LLM Agent Memory

LLM: Large Language Model

LLM agent memory studies how systems preserve useful facts, user preferences, task state, and reusable procedures beyond the current context window, so they can act consistently across sessions and long workflows. Common approaches include summarizing or retaining raw episodes, structured temporal or graph stores, retrieval-augmented memory, hierarchical episodic and procedural tiers, and learned policies for writing, consolidation, selection, and forgetting. Work is tested on long-horizon dialogue and personalization benchmarks such as LoCoMo and LongMemEval, as well as coding, web, tool-use, and embodied tasks; newer evaluations probe multi-turn learning, changing facts, and whether agents actually use retrieved evidence. The central tension is that compression and selective retrieval save tokens but can discard evidence or introduce stale, conflicting, biased, or poisoned memories, while larger stores and elaborate retrieval add cost and raise privacy and governance concerns.

Latest papers

376 papers71 papers in the last four weeks, up 51% on the four weeks before. 0.4% of all new papers.

Memex(RL): Scaling Long-Horizon LLM Agents via Indexed Experience Memory
Mar 4, 2026Zhenting Wang, Huancheng Chen, Jiayun Wang +1LLM Agent Memory Memory Management for LLM Agents
Center for Advanced AI, Accenture
Diagnosing Retrieval vs. Utilization Bottlenecks in LLM Agent Memory
Mar 2, 2026Boqin Yuan, Yue Su, Kun YaoLLM Agent Memory Agent Memory Management
University of California, San Diego · Carnegie Mellon University · University of North Carolina
NextMem: Towards Latent Factual Memory for LLM-based Agents
Feb 26, 2026Zeyu Zhang, Rui Li, Xiaoyan Zhao +4Quantization LLM Agent Memory
Renmin University of China · The Chinese University of Hong Kong · National University of Singapore +1
Structurally Aligned Subtask-Level Memory for Software Engineering Agents
Feb 25, 2026Kangning Shen, Jingyuan Zhang, Chenxi Sun +2LLM Agent Memory Agent Memory
Kuaishou Technology, Beijing, China
ActionEngine: From Reactive to Programmatic Web Agents via State Machine Memory
Feb 24, 2026Hongbin Zhong, Fazle Faisal, Luis França +4Agent Memory LLM Agent Memory
Georgia Tech USA · Microsoft Research USA
OpenSage: Self-programming Agent Generation Engine
Feb 18, 2026Hongwei Li, Zhun Wang, Qinrun Dai +11LLM Agent Orchestration Self-Evolving Agents
UC Santa Barbara · UC Berkeley · University of Colorado Boulder +4
MemoryArena: Benchmarking Agent Memory in Interdependent Multi-Session Agentic Tasks
Feb 18, 2026Zexue He, Yu Wang, Churan Zhi +11Agent Memory LLM Agent Memory
Stanford University · UCSD · UIUC +3
Agent Memory Below the Prompt: Persistent Q4 KV Cache for Multi-Agent LLM Inference on Edge Devices
Feb 17, 2026Yakov Pyotr ShkolnikovMemory Management for LLM Agents LLM Agent Memory
Evaluating Memory Structure in LLM Agents
Feb 11, 2026Alina Shutova, Alexandra Olenina, Ivan Vinogradov +1LLM Agent Memory LLM Agent Evaluation
HSE University · Yandex · YSDA +1
Toward Efficient Agents: Memory, Tool learning, and Planning
Jan 20, 2026Xiaofang Yang, Lijun Li, Heng Zhou +12LLM Agent Memory AI Agent Benchmarks
Shanghai Artificial Intelligence Laboratory · Fudan University · University of Science and Technology of China +5
Beyond Dialogue Time: Temporal Semantic Memory for Personalized LLM Agents
Jan 12, 2026Miao Su, Yucan Guo, Zhongni Hou +8LLM Agent Memory Conversational Memory
Institute of Computing Technology, Chinese Academy of Sciences · State Key Laboratory of AI Safety · School of Computer Science, University of Chinese Academy of Sciences +1
LLMs Can't Play Hangman: On the Necessity of a Private Working Memory for Language Agents
Jan 11, 2026Davide Baldelli, Ali Parviz, Amal Zouaq +1LLM Agent Memory LLM Agent Reliability
Chandar Research Lab · LAMA-WeST Lab · Mila – Quebec AI Institute +2
Beyond Static Summarization: Proactive Memory Extraction for LLM Agents
Jan 8, 2026Chengyuan Yang, Zequn Sun, Wei Wei +1LLM Agent Memory Persistent Memory for Language Models
State Key Laboratory for Novel Software Technology, Nanjing University, China · National Institute of Healthcare Data Science, Nanjing University, China
CaveAgent: Transforming LLMs into Stateful Runtime Operators
Jan 4, 2026Maohao Ran, Zhenglin Wan, Cooper Lin +21LLM Agent Memory LLM Agents
HKUST · HKBU · NUS +8
SEDM: Selective Self-Evolving Distributed Memory for Agents
Sep 11, 2025Yichen Liu, Anna Min, Yinzhi Wang +14LLM Agent Memory Agent Memory Management
Gradient · Zhejiang University · Rice University
Evaluating Memory in LLM Agents via Incremental Multi-Turn Interactions
Jul 7, 2025Yuanzhe Hu, Yu Wang, Julian McAuleyLLM Agent Memory LLM Agent Evaluation
University of California, San Diego

Mar 4, 2026J/K move · Enter open · S save

Memex(RL): Scaling Long-Horizon LLM Agents via Indexed Experience Memory

Zhenting Wang, Huancheng Chen, Jiayun Wang, Wei Wei

Center for Advanced AI, Accenture

LLM Agent Memory Memory Management for LLM Agents Agent Memory Management LLM Context Management Long-Horizon Agent Tasks RL for LLM Agents

Open paper AskPDF arXiv

Large language model (LLM) agents are fundamentally bottlenecked by finite context windows on long-horizon tasks. As trajectories grow, retaining tool outputs and intermediate reasoning in-context quickly becomes infeasible: the working context becomes prohibitively long, eventually exceeds the context budget, and makes distant evidence harder to use even when it is still present. Existing solutions typically shorten context through truncation or running summaries, but these methods are fundamentally lossy because they compress or discard past evidence itself. We introduce Memex, an indexed experience memory mechanism that instead compresses context without discarding evidence. Memex maintains a compact working context consisting of concise structured summaries and stable indices, while storing full-fidelity underlying interactions in an external experience database under those indices. The agent can then decide when to dereference an index and recover the exact past evidence needed for the current subgoal. We optimize both write and read behaviors with our reinforcement learning framework MemexRL, using reward shaping tailored to indexed memory usage under a context budget, so the agent learns what to summarize, what to archive, how to index it, and when to retrieve it. This yields a substantially less lossy form of long-horizon memory than summary-only approaches. We further provide a theoretical analysis showing the potential of the Memex loop to preserve decision quality with bounded dereferencing while keeping effective in-context computation bounded as history grows. Empirically, on challenging long-horizon tasks, Memex agent trained with MemexRL improves task success while using a significantly smaller working context.