README.md

April 13, 2026 · View on GitHub

Evolution Path of LLM Agent Memory

From Storage to Experience:
A Survey on the Evolution of LLM Agent Memory Mechanisms

Paper GitHub Stars License Last Commit Issues Paper Count PRs Welcome

ACL 2026 (Findings)  |  ICLR 2026 Workshop MemAgents  |  Actively maintained

A curated and continuously updated collection of 140+ papers and 40+ benchmarks
on the evolutionary framework of LLM agent memory mechanisms.

If you find this survey useful, please consider giving us a :star: to stay updated with the latest additions!

We welcome contributions! If you know of a relevant paper we missed, please open an issue or submit a pull request. See our Contributing Guidelines.


Table of Contents


Overview

While memory mechanisms have emerged as the architectural cornerstone of LLM agents, current research remains fragmented between operating system engineering and cognitive science. This theoretical divide prevents a unified view of technological synthesis.

We propose a novel Evolutionary Framework that formalizes the development of LLM agent memory into three progressive stages:

StageCore IdeaKey Transformation
StorageTrajectory PreservationFaithfully recording raw interaction traces via linear, vector, or structured storage
ReflectionTrajectory RefinementActively evaluating and correcting stored memories through introspection, environment feedback, or coordination
ExperienceTrajectory AbstractionCompressing redundant trajectories into transferable heuristic wisdom via cross-trajectory abstraction

Evolution Path

The agent's decision-making process forms a dynamic closed loop enabled by memory. During each task execution cycle, the agent leverages two core capabilities:

  1. Memory Read — Actively retrieves relevant knowledge from the memory bank to supplement the current context.
  2. Memory Write — Records generated interaction sequences as historical trajectories into the memory system.

Building upon this foundation, we conceptualize the memory mechanism as an evolutionary pathway structured into three stages:


Storage (Trajectory Preservation)

Storage serves as the cornerstone of memory evolution, emphasizing the faithful preservation of interaction history.

ParadigmCharacteristicsStrengthsLimitations
LinearToken stream ordered chronologicallyMinimal information loss, maximal logical completenessEarly critical information irreversibly forgotten
VectorEmbeddings in high-dimensional spaceMassive storage capacityHigh retrieval difficulty, limited relevance
StructuredRelational structures (tables, graphs, tiers)Precise operations, multi-hop retrievalRequires schema maintenance, less flexible at scale
Reflection (Trajectory Refinement)

Drivers in Dynamic Environments
Figure 2: Temporal Validity & Causal Structure in Dynamic Environments.

ParadigmCharacteristicsStrengthsLimitations
IntrospectionInternal knowledge for self-evaluationError correction without external feedbackRisk of reinforcing biases
EnvironmentExecution outcomes as refinement signalsGreater adaptability to dynamic environmentsSparse rewards, ambiguous settings
CoordinationMulti-agent collective reflectionReduced hallucination, enriched perspectivesCommunication overhead, memory conflicts
Experience (Trajectory Abstraction)

Cross-Trajectory Abstraction
Figure 3: Overview of Cross-Trajectory Abstraction Techniques.

ParadigmCharacteristicsStrengthsLimitations
ExplicitHuman-readable patterns from trajectory clustersHighly interpretable and editableLacks precision in complex boundaries
ImplicitInternalized into model parameters / latent variablesNear-zero retrieval overheadReduced interpretability, forgetting risk
HybridDynamic "accumulate–internalize" cycleBalances interpretability with efficiencyRequires careful transfer coordination

Paper List

Papers are organized following our three-stage evolutionary framework.

Storage

Linear Storage

Context Window Adaptation

PaperVenueYear
Parallel Context Windows for Large Language ModelsACL2022
Efficient Streaming Language Models with Attention SinksarXiv2023
LLM Maybe LongLM: Self-Extend LLM Context Window Without TuningarXiv2024

Information Sparsification

PaperVenueYear
H2O: Heavy-Hitter Oracle for Efficient Generative Inference of Large Language ModelsarXiv2023
LLMLingua: Compressing Prompts for Accelerated Inference of Large Language ModelsEMNLP2023
Quest: Query-Aware Sparsity for Efficient Long-Context LLM InferencearXiv2024
InfLLM: Training-Free Long-Context Extrapolation for LLMs with an Efficient Context MemoryNeurIPS2024
LightThinker++: From Reasoning Compression to Memory ManagementarXiv2026
Understand and Accelerate Memory Processing Pipeline for Disaggregated LLM InferencearXiv2026
MemBoost: A Memory-Boosted Framework for Cost-Aware LLM InferencearXiv2026

Vector Storage

Semantic Retrieval

PaperVenueYear
Enhancing LLM Intelligence with ARM-RAG: Auxiliary Rationale Memory for Retrieval Augmented GenerationarXiv2023
Larimar: Large Language Models with Episodic Memory ControlarXiv2024
MemLong: Memory-Augmented Retrieval for Long Text ModelingarXiv2024
SelRoute: Query-Type-Aware Routing for Long-Term Conversational Memory RetrievalarXiv2026
MemSifter: Offloading LLM Memory Retrieval via Outcome-Driven Proxy ReasoningarXiv2026
TA-Mem: Tool-Augmented Autonomous Memory Retrieval for LLM in Long-Term Conversational QAarXiv2026
Evoking User Memory: Personalizing LLM via Recollection-Familiarity Adaptive RetrievalarXiv2026

Weighted Retrieval

PaperVenueYear
Generative Agents: Interactive Simulacra of Human BehaviorUIST2023
MemoryBank: Enhancing Large Language Models with Long-Term MemoryarXiv2023

Structured Storage

Tabular Database

PaperVenueYear
ChatDB: Augmenting LLMs with Databases as Their Symbolic MemoryarXiv2023
DB-GPT: Empowering Database Interactions with Private Large Language ModelsarXiv2023
Training a Team of Language Models as Options to Build an SQL-Based MemoryApplied Sciences2025

Tiered Architectures

PaperVenueYear
RecurrentGPT: Interactive Generation of (Arbitrarily) Long TextarXiv2023
SwiftSage: A Generative Agent with Fast and Slow Thinking for Complex Interactive TasksarXiv2023
MemoChat: Tuning LLMs to Use Memos for Consistent Long-Range Open-Domain ConversationarXiv2023
MemGPT: Towards LLMs as Operating SystemsarXiv2023
MemOS: A Memory OS for AI SystemarXiv2025
Pancake: Hierarchical Memory System for Multi-Agent LLM ServingarXiv2026
Multi-Layered Memory Architectures for LLM AgentsarXiv2026

Semantic Graphs

PaperVenueYear
MemLLM: Finetuning LLMs to Use an Explicit Read-Write MemoryarXiv2024
GraphReader: Building Graph-based Agent to Enhance Long-Context Abilities of Large Language ModelsEMNLP2024
AriGraph: Learning Knowledge Graph World Models with Episodic Memory for LLM AgentsIJCAI2024
Codebase-Memory: Tree-Sitter-Based Knowledge Graphs for LLM Code Exploration via MCParXiv2026

Reflection

Introspection

Error Rectification

PaperVenueYear
Reflexion: Language Agents with Verbal Reinforcement LearningNeurIPS2023
Think-in-Memory: Recalling and Post-Thinking Enable LLMs with Long-Term MemoryarXiv2023
Constructing Coherent Spatial Memory in LLM Agents through Graph RectificationarXiv2025
Enhancing LLM Planning Capabilities through Intrinsic Self-CritiquearXiv2025

Dynamic Maintenance

PaperVenueYear
Zep: A Temporal Knowledge Graph Architecture for Agent MemoryarXiv2025
Mem0: Building Production-Ready AI Agents with Scalable Long-Term MemoryarXiv2025
CAMA: A Constructivist View of Agentic Memory for LLM-Based Reading ComprehensionarXiv2025
Memory OS of AI AgentarXiv2025
Mem1: Learning to Synergize Memory and Reasoning for Efficient Long-Horizon AgentsarXiv2025

Knowledge Compression

PaperVenueYear
Coarse-to-Fine Grounded Memory for LLM Agent PlanningarXiv2025
MemOrb: A Plug-and-Play Verbal-Reinforcement Memory Layer for E-Commerce Customer ServicearXiv2025
LEGOMem: Modular Procedural Memory for Multi-Agent LLM Systems for Workflow AutomationarXiv2025
Scaling Long-Horizon LLM Agent via Context-FoldingarXiv2025
DeepAgent: A General Reasoning Agent with Scalable ToolsetsarXiv2025
AgentFold: Long-Horizon Web Agents with Proactive Context ManagementarXiv2025
Structured Distillation for Personalized Agent Memory: 11x Token ReductionarXiv2026
Trajectory-Informed Memory Generation for Self-Improving Agent SystemsarXiv2026
ParamMem: Augmenting Language Agents with Parametric Reflective MemoryarXiv2026
MEMO: Memory-Augmented Model Context Optimization for Multi-Agent LLM GamesarXiv2026

Environment

Environment Modeling

PaperVenueYear
CLIN: A Continually Learning Language Agent for Rapid Task Adaptation and GeneralizationarXiv2023
Explicit Memory Learning with Expectation MaximizationEMNLP2024
DEAL: Enhancing Agent Learning through World Dynamics ModelingarXiv2024
ToolMem: Enhancing Multimodal Agents with Learnable Tool Capability MemoryarXiv2025
Preference-Aware Memory Update for Long-Term LLM AgentsarXiv2025
Aligning Progress and Feasibility: A Neuro-Symbolic Dual Memory FrameworkarXiv2026
OBLIVION: Self-Adaptive Agentic Memory Control through Decay-Driven ActivationarXiv2026
Adaptive Memory Admission Control for LLM AgentsarXiv2026
Governing Evolving Memory in LLM Agents: The SSGM FrameworkarXiv2026

Decision Optimization

PaperVenueYear
Self-Goal: Your Language Agents Already Know How to Achieve High-Level GoalsarXiv2024
Memory-R1: Enhancing Large Language Model Agents to Manage and Utilize Memories via Reinforcement LearningarXiv2025
Memory-Driven Self-Improvement for Decision Making with Large Language ModelsarXiv2025

Coordination

Multi-dimensional Calibration

PaperVenueYear
Reflective Multi-Agent Collaboration Based on Large Language ModelsNeurIPS2024
GraphCoGent: Mitigating LLMs' Working Memory Constraints via Multi-Agent CollaborationarXiv2025
MAR: Multi-Agent Reflexion Improves Reasoning Abilities in LLMsarXiv2025
MIRIX: Multi-Agent Memory System for LLM-Based AgentsarXiv2025
Narrative Memory in Machines: Multi-Agent Arc Extraction in Serialized TVarXiv2025
GEMS: Agent-Native Multimodal Generation with Memory and SkillsarXiv2026
Scaling Teams or Scaling Time? Memory Enabled Lifelong Learning in Multi-Agent SystemsarXiv2026
Collaborative Multi-Agent Optimization for Personalized Memory SystemarXiv2026

Experience

Explicit Experience

Heuristic Guidelines

PaperVenueYear
Dynamic Cheatsheet: Test-Time Learning with Adaptive MemoryarXiv2025
G-Memory: Tracing Hierarchical Memory for Multi-Agent SystemsarXiv2025
Agent KB: Leveraging Cross-Domain Experience for Agentic Problem SolvingarXiv2025
SWE-Exp: Experience-Driven Software Issue ResolutionarXiv2025
ArcMemo: Abstract Reasoning Composition with Lifelong LLM MemoryarXiv2025
SEDM: Scalable Self-Evolving Distributed Memory for AgentsarXiv2025
ReasoningBank: Scaling Agent Self-Evolving with Reasoning MemoryarXiv2025
LightMem: Lightweight and Efficient Memory-Augmented GenerationarXiv2025
Learning from Supervision with Semantic and Episodic MemoryarXiv2025
FLEX: Continuous Agent Evolution via Forward Learning from ExperiencearXiv2025
Remember Me, Refine Me: A Dynamic Procedural Memory FrameworkarXiv2025
LoongFlow: Directed Evolutionary Search via Plan-Execute-SummarizearXiv2025
AutoAgent: Evolving Cognition and Elastic Memory Orchestration for Adaptive AgentsarXiv2026
MemPO: Self-Memory Policy Optimization for Long-Horizon AgentsarXiv2026
Towards Autonomous Memory AgentsarXiv2026
Structurally Aligned Subtask-Level Memory for Software Engineering AgentsarXiv2026

Procedural Primitives

PaperVenueYear
Agent Workflow MemoryarXiv2024
Inducing Programmatic Skills for Agentic TasksarXiv2025
Automated Skill Discovery for Language Agents through Exploration and Iterative FeedbackarXiv2025
MemP: Exploring Agent Procedural MemoryarXiv2025
PolySkill: Learning Generalizable Skills through Polymorphic AbstractionarXiv2025
CASCADE: Cumulative Agentic Skill Creation through Autonomous Development and EvolutionarXiv2025
AccelOpt: A Self-Improving LLM Agentic System for AI Accelerator Kernel OptimizationarXiv2025
Contextual Experience Replay for Self-Improvement of Language AgentsACL2025
MemEvolve: Meta-Evolution of Agent Memory SystemsarXiv2025
Youtu-Agent: Scaling Agent Productivity with Hybrid Policy OptimizationarXiv2025

Implicit Experience

Latent Modulation

PaperVenueYear
Titans: Learning to Memorize at Test TimearXiv2024
MemGen: Weaving Generative Latent Memory for Self-Evolving AgentsarXiv2025
LatentEvolve: Self-Evolving Test-Time Scaling in Latent SpacearXiv2025
Tell Me What To Learn: Generalizing Neural Memory to be Controllable in Natural LanguagearXiv2026
ReMem-VLA: Empowering Vision-Language-Action Model with Memory via Dual-Level Recurrent QueriesarXiv2026
Think While Watching: Online Streaming Segment-Level Memory for Multi-Turn Video ReasoningarXiv2026

Parameter Internalization

PaperVenueYear
AgentRefine: Enhancing Agent Generalization through Refinement TuningarXiv2025
Memento No More: Coaching AI Agents to Master Multiple Tasks via Hints InternalizationarXiv2025
Group-in-Group Policy Optimization for LLM Agent TrainingarXiv2025
Agent Lightning: Train Any AI Agents with Reinforcement LearningarXiv2025
VerlTool: Towards Holistic Agentic Reinforcement Learning with Tool UsearXiv2025
From Correction to Mastery: Reinforced Distillation of Large Language Model AgentsarXiv2025
Agent Learning via Early ExperiencearXiv2025
Analyzing and Internalizing Complex Policy Documents for LLM AgentsarXiv2025
Internalizing World Models via Self-Play Finetuning for Agentic RLarXiv2025
AgentEvolver: Towards Efficient Self-Evolving Agent SystemarXiv2025
Agent-R1: Training Powerful LLM Agents with End-to-End Reinforcement LearningarXiv2025
Rethinking Expert Trajectory Utilization in LLM Post-TrainingarXiv2025
Guided Self-Evolving LLMs with Minimal Human SupervisionarXiv2025
End-to-End Test-Time Training for Long ContextarXiv2025

Hybrid Experience

Experience Transfer

PaperVenueYear
ReasoningBank: Scaling Agent Self-Evolving with Reasoning MemoryarXiv2025
EvolveR: Self-Evolving LLM Agents through an Experience-Driven LifecyclearXiv2025
Exploratory Memory-Augmented LLM Agent via Hybrid On- and Off-Policy OptimizationICLR2026
Exploratory Memory-Augmented LLM Agent via Hybrid On- and Off-Policy OptimizationICLR2026
ByteRover: Agent-Native Memory Through LLM-Curated Hierarchical ContextarXiv2026
MemFactory: Unified Inference & Training Framework for Agent MemoryarXiv2026
OMNI-SIMPLEMEM: Autoresearch-Guided Discovery of Lifelong Multimodal Agent MemoryarXiv2026
GAM-RAG: Gain-Adaptive Memory for Evolving Retrieval in RAGarXiv2026
MEMRERANK: Preference Memory for Personalized Product RerankingarXiv2026

Benchmarks & Datasets

Storage Stage

BenchmarkVenueYearFocus
HotpotQAEMNLP2018Multi-hop cross-document reasoning
LongBencharXiv2023Bilingual, multitask long-context understanding
MemoryBankarXiv2023Long-term dialog memory
Multimodal Needle in a HaystackNAACL2024Long-context multimodal retrieval
LongBench v2arXiv2024Realistic long-context reasoning
RULERarXiv2024Effective context window measurement
BABILongarXiv2024Synthetic long-context reasoning-in-a-haystack
DialSimarXiv2024Real-time long-term dialogue simulation
Evaluating Very Long-Term Conversational Memory of LLM AgentsarXiv2024Very long-term conversational memory
MADial-BenchNAACL2024Memory-augmented dialogue
HELMETarXiv2024Comprehensive long-context evaluation
Explicit vs. Implicit MemoryarXiv2025Multi-hop personalized reasoning
Evaluating the Long-Term Memory of Large Language ModelsACL2025Long-term memory evaluation protocol
Beyond a Million TokensarXiv2025Ultra-long memory (>1M tokens)
LoCoBench-AgentarXiv2025Long-context software engineering agents
MemoryRewardBencharXiv2026Reward models for long-term memory management
AgentLongBencharXiv2026Dynamic long-context agent evaluation

Reflection Stage

BenchmarkVenueYearFocus
Personalized Large Language Model Assistant with Evolving Conditional MemoryCOLING2023Evolving conditional memory
PerLTQAarXiv2024Personal long-term memory QA
Evaluating Very Long-Term Conversational Memory of LLM AgentsarXiv2024Conversational memory consistency
On the Multi-Turn Instruction Following for Conversational Web AgentsACL2024Web agent multi-turn memory
SHAREarXiv2024Shared memory in dialogue
MinervaarXiv2025Programmable memory read-write tests
Personalized Preference FollowingarXiv2025User preference memory
Multi-Session Personalized ConversationarXiv2025Multi-session implicit reasoning
PersonaMem-v2arXiv2025Implicit user persona memory
Mem-PALarXiv2025Personalized long-term dialogue
WebChoreArenaarXiv2025Web agent intermediate state memory
StoryBencharXiv2025Long-term narrative tracking
Evaluating Memory in LLM Agents via Incremental Multi-Turn InteractionsarXiv2025Memory retention across turns
LLM Self-Awareness via Internal CircuitsarXiv2025Internal memory limit awareness
HaluMemarXiv2025Hallucination in memory systems
ConvoMem BenchmarkarXiv2025Conversational memory (75K+ QA pairs)
KnowMe-BencharXiv2026Person understanding from narratives
RealMemarXiv2026Real-world project-oriented memory
StructMemEvalarXiv2026Memory structure organization
Beyond the Context Window: Cost-Performance Analysis of Fact-Based Memory vs. Long-Context LLMsarXiv2026Fact-based memory vs. long-context cost tradeoff
Diagnosing Retrieval vs. Utilization Bottlenecks in LLM Agent MemoryarXiv2026Write strategy x retrieval method diagnostics

Experience Stage

BenchmarkVenueYearFocus
StreamBencharXiv2024Continuous improvement over task streams
LifelongAgentBencharXiv2025Lifelong learning agents
MEMTRACKNeurIPS Workshop2025Multi-platform state tracking
MemoryBencharXiv2025Memory and continual learning metrics
Evo-MemoryarXiv2025Test-time self-evolving memory
MemoryArenaarXiv2026Interdependent multi-session agentic tasks
AMA-BencharXiv2026Long-horizon agentic memory

Citation

If you find this survey useful in your research, please consider citing our paper:

@inproceedings{luo2026from,
  title   = {From Storage to Experience: A Survey on the Evolution of LLM Agent Memory Mechanisms},
  author  = {Luo, Jinghao and Tian, Yuchen and Cao, Chuxue and Luo, Ziyang and Lin, Hongzhan and Li, Kaixin and Kong, Chuyi and Yang, Ruichao and Ma, Jing},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year    = {2026}
}

The BibTeX entry will be updated with the official ACL 2026 proceedings metadata once available.


Contributing

We welcome contributions from the community! Please see our Contributing Guidelines for how to:

  • Suggest new papers
  • Report broken links
  • Propose new categories

This project is licensed under the MIT License.