexts. [paper]
• [2024/10] AlphaEdit: Null-Space Constrained Knowledge Editing for Language Models. [paper]
• [2024/09] Neighboring Perturbations of Knowledge Editing on Large Language Models. [paper]
• [2024/04] Character-LLM: A Trainable Agent for Role-Playing. [paper]
• [2023/03] K-Adapter: Infusing Knowledge into Pre-Trained Models with Adapters. [paper]
• [2022/08] Fast Model Editing at Scale. [paper]
• [2021/04] Editing Factual Knowledge in Language Models. [paper]
• [2013/02] ELLA: An Efficient Lifelong Learning Algorithm. [paper]

• [2025/10] Memory$^3$: Language Modeling with Explicit Memory. [paper]
• [2025/08] Towards General Continuous Memory for Vision-Language Models. [paper]
• [2025/03] M+: Extending MemoryLLM with Scalable Long-Term Memory. [paper]
• [2025/02] R3Mem: Bridging Memory Retention and Retrieval via Reversible Compression [paper]
• [2024/07] Efficient Episodic Memory Utilization of Cooperative Multi-Agent Reinforcement Learning. [paper]

• [2025/12] Hindsight is 20/20: Building Agent Memory that Retains, Recalls, and Reflects. [paper
• [2025/11] Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models. [paper]
• [2025/11] FLEX: Continuous Agent Evolution via Forward Learning from Experience. [paper]
• [2025/11] Scaling Agent Learning via Experience Synthesis. [paper]
• [2025/11] UFO2: The Desktop AgentOS. [paper]
• [2025/10] PRINCIPLES: Synthetic Strategy Memory for Proactive Dialogue Agents. [paper]
• [2025/10] Training-Free Group Relative Policy Optimization. [paper]
• [2025/10] ToolMem: Enhancing Multimodal Agents with Learnable Tool Capability Memory. [paper]
• [2025/10] H(^2)R: Hierarchical Hindsight Reflection for Multi-Task LLM Agents. [paper]
• [2025/10] BrowserAgent: Building Web Agents with Human-Inspired Web Browsing Actions. [paper]
• [2025/10] LEGOMem: Modular Procedural Memory for Multi-agent LLM Systems for Workflow Automation. [paper]
• [2025/10] Alita-G: Self-Evolving Generative Agent for Agent Generation. [paper]
• [2025/10] SAGE: Self-evolving Agents with Reflective and Memory-augmented Abilities. [paper]
• [2025/09] ReasoningBank: Scaling Agent Self-Evolving with Reasoning Memory. [paper]
• [2025/09] Memento: Fine-tuning LLM Agents without Fine-tuning LLMs. [paper]
• [2025/08] Memp: Exploring Agent Procedural Memory. [paper]
• [2025/08] SEAgent: Self-Evolving Computer Use Agent with Autonomous Learning from Experience. [paper]
• [2025/07] Agent KB: Leveraging Cross-Domain Experience for Agentic Problem Solving. [paper]
• [2025/07] MemTool: Optimizing short-term memory management for dynamic tool calling in llm agent multi-turn conversations. [paper]
• [2025/06] JARVIS-1: Open-World Multi-Task Agents With Memory-Augmented Multimodal Language Models. [paper]
• [2025/05] Agent Workflow Memory. [paper]
• [2025/05] Darwin Godel Machine: Open-Ended Evolution of Self-Improving Agents. [paper]
• [2025/05] Alita: Generalist Agent Enabling Scalable Agentic Reasoning with Minimal Predefinition and Maximal Self-Evolution. [paper]
• [2025/05] SkillWeaver: Web Agents can Self-Improve by Discovering and Honing Skills. [paper]
• [2025/05] LearnAct: Few-Shot Mobile GUI Agent with a Unified Demonstration Benchmark. [paper]
• [2025/05] Retrieval Models Aren't Tool-Savvy: Benchmarking Tool Retrieval for Large Language Models. [paper]
• [2025/04] Dynamic Cheatsheet: Test-Time Learning with Adaptive Memory. [paper]
• [2025/04] Inducing Programmatic Skills for Agentic Tasks. [paper]
• [2025/03] COLA: A Scalable Multi-Agent Framework For Windows UI Task Automation. [paper]
• [2025/03] Memory-augmented Query Reconstruction for LLM-based Knowledge Graph Reasoning. [paper]
• [2025/02] Buffer of Thoughts: Thought-Augmented Reasoning with Large Language Models. [paper]
• [2025/02] From Exploration to Mastery: Enabling LLMs to Master Tools via Self-Driven Interactions. [paper]
• [2025/02] From RAG to Memory: Non-Parametric Continual Learning for Large Language Models. [paper]
• [2024/12] Planning from Imagination: Episodic Simulation and Episodic Memory for Vision-and-Language Navigation. [paper]
• [2024/11] ExpeL: LLM Agents Are Experiential Learners. [paper]
• [2024/10] RepairAgent: An Autonomous, LLM-Based Agent for Program Repair. [paper]
• [2024/07] Fincon: A synthesized llm multi-agent system with conceptual verbal reinforcement for enhanced financial decision making. [paper]
• [2024/05] COLT: Towards Completeness-Oriented Tool Retrieval for Large Language Models. [paper]
• [2023/08] RecMind: Large Language Model Powered Agent For Recommendation. [paper]
• [2023/07] ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIs. [paper]
• [2023/05] CREATOR: Tool Creation for Disentangling Abstract and Concrete Reasoning of Large Language Models. [paper]
• [2023/03] Reflexion: Language agents with verbal reinforcement learning. [paper]
• [2023/02] Toolformer: Language models can teach themselves to use tools. [paper]

• [2025/11] AgentEvolver: Towards Efficient Self-Evolving Agent System. [paper]
• [2025/10] Agent Learning via Early Experience. [paper]
• [2025/10] Scaling Agents via Continual Pre-training. [paper]
• [2024/10] ToolGen: Unified Tool Retrieval and Calling via Generation. [paper]
• [2023/09] A Machine with Short-Term, Episodic, and Semantic Memory Systems. [paper]
• [2023/08] Retroformer: Retrospective Large Language Agents with Policy Gradient Optimization. [paper]

• [2025/11] Auto-scaling Continuous Memory for GUI Agent. [paper]

• [2025/11] Memory as Action: Autonomous Context Curation for Long-Horizon Agentic Tasks. [paper]
• [2025/11] IterResearch: Rethinking Long-Horizon Agents via Markovian State Reconstruction. [paper]
• [2025/11] MemSearcher: Training LLMs to Reason, Search and Manage Memory via End-to-End Reinforcement Learning. [paper]
• [2025/10] AgentFold: Long-Horizon Web Agents with Proactive Context Management. [paper]
• [2025/10] PRIME: Planning and Retrieval-Integrated Memory for Enhanced Reasoning. [paper]
• [2025/10] Context as Memory: Scene-Consistent Interactive Long Video Generation with Memory Retrieval. [paper]
• [2025/10] DeepAgent: A General Reasoning Agent with Scalable Toolsets. [paper]
• [2025/10] ACON: Optimizing Context Compression for Long-Horizon LLM Agents. [paper]
• [2025/09] ReSum: Unlocking Long-Horizon Search Intelligence via Context Summarization. [paper]
• [2025/07] MemAgent: Reshaping Long-Context LLM with Multi-Conv RL-based Memory Agent. [paper]
• [2024/10] Agent S: An Open Agentic Framework That Uses Computers Like a Human. [paper]

• [2025/04] Various Lengths, Constant Speed: Efficient Language Modeling with Lightning Attention. [paper]
• [2025/02] Efficient Streaming Language Models with Attention Sinks. [paper]

• [2025/11] Time-VLM: Exploring Multimodal Vision-Language Models for Augmented Time Series Forecasting. [paper]
• [2025/11] SoftCoT: Soft Chain-of-Thought for Efficient Reasoning with LLMs. [paper]
• [2025/11] MemoRAG: Boosting Long Context Processing with Global Memory-Enhanced Retrieval Augmentation. [paper]
• [2025/09] MemGen: Weaving Generative Latent Memory for Self-Evolving Agents. [paper]
• [2025/09] Conflict-Aware Soft Prompting for Retrieval-Augmented Generation. [paper]
• [2025/09] MemoryVLA: Perceptual-Cognitive Memory in Vision-Language-Action Models for Robotic Manipulation. [paper]
• [2025/06] MEM$1$: Learning to Synergize Memory and Reasoning for Efficient Long-Horizon Agents. [paper]
• [2025/06] Taking a Deep Breath: Enhancing Language Modeling of Large Language Models with Sentinel Tokens. [paper]
• [2025/05] H2O: Heavy-Hitter Oracle for Efficient Generative Inference of Large Language Models. [paper]
• [2025/05] RazorAttention: Efficient KV Cache Compression Through Retrieval Heads. [paper]
• [2025/04] SnapKV: LLM Knows What You are Looking for Before Generation. [paper]
• [2025/03] LM2: Large Memory Models. [paper]
• [2025/02] Titans: Learning to Memorize at Test Time. [paper]
• [2024/08] Augmenting Language Models with Long-Term Memory. [paper]
• [2024/04] Adapting Language Models to Compress Contexts. [paper]
• [2024/03] Learning to Compress Prompts with Gist Tokens. [paper]
• [2024/03] Scissorhands: Exploiting the Persistence of Importance Hypothesis for LLM KV Cache Compression at Test Time. [paper]
• [2024/03] Focused Transformer: Contrastive Training for Context Scaling. [paper]
• [2023/07] In-Context Autoencoder for Context Compression in a Large Language Model. [paper]
• [2022/08] Memorizing Transformers. [paper]
• [2022/07] XMem: Long-Term Video Object Segmentation with an Atkinson-Shiffrin Memory Model. [paper]