A-Mem: Agentic Memory for LLM Agents
Open accessThe DOI matches its Crossref record; 33 of the 33 references it cites were found in a registry or an index; 11 of the 17 sources its claims cite were checked for support.
Abstract
This paper proposes a novel agentic memory system for LLM agents that can dynamically organize memories in an agentic way following Zettelkasten note-taking principles with dynamic indexing and linking.
Overview
CoChat Summary
TL;DR
A-MEM is an agentic memory system for LLM agents that, inspired by the Zettelkasten method, builds structured notes with LLM-generated keywords, tags, and contextual descriptions, then autonomously links and evolves them as new memories arrive. Across six foundation models on LoCoMo and DialSim, it outperforms LoCoMo, ReadAgent, MemoryBank, and MemGPT while using far fewer tokens.
Key findings
- On LoCoMo with GPT-4o-mini, A-MEM reaches 27.02 F1 / 20.09 BLEU-1 on Multi Hop and 45.85 / 36.67 on Temporal, versus MemGPT's 26.65 / 17.72 and 25.52 / 19.44; A-MEM's average ranking is 1.2 versus 2.4 for both LoCoMo and MemGPT.
- On DialSim, A-MEM scores F1 3.45, BLEU-1 3.37, ROUGE-L 3.54, ROUGE-2 3.60, METEOR 2.05, SBERT 19.51, versus LoCoMo (2.55, 3.13, 2.75, 0.90, 1.64, 15.76) and MemGPT (1.18, 1.07, 0.96, 0.42, 0.95, 8.54) — a 35% F1 improvement over LoCoMo and 192% over MemGPT.
- Ablation with GPT-4o-mini: removing both link generation and memory evolution drops Multi Hop F1 to 9.65; removing only memory evolution gives 21.35; full A-MEM gives 27.02.
- Cost/scaling: ~1,200 tokens per memory operation (85–93% reduction vs. ~16,900 for LoCoMo/MemGPT), under $0.0003 per operation, 5.4 s average with GPT-4o-mini and 1.1 s with local Llama 3.2 1B; retrieval time grows only from 0.31 µs to 3.70 µs from 1,000 to 1,000,000 memories.
Methods
- Design: each memory note is `{content, timestamp, keywords, tags, context, embedding, links}`; an LLM generates keywords/tags/context, a text encoder (all-minilm-l6-v2) produces embeddings, top-k nearest neighbors are retrieved by cosine similarity, and an LLM decides link generation and memory evolution (updating neighbors' context, keywords, tags).
- Data: LoCoMo (7,512 QA pairs, dialogues averaging 9K tokens over up to 35 sessions; single-hop, multi-hop, temporal, open-domain, adversarial) and DialSim (multi-party dialogues from Friends, The Big Bang Theory, The Office; 1,300 sessions, ~350,000 tokens, >1,000 questions per session).
- Evaluation: F1 and BLEU-1 as primary metrics, plus ROUGE-L, ROUGE-2, METEOR, and SBERT similarity; baselines LoCoMo, ReadAgent, MemoryBank, MemGPT; six foundation models (GPT-4o-mini, GPT-4o, Qwen2.5-1.5B/3B, Llama 3.2 1B/3B) plus DeepSeek-R1-32B, Claude 3.0 Haiku, and Claude 3.5 Haiku in the appendix; top-k retrieval with k=10 default and category-specific tuning.
Limitations
- Stated by the paper: memory organization quality may still depend on the underlying LLM's capabilities, and different LLMs may generate different contextual descriptions or connections; the implementation is text-only, with multimodal (image/audio) extension left to future work.
- Evident from the text: no error bars or significance tests are reported (the checklist answers "No" to statistical significance, citing API cost); the main-text results are reported for a single dataset family of long-term conversation QA, and the paper's own checklist notes broader societal impacts are not discussed.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
5 Valid · 0 Not valid · 3 Not checked · 55 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“These systems require agent developers to predefine memory storage structures, specify storage points within the workflow, and establish retrieval timing.”
What the sources say
- Partly supportedcites [25] [25] Charles Packer, Sarah Wooders, Kevin Lin, Vivian Fang, Shishir G Patil, Ion Stoica, and Joseph E Gonzalez. Memgpt: Towards llms as operating systems. arXiv preprint arXiv:2310.08560, 2023. sourceExperimental — accuracy not yet measured
Section 3.2.1 (p. 6) says the fixed-context baselines fetch documents independently of LLM inference, supporting the idea that retrieval is set by the workflow rather than chosen by the agent. Section 2.3 (p. 3) contrasts this with MemGPT’s self-directed memory edits and retrieval. The paper does not establish that existing memory systems generally require developers to predefine storage structures or specify storage points. (LLM, conf=0.78)
- Could not verifycites [39] [39] Wanjun Zhong, Lianghong Guo, Qiqi Gao, He Ye, and Yanlin Wang. Memorybank: Enhancing large language models with long-term memory. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 19724–19731, 2024. 12 Contents sourceExperimental — accuracy not yet measured
Sections 2.1 and 2.2 describe MemoryBank’s storage structure and retrieval mechanism, but do not say that existing systems require agent developers to predefine structures, choose workflow storage points, or set retrieval timing. Searches for those developer requirements found no passage addressing them. (LLM, conf=0.73)
- Not matchedcites [28] [28] Aymeric Roucher, Albert Villanova del Moral, Thomas Wolf, Leandro von Werra, and Erik Kaunismäki. 'smolagents': a smol library to build great agentic systems. https://github. com/huggingface/smolagents, 2025.
Its entry has no title and year that we can search by.
- Could not verifycites [21] [21] Zhiwei Liu, Weiran Yao, Jianguo Zhang, Liangwei Yang, Zuxin Liu, Juntao Tan, Prafulla K Choubey, Tian Lan, Jason Wu, Huan Wang, et al. Agentlite: A lightweight library for building and advancing task-oriented llm agent system. arXiv preprint arXiv:2402.15538, 2024. sourceExperimental — accuracy not yet measured
Section 2.1 describes when AgentLite retrieves an action-observation history and stores an observation; the Figure 1 description in Section 6 also describes its memory workflow. Neither passage says that existing memory systems require developers to predefine storage structures, storage points, or retrieval timing. Searches for those requirements found no passage addressing them. (LLM, conf=0.72)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Existing memory systems [25, 39, 28, 21] for LLM agents provide basic memory storage functionality.”
What the sources say
- Supportedcites [25] [25] Charles Packer, Sarah Wooders, Kevin Lin, Vivian Fang, Shishir G Patil, Ion Stoica, and Joseph E Gonzalez. Memgpt: Towards llms as operating systems. arXiv preprint arXiv:2310.08560, 2023. source
MemGPT is an LLM-agent memory system that stores information outside the context window. Section 2 (p. 2) describes its external context and memory-management functions, and the Figure 3 description (p. 3) identifies archival and recall storage databases. The claim does not say these systems provide only basic storage. (LLM, conf=0.88)
- Supportedcites [39] [39] Wanjun Zhong, Lianghong Guo, Qiqi Gao, He Ye, and Yanlin Wang. Memorybank: Enhancing large language models with long-term memory. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 19724–19731, 2024. 12 Contents source
MemoryBank is an LLM memory system with a memory-storage component. Section 2 describes that component as its primary data repository, and §2.1 says it stores conversation records, event summaries, and user information. This supports citing MemoryBank as an example of a system that provides memory storage functionality. (LLM, conf=0.91)
- Not matchedcites [28] [28] Aymeric Roucher, Albert Villanova del Moral, Thomas Wolf, Leandro von Werra, and Erik Kaunismäki. 'smolagents': a smol library to build great agentic systems. https://github. com/huggingface/smolagents, 2025.
Its entry has no title and year that we can search by.
- Supportedcites [21] [21] Zhiwei Liu, Weiran Yao, Jianguo Zhang, Liangwei Yang, Zuxin Liu, Juntao Tan, Prafulla K Choubey, Tian Lan, Jason Wu, Huan Wang, et al. Agentlite: A lightweight library for building and advancing task-oriented llm agent system. arXiv preprint arXiv:2402.15538, 2024. source
Section 2.1, “Individual Agent” (page not available), describes AgentLite’s Memory component as storing an agent’s action-observation chains and saving new actions and observations during execution. This supports citing AgentLite as an example of an LLM-agent system with basic memory storage functionality. (LLM, conf=0.90)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“While previous datasets contain dialogues with around 1K tokens over 4-5 sessions, LoCoMo features much longer conversations averaging 9K tokens spanning up to 35 sessions”
§ 4 Experiment > 4.1 Dataset and Evaluation
What the sources say
- Supportedcites [22] [22] Adyasha Maharana, Dong-Ho Lee, Sergey Tulyakov, Mohit Bansal, Francesco Barbieri, and Yuwei Fang. Evaluating very long-term conversational memory of llm agents. arXiv preprint arXiv:2402.17753, 2024. source
The Introduction (p. 2) reports that LoCoMo conversations average 9K tokens and span up to 35 sessions. Its comparison table (p. 2) lists earlier multi-session datasets at about 1,055–1,226 tokens across four or five sessions, supporting the stated contrast. (LLM, conf=0.94)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“T-SNE Visualization of Memory Embeddings Showing More Organized Distribution with A-MEM (blue) Compared to Base Memory (red) Across Different Dialogues.”
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“with each memory operation costing less than $0.0003 when using commercial API services”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
25 more claims
Create a free account to see all 63 claims and every source verdict.
- Probably not a substantial claim. 16
- It may state two separate claims as one. 20
- The quote only partly states this claim. 2
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The cited source doesn’t support the claim as written.
In the citing paper
“Relative to the O (1) dictionary lookup used by all prior systems [4], [5], [6], [7], [8], [9], lineage-gated retrieval trades at most 13 . 82 µ s in absolute terms for the safety guarantee of Theorem 1-a tradeoff Experiments 1-2 show buys elimination of an 18 . 8 -25 . 5% structural leak rate at a 13 -14 percentage-point reuse cost.”
From Lineage-Aware Memory Governance: A Derivation-Gated Framework for Privacy-Preserving Column-Level Access Control in Enterprise AI Agents (2026) · cites this paper as [6]
Checked by a reasoning model — may be wrong.
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“This emerging class of systems extends the capabilities of traditional AI Agents by enabling multiple intelligent entities to collaboratively pursue goals through structured communication [144-146], shared memory [147,148], and dynamic role assignment [22].”
From AI Agents vs. Agentic AI: A Conceptual taxonomy, applications and challenges (2026) · cites this paper as [147]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“from Wang et al. [55], Xu et al. [45] build memory to improve long context performance for specific workflows which suggest that diverse evaluation workflows are needed for better testing of generalizable memory components.”
From MEMTRACK: Evaluating Long-Term Memory and State Tracking in Multi-Platform Dynamic Agent Environments (2025) · cites this paper as [45]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Some other approaches introduce external memory modules ( e.g., summarizers or retrievers) [63, 29, 13, 57], but these are typically trained separately and cannot be optimized end-to-end with the agent's policy. This also introduces additional engineering overhead, as engineers must manage and integrate two separate models.”
From MEM1: Learning to Synergize Memory and Reasoning for Efficient Long-Horizon Agents (2025) · cites this paper as [57]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Some others, such as HiAgent[5], A-Mem[23], MemAgent [28] use hierarchical designs to separate working buffers from episodic and long-term stores to relieve context pressure and improve persistence.”
From Learning Hierarchical Procedural Memory for LLM Agents through Bayesian Selection and Contrastive Refinement (2025) · cites this paper as [23]
Experimental — accuracy not yet measured
9 more claims about this paper
Create a free account to see every claim other papers make about this one.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
§ APPENDIX
“Using GPT-4o-mini, A-MEM achieves a ROUGE-L score of 44.27 in Multi-Hop tasks, more than doubling LoComo's 18.09.”
§ APPENDIX
“LOCOMO 9.64 23.92 2.01 18.09 3.40 11.58 26.48 40.20 60.46 69.59 READAGENT 2.47 9.45 0.95 13.12 0.55 5.76 2.99 9.92 6.66 9.79 MEMORYBANK 1.18 5.43 0.52 9.64 0.97 5.77 1.64 6.63 4.55 7.35 MEMGPT 10.58 25.60 4.76 25.22 0.76 9.14 28.44 42.24 36.62 43.75 A-MEM 10.61 25.86 21.39 44.27 3.42 12.09 29.50 45.18 42.62 50.04”
Rated major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
§ APPENDIX
“The Multi-Hop category showcases particularly striking results, where Qwen2.5-15b with A-MEM achieves a ROUGE-L score of 27.23, dramatically surpassing LoComo's 4.68 and ReadAgent's 2.81 - representing a nearly six-fold improvement.”
§ APPENDIX
“Multi Hop Temporal Open Domain Single Hop Adversial RGE-2 RGE-L RGE-2 RGE-L RGE-2 RGE-L RGE-2 RGE-L RGE-2 RGE-L”
Rated major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
§ APPENDIX
“This significant advantage maintains consistency across other metrics, with METEOR scores of 23.43 versus 7.61 and SBERT scores of 70.49 versus 52.30.”
§ APPENDIX
“Multi Hop Temporal Open Domain Single Hop Adversial ME SBERT ME SBERT ME SBERT ME SBERT ME SBERT”
Rated major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
§ 4 Experiment > 4.3 Empricial Results
“our A-MEM demonstrates superior performance in Multi-Hop tasks achieves at least two times better performance”
§ 4 Experiment > 4.1 Dataset and Evaluation
“MEMGPT 26.65 17.72 25.52 19.44 9.15 7.44 41.04 34.34 43.29 42.73 2.4 2.4 16,977 A-MEM 27.02 20.09 45.85 36.67 12.14 12.00 44.65 37.06 50.03 49.47 1.2 1.2 2,520”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 10 20 30 40 50
“44.55 25.43 28.31 32.12 34.32 37.02”
§ 4 Experiment > 4.1 Dataset and Evaluation
“A-MEM 27.02 20.09 45.85 36.67 12.14 12.00 44.65 37.06 50.03 49.47 1.2 1.2 2,520”
Citations
Cites
The publisher hasn’t deposited this paper’s reference list.
Research works printed but not linked: 4
- [1][1] Sönke Ahrens. How to Take Smart Notes: One Simple Technique to Boost Writing, Learning and Thinking. Amazon, 2017. Second Edition.
- [5][5] Satanjeev Banerjee and Alon Lavie. Meteor: An automatic metric for mt evaluation with improved correlation with human judgments. In Proceedings of the acl workshop on intrinsic and extrinsic evaluation measures for machine translation and/or summarization, pages 65–72, 2005.
- [18][18] Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, et al. Retrieval-augmented generation for knowledge-intensive nlp tasks. Advances in Neural Information Processing Systems, 33:9459–9474, 2020.
- [19][19] Chin-Yew Lin. Rouge: A package for automatic evaluation of summaries. In Text summarization branches out, pages 74–81, 2004.
Cited by
Items 201–300 of 1,030
Tree Training: Accelerating Agentic LLMs Training via Shared Prefix ReusePreprint
Wang, J., Wang, S., Cui, Y., Chen, X., Wang, C., Huang, L., Tang, C., Zhang, X., Peng, J., Wan, L., et al. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2511.00413 · Cited by 10
“…Zhang et al., 2024) and sub-agents(Team et al., 2026)), concurrent tool invocations (Kim et al., 2024; Abdelaziz et al., 2024), context management (Xu et al., 2025; Packer et al., 2024; Zhong et al., 2023), as well as runtime behaviors such as think-mode (discarding/replacing intermediate…”
DeepSieve: Information Sieving via LLM-as-a-Knowledge-RouterPreprint
Guo, M., Zeng, Q., Zhao, X., Liu, Y., Yu, W., Du, M., Chen, H., & Cheng, W. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2507.22050 · Cited by 9
“This source-aware routing builds on the idea of using LLMs as controllers for modular tool access (Schick et al., 2023; Xu et al., 2025).”
Memory-T1: Reinforcement Learning for Temporal Reasoning in Multi-session AgentsPreprint
Du, Y., Wang, B., Xiang, Y., Wang, Z., Huang, W., Xue, B., Liang, B., Zeng, X., Mi, F., Bai, H., et al. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2512.20092 · Cited by 14
Live-Evo: Online Evolution of Agentic Memory from Continuous FeedbackPreprint
Zhang, Y., Wu, Y., Yu, Y., Wu, Q., & Wang, H. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.02369 · Cited by 9
“While existing systems retrieve through similarity matching (Xu et al., 2025; Park et al., 2023)) or active exploration strategies (e.g., in which the agent probes the memory bank iteratively) (Chhikara et al., 2025a; Long et al., 2025), our active retrieval design enables the agent to retrieve…”
InfluentialSelf-citationCosta, I. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.07072 · Cited by 8
“A-MEM [25] implements Zettelkasten-style interconnected knowledge.”
PreFlect: From Retrospective to Prospective Reflection in Large Language Model AgentsPreprint
Wang, H., Cao, Y., Lin, L., & Chen, J. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.07187 · Cited by 8
“…established the iterative “reason-act” paradigm, which has since evolved into specialized research on test-time searching (Yao et al., 2023; Besta et al., 2024), tool tuning (Zeng et al., 2025; Qian et al., 2025; Jiang et al., 2025), and memory structuring (Xu et al., 2025; Chhikara et al., 2025).”
UMEM: Unified Memory Extraction and Management Framework for Generalizable MemoryPreprint
Ye, Y., Jiang, H., Jiang, F., Lan, T., Du, Y., Fu, B., Shi, X., Jia, Q., Wang, L., & Luo, W. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.10652 · Cited by 10
Recursive Models for Long-Horizon ReasoningPreprint
Yang, C., Srebro, N., & Li, Z. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2603.02112 · Cited by 9
“…approaches write and retrieve relevant information in external storage (Packer et al., 2024; Chhikara et al., 2025; Suzgun et al., 2025; Xu et al., 2025); and in agentic systems, subtasks are distributed across agents, each operating in its own context while collaborating toward a…”
Scaling Teams or Scaling Time? Memory Enabled Lifelong Learning in LLM Multi-Agent SystemsPreprint
Wu, S., Luo, Y., Liang, Y., Shi, K., Ye, Y., Payani, A., & Shu, K. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.03295 · Cited by 8
State Contamination in Memory-Augmented LLM AgentsPreprint
Wang, Y., Goyal, A., Chen, Y., & Sundaram, H. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.16746 · Cited by 11
AutoSci: A Memory-Centric Agentic System for the Full Scientific Research LifecyclePreprint
Qian, W., Xu, B., Xie, Z., Fan, B., Tang, G., Chen, J., Wu, X., Yang, M., Di, C., Li, J., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.31468 · Cited by 8
“A-MEM (Xu et al., 2025) links memory notes in an agentic network, AriGraph (Anokhin et al., 2024) and Zep (Rasmussen et al., 2025) build graph-based world or temporal memories, Hip-poRAG (Jimenez Gutierrez et al., 2024)”
TellOnce: Correcting Coding Agents Once, Improving Their Future ActionsPreprint
Zhou, Y., Guo, K., Zhuang, H., Wang, X., Huang, Y., Liang, Z., Chen, P., Gao, T., Moniz, N., Chawla, N. V., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.13174 · Cited by 9
Self-citationStructAgent: Harness Long-horizon Digital Agents with Unified Causal StructurePreprint
Wu, W., Zhu, S., Zhou, K., Salvi, A., Song, Z., & Huang, B. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.11388 · Cited by 10
“Prior work stores reusable experience, induces workflows or skills, and builds tool surfaces that make execution more reliable across repeated tasks [6, 21–23 , 37– 45”
From AI for Science to Agentic Science: A Survey on Autonomous Scientific Discovery
Wei, J., Yang, Y., Zhang, X., Chen, Y., Zhuang, X., Gao, Z., Zhou, D., Wang, G., Gao, Z., Cao, J., et al. (2025) · arXiv · doi:10.48550/arxiv.2508.14111 · Cited by 7
G-Memory: Tracing Hierarchical Memory for Multi-Agent Systems
Zhang, G., Fu, M., Wang, K., Wan, F., Yu, M., & Yan, S. (2025) · Advances in Neural Information Processing Systems 38 · doi:10.52202/085713-0439 · Cited by 7
Towards Scientific Intelligence: A Survey of LLM-based Scientific Agents
Ren, S., Xie, C., Jian, P., Ren, Z., Leng, C., & Zhang, J. (2025) · arXiv · doi:10.48550/arxiv.2503.24047 · Cited by 7
“Similarly, ClimateGPT (Thulke et al., 2024) integrates a retrieval mechanism to access a curated collection of climatological research reports and peer-reviewed papers, thereby enhancing response accuracy in climate science applications.”
CogMem: A Cognitive Memory Architecture for Sustained Multi-Turn Reasoning in Large Language Models
Zhang, Y., Hu, J., Dras, M., & Naseem, U. (2026) · International Joint Conference on Natural Language Processing and Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (IJCNLP-AACL) · doi:10.48550/arxiv.2512.14118 · Cited by 8
“Later systems, including MemBank (Zhong et al., 2024), Mem0 (Chhikara et al., 2025), Mem1 (Zhou et al., 2025), A-Mem (Xu et al., 2025), and MemOS (Li et al., 2025), introduced structured, evolving memory architectures that enable dynamic extraction, updating, and long-term storage of relevant knowledge.”
Self-citationMAS-on-the-Fly: In-Context Structural Adaptation of LLM-Based Multi-Agent Systems
Liu, G., Lin, H., Zeng, H., Wang, H., & Yao, Q. (2026) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/2026.emnlp-main.187 · Cited by 9
Li, C., & Lu, W. (2026) · Conference of the European Chapter of the Association for Computational Linguistics · doi:10.18653/v1/2026.eacl-long.141 · Cited by 16
GSEM: Graph-based Self-Evolving Memory for Experience Augmented Clinical Reasoning
Han, X., Fan, Y., Zhao, S., Wang, H., & Qin, B. (2026) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/2026.emnlp-main.507 · Cited by 8
DeltaMem: Towards Agentic Memory Management via Reinforcement Learning
Zhang, Q., Huang, S., Liu, C., Yang, S., Hu, X., Xie, P., Wang, H., & Zhao, J. (2026) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/2026.emnlp-main.1018 · Cited by 8
RECALLbot: Designing Agentic Memory and Reciprocal Disclosure for Human–Chatbot Relationships
Jiang, Z., Zheng, C., Chen, H., & Chen, L. (2026) · CHI Conference on Human Factors in Computing Systems · doi:10.1145/3772318.3790714 · Cited by 9
“Furthermore, recent machine learning research has explored broader memory architectures for LLM-based agents, including long-term structured memory for autonomous agents [104], modular and evaluative memory frameworks [99], and multi-component memory systems [94].”
MemGuard: Preventing Memory Contamination in Long-Term Memory-Augmented Large Language Models
Ha, H., Kim, J., Qian, C., Liu, J., Campbell, W. M., Wu, Y., Zhang, Y., McKeown, K. R., Hakkani-Tur, D. Z., & Ji, H. (2026) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/2026.findings-emnlp.571 · Cited by 8
Self-citationInside Out: Evolving User-Centric Core Memory Trees for Long-Term Personalized Dialogue Systems
Zhao, J., Chen, D., Fan, Z., Xu, K., Hu, M., Tang, B., Xiong, F., & li, Z. (2026) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2026.acl-long.614 · Cited by 8
Cache Mechanism for Agent RAG SystemsPreprint
Lin, S.-H., Peng, Z., Li, L., Lin, X., Zhu, X., & Zhang, Y.-F. (2025) · arXiv.org · doi:10.48550/arxiv.2511.02919 · Cited by 8
“…which exhibit remarkable capabilities in performing complex reasoning tasks (Wang et al., 2024b; Jin et al., 2024; Hua et al., 2024), executing knowledge-intensive operations (Jiang et al., 2024), and implementing customized automated workflows (Li et al., 2024; Xu et al., 2025; Mei et al., 2025b).”
Self-citationBeyond Heuristics: A Decision-Theoretic Framework for Agent Memory ManagementPreprint
Sun, C., Chen, X., Luo, J., Zhang, D., & Li, X. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2512.21567 · Cited by 7
“Extending this to settings involving multiple users or shared memory pools requires augmenting S t to capture distribution shifts and interference effects (Packer et al., 2023; Xu et al., 2025).”
ES-Mem: Event Segmentation-Based Memory for Long-Term Dialogue AgentsPreprint
Zou, H., Sun, T., He, C., Tian, Y., Li, Z., Jin, L., Liu, N., Zhong, J., & Wei, K. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2601.07582 · Cited by 7
“We follow A-Mem (Xu et al., 2025) to use F1 and BLEU-1 (B1) as the evaluation metrics of LoCoMo benchmark.”
InfluentialFrom Atom to Community: Structured and Evolving Agent Memory for User Behavior ModelingPreprint
Liao, Y., Wu, L., Hou, M., Wang, Y., Wu, H., & Wang, M. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2601.16872 · Cited by 9
“(ii) Planar Memory [23, 27, 31, 34, 46] extends the flat organization by explicitly capturing relations between memory units, typically using graphs [23, 46], trees [31], or hybrid architectures [19] to represent their connections, thereby enabling more structured retrieval and reasoning.”
AMA: Adaptive Memory via Multi-Agent CollaborationPreprint
Huang, W., Wang, Z., Lin, H., Wang, S., Xu, B., Li, Q., Zhu, B., Yang, L., & Qin, C. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2601.20352 · Cited by 9
“A-Mem (Xu et al., 2025) A-Mem is an agentic memory system designed for LLM-based agents that enables dynamic organization and evolution of long-term memory.”
Self-Consolidation for Self-Evolving AgentsPreprint
Yu, H., Zhu, F., Xie, G.-S., & Shao, L. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.01966 · Cited by 7
“Self-Controlled Memory (Wang et al., 2025) devises a dedicated memory framework consisting of the agent, stream and controller modules, which is specialized for processing ultra-long input sequences.”
InfluentialWang, Q., Wang, S., Zeng, Y., Zhang, Q., Zhang, F., Guo, Z., Zhang, B., Huang, W., Lin, C., Chen, Z., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.12735 · Cited by 7
“…strategies have been recognized as effective approaches to optimize long-context tasks Zhou et al. (2025); Yu et al. (2025a); Wu et al. (2025); Xu et al. (2025); Chhikara et al. (2025); Chen et al. (2024a); Ye et al. (2025), transferring this paradigm to efficiently manage massive visual…”
Xu, Y., Li, L., Sleem, L., Gentile, N., Song, Y., Wang, Y., Ji, S., Wu, W., Huang, H., & State, R. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.16653 · Cited by 8
“…(Zhang et al., 2024b; Salama et al., 2025; Hu et al., 2025) leveraging file systems, vector databases, knowledge graphs (Edge et al., 2025), and memory buffers (Xu et al., 2025) to achieve comprehensive enhancements in contextual coherence, personalized learning, and complex task decision-making.”
A Scalable Benchmark for Repository-Oriented Long-Horizon Conversational Context ManagementPreprint
Liu, Y., Zhang, L., Liu, F., Lin, P., & Li, X. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2603.06358 · Cited by 7
“For example, Mem0 [8] maintains memory through a database-like manner, whereas Agentic Memory [53] organizes memories in a graph structure to establish semantic links among them.”
HAGE: Harnessing Agentic Memory via RL-Driven Weighted Graph EvolutionPreprint
Jiang, D., Li, Y., Li, G., Li, Q., & Li, B. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.09942 · Cited by 7
“Feeds the entire conversation history into the LLM. A-MEM (Xu et al., 2025).”
InfluentialMemory Retrieval for Changing PreferencesPreprint
Qin, Y., Li, L., Song, L., (109917), W. Y., Li, J., Yang, Y., & (186573), Y. Z. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.02976 · Cited by 7
End-to-End Context Compression at ScalePreprint
Li, A., McLeish, S., Chen, H., Kalra, N., Chen, Z., Gazizov, A., Morisetty, V. A. S. K., Kailkhura, B., Menon, H., Liu, Z., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.09659 · Cited by 8
User as Engram: Internalizing Per-User Memory as Local Parametric EditsPreprint
Li, B. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.19172 · Cited by 7
Singh, P. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.22030 · Cited by 7
“More recent systems including Mem0 [2] and A-MEM [17] combine graph and vector representations to improve retrieval precision.”
InfluentialWhat to Keep, What to Forget: A Rate--Distortion View of Memory Compaction in LLMs and AgentsPreprint
Colaco, A. G., & Lahjouji, N. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.08032 · Cited by 7
“…consistent store [21]; A-MEM builds Zettelkasten-style atomic notes that are dynamically linked to and evolve related notes into a self-organizing network [125]; Memory-Bank adds explicit forgetting, decaying memories by an Ebbinghaus curve so that significance and recency, not capacity alone,…”
ABot-AgentOS: A General Robotic Agent OS with Lifelong Multi-modal MemoryPreprint
Tian, J., Liu, S., Xu, Y., Lu, J., Guan, Z., Han, H., Yang, D., Gu, M., Qian, Y., Zhang, T., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.10350 · Cited by 8
SafeCommit: Certifying When Memory-Grounded Agents May Safely ActPreprint
Akewar, M., & Ranjan, R. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.04289 · Cited by 7
“ReAct established the value of interleaving reasoning and tool use [21], while Reflexion and newer memory systems use persistent feedback and organized memory to improve future behavior [17, 20].”
InjecMEM: Memory Injection Attack on LLM Agent Memory SystemsPreprint
Tian, H., Zhang, G., Sha, Z., Wang, J., Liu, Y., Huang, Z., Yang, K., & Huang, X. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.23471 · Cited by 13
Lifelong Learning of Large Language Model Based Agents: A Roadmap
Zheng, J., Shi, C., Cai, X., Li, Q., Zhang, D., Li, C., Yu, D., & Ma, Q. (2026) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2025.3650546 · Cited by 7
Ying, T., Wang, L., Hu, P., & Ding, T. (2026) · International Conference on Electronic Technology and Information Science (ICETIS) · doi:10.1109/icetis70504.2026.11633437 · Cited by 7
“…understanding, complex reasoning, task planning, and tool invocation, building intelligent agents with autonomous perception, decision-making, and execution capabilities based on large language models is becoming an important development direction in the field of artificial intelligence [1,2].”
Fang, J., Peng, Y., Zhang, X., Wang, Y., Yi, X., Zhang, G., Xu, Y., Wu, B., Liu, S., Li, Z., et al. (2025) · arXiv · doi:10.48550/arxiv.2508.07407 · Cited by 6
Xu, B., Zheng, T., Wang, Z., Tsang, H. T., Wang, W., Fang, T., & Song, Y. (2026) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/2026.emnlp-main.755 · Cited by 6
Yu, Z., Xiao, K., Zhao, H., Luo, T., & Zeng, X. (2026) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/2026.findings-emnlp.1669 · Cited by 8
Toward Sustainable Agentic AI Systems: A Survey of Architectures and Methodologies
Lee, Y., & Park, E. (2026) · Sustainable Development · doi:10.1002/sd.70942 · Cited by 13
Wang, F., Yang, J., Chen, J., Liu, Y., Li, K., Wei, Y., Guo, D., & Wang, M. (2026) · International World Wide Web Conference (WWW) · doi:10.1145/3774904.3793062 · Cited by 13
“Following the Atomic Structured Tool-Taking scheme illustrated in Figure 3, inspired by the Zettelkasten method [1, 26, 48], traditional memory [30, 47] relies on predefined storage frameworks and is unsuitable for the multi-stage dynamic memory of counseling.”
Benchmarking Knowledge-Extraction Attack and Defense on Retrieval-Augmented Generation (RAG)
Qi, Z., Sahu, U., Ma, L., Han, H., Rossi, R. A., Dernoncourt, F., Halappanavar, M. M., Ahmed, N., Dong, Y., Zhao, Y., et al. (2026) · ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) · doi:10.1145/3770855.3817524 · Cited by 8
“This threat is further amplified by the growing adoption of RAG as memory management in Agentic systems [19, 20] in high-stakes domains such as personal healthcare [21, 22] and proprietary financial transactions [23] Therefore, successful knowledge extraction attacks can lead to severe privacy…”
Self-citationStructured Episodic Event Memory
Lu, Z., Li, D., Shi, Y., Wang, B., Wang, L., & Hu, B. (2026) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2026.acl-long.277 · Cited by 6
Fine-Mem: Fine-Grained Feedback Alignment for Long-Horizon Memory Management
Ma, W., Feng, X., Huang, L., Feng, X., Ma, Z., Xu, J., Gao, J., Hao, J., He, R., & Qin, B. (2026) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2026.acl-long.900 · Cited by 12
Zhang, S., Li, J. J., Zhang, Y., Jia, P., Wang, Y., Qian, X., Zhang, W., Wang, M., Liu, Y., & Zhao, X. (2026) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2026.acl-long.41 · Cited by 12
Self-citationGated Differentiable Working Memory for Long-Context Language Modeling
Mei, L., Liu, S., Wang, Y., Ge, Y., Bi, B., Yao, J., Wan, J., Yin, Z., Guo, J., & Cheng, X. (2026) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2026.acl-long.1471 · Cited by 9
Individual Turing Test: A Case Study of LLM-based Simulation Using Longitudinal Personal Data
Guo, M., Ye, Z., Xu, W., Zhu, X., Hua, W., & Metaxas, D. (2026) · International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR) · doi:10.1145/3805712.3809868 · Cited by 7
“For non-parametric modeling, we consider both retrieval-based RAG [17] and memory-based A-Mem [30].”
Self-citationApplying Cognitive Design Patterns to General LLM Agents
Wray, R. E., Kirk, J. R., & Laird, J. E. (2025) · Lecture notes in computer science · doi:10.1007/978-3-032-00800-8_28 · Cited by 9
“MemGPT [32], LONGMEM [52] Ahistorical KR/memory A-MEM [63], WISE [50], MemoryBank [69] Historical KR/memory Generative Agents [33], MemoryBank [69] Procedural KR/memory ProgPROMPT [41], Voyager [49], Self-evolving GPT [12] Observe-Decide-Act in ReAct Generally, LLM performance improves when the…”
SignalLLM: A General-Purpose LLM Agent Framework for Automated Signal ProcessingPreprint
Ke, J., Hu, Q., Yuan, S., Xu, Y., & Yang, J. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2509.17197 · Cited by 8
“To further improve its effectiveness, we suggest incorporating more advanced RAG techniques and agent memory mechanisms [37], or fine-tuning the model with reinforcement learning methods [38], thereby enabling more informed and adaptive decision-making.”
Zeng, Q., Fan, B., Chen, Z., Ren, S., Zhou, Z., Zhang, X., Zhen, Y., Xu, F., Yong, L., & Liu, T. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2511.16997 · Cited by 6
“8 Related Work [25] and Mem0 [3] capture and structure knowledge to enable session-spanning reasoning, while Retrieval-Augmented Generation (RAG) [10], MemoryBank [28], and the HippoARG series [8, 6] enrich LLMs with external memory libraries for context-aware generation.”
PRELUDE: A Benchmark Designed to Require Global Comprehension and Reasoning over Long ContextsPreprint
Yu, M., Chung, T. T., Zhou, C., Li, T., Lu, R., Li, J., Xu, L., Lu, H., Zhang, N., Li, J., et al. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2508.09848 · Cited by 7
“…analysis (Google DeepMind, 2024; Wang et al., 2024d; Gutierrez et al., 2024), personal assistants with chat histories (Wu et al., 2024; Xu et al., 2025a), autonomous agents (Hu et al., 2025; OpenAI, 2025; Wang et al., 2025), and repository-level coding tools (Jimenez et al., 2024),…”
Evaluating Long-Term Memory for Long-Context Question Answering
Terranova, A., Roß, B., & Birch, A. (2025) · Edinburgh Research Explorer (University of Edinburgh) · doi:10.48550/arxiv.2510.23730 · Cited by 8
“Unlike previous research that used different prompts depending on the type of question [11, 18], we prompt the models in the same way regardless of the question type, making the task more realistic and allowing better isolation of memory mechanism effects.”
Experience-Evolving Multi-Turn Tool-Use Agent with Hybrid Episodic-Procedural MemoryPreprint
Li, S., Huang, Y., Liu, Z., Li, Z., Fu, J., Song, L., Bian, J., Zhang, J., & Wang, R. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2512.07287 · Cited by 7
“Graph-based methods continuously update the memory graph by adding new episodic events or semantic knowledge, dynamically creating nodes and relational links that refine the agent’s understanding as it interacts with the environment (Xu et al., 2025; Anokhin et al., 2024; Jiang et al., 2025a).”
ARC: Active and Reflection-driven Context Management for Long-Horizon Information Seeking AgentsPreprint
Yao, Y., Huang, S., Dai, E., Tan, Z., Duan, Z., Jia, S., Jiang, Y., & Yang, T. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2601.12030 · Cited by 6
“MemoryBank (Zhong et al., 2023) models memory decay via the Ebbinghaus curve, while Mem0 (Chhikara et al., 2025) and A-MEM (Xu et al., 2025) utilize dynamic graphs or triplets.”
U-Fold: Dynamic Intent-Aware Context Folding for User-Centric AgentsPreprint
Su, J., Fang, R., Li, Y., Wang, X., Cai, S., Xie, P., Zhang, N., & Yuan, F. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2601.18285 · Cited by 6
“To mitigate this, one line of work adds external memory (e.g., A-Mem (Xu et al., 2025), Mem-OS (Li et al., 2025)), often at the cost of higher system complexity.”
$G^2$-Reader: Dual Evolving Graphs for Multimodal Document QAPreprint
Du, Y., Song, J., Zhou, Y., Wang, C., Gu, J., Chen, Z., Chen, M., Yao, W., Yang, Y., Wen, Y., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2601.22055 · Cited by 7
Pan, Z., Shen, Y., Hu, Z., Wang, Z., Manocha, A., Wang, Z., Yu, Z., Guan, Y., & Ding, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2601.21473 · Cited by 10
“A key bottleneck arises from the agent-specific memory , which includes private models (Park et al., 2025), LoRA adapters (Yu et al., 2024), prefix caches (Zheng et al., 2024a; Gim et al., 2024; Pan et al., 2025b), and action histories 1 (Packer et al., 2023; Chhikara et al., 2025; Xu et al., 2025).”
Arming Data Agents with Tribal KnowledgePreprint
Agarwal, S., Biswal, A., Zeighami, S., Cheung, A., Gonzalez, J., & Parameswaran, A. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.13521 · Cited by 10
“…to store the agent’s interactions with the database (i.e., queries and answers)—or their summaries—and retrieve relevant ones to include in the agent’s prompts for future queries via a form of RAG (retrieval augmented generation) using LLM memory stores [22, 23, 30, 38, 41, 45] such as Mem0 [30].”
Evolutionary System Prompt Learning for Reinforcement Learning in LLMsPreprint
Zhang, L., Chen, R., & Stadie, B. C. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.14697 · Cited by 6
See, Remember, Explore: A Benchmark and Baselines for Streaming Spatial ReasoningPreprint
Wei, Y., Huang, W. H., Chen, Q., Hou, L., & Qi, X. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2603.23864 · Cited by 8
“While memory mechanisms like [12, 22, 39, 42, 43, 55, 57,67] are well-established in LLMs for long sequences, existing VLMs lack dedicated architectures to prevent forgetting in continuous spatial environments.”
SkillLens: Adaptive Multi-Granularity Skill Reuse for Cost-Efficient LLM AgentsPreprint
Yu, Z., Miao, Y., Zhao, L., Zhu, B., & Haque, H. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.08386 · Cited by 9
“MemoryBank [36], A-MEM [25], Mem0 [4], and Memp [7] organize, retrieve, and revise long-term or procedural memories across interactions.”
EquiMem: Calibrating Shared Memory in Multi-Agent Debate via Game-Theoretic EquilibriumPreprint
孟宇桥, Tang, L., Narvekar, S. S., Vaje, R. R., Zhang, Y., Ye, M., & Xi, Z. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.09278 · Cited by 7
“The effort cost c i is estimated online as the empirical computational overhead of a i ’s scrutiny actions across rounds, requiring no pre-specification (see App.”
Self-citationRethinking How to Remember: Beyond Atomic Facts in Lifelong LLM Agent MemoryPreprint
Sun, J., Zhu, J., Yao, J., Liu, T., & Han, B. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.19952 · Cited by 6
“For memory construction, Mem0 [3], A-Mem [24], and MemoryOS [10] prompt LLMs to extract atomic facts from dialogues and consolidate them via dynamic updates or hierarchical stores.”
Worth Remembering: Surprise-Gated Robot Episodic MemoryPreprint
Gorlo, N., Wise, D. K., Speranzon, A., & Carlone, L. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.03787 · Cited by 6
“D-Mem [47] uses a reward prediction error signal to determine the utility of language inputs and whether they should be integrated into an actively managed agent knowledge graph [48].”
MemToolAgent: Leveraging Memory for Tool Using Agents Based on Environment and User FeedbackPreprint
Er, S. A., Ribeiro, D. N., Virkar, Y. S., Lakew, S. M., Kalyanpur, A., Gung, J., Delteil, T., & Gupta, A. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.07909 · Cited by 6
“Agent Workflow Memory (Wang et al., 2024b) stores workflows in the memory to guide the agent in web navigation tasks.”
InfluentialBeyond Self-Reflection: An Execute-Distill-Verify Framework for Agentic Experience LearningPreprint
Zhu, S. J., Qi, Y., Wang, Y., Li, J., Song, C., Shi, Y., Miao, Y., Gao, H., & Zhang, K. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.24428 · Cited by 6
“This motivates experience learning , a paradigm in which agents distill reusable lessons from historical executions, store them as external memory, and retrieve them to support future decisions(Xu et al., 2025b).”
Self-citationAutoMem: Automated Learning of Memory as a Cognitive SkillPreprint
Wu, S., Zhu, H., Zhang, Y., Wang, X., & Yeung, S. Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.01224 · Cited by 7
“…memory management; Generative Agents [Park et al., 2023] maintain a timestamped memory stream from which they retrieve, reflect, and plan; A-MEM [Xu et al., 2025] equips LLMs with active decisions on what to retain and forget; MemoryBank [Zhong et al., 2024] maintains a persistent long-term…”
Self-citationPM-Bench: Evaluating Prospective Memory in LLM AgentsPreprint
Liu, G., & Gabriel, S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.12385 · Cited by 9
Spencer, Q. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.21962 · Cited by 8
“…unbounded stores : vector retrieval with decay (Zhong et al., 2024), graph-structured memory with associative retrieval (Gutiérrez et al., 2024; Xu et al., 2025), and commercial systems — Mem0’s extraction pipeline (Chhikara et al., 2025) and Zep’s bi-temporal knowledge graph (Rasmussen et…”
Filesystem-Based Memory for LLM Agents: Organization, Evolution, and SustainabilityPreprint
Zhou, S., Yu, S., Wei, H., Wu, J., Ouyang, S., Jiao, Y., Pan, S., McAuley, J., Zhang, Y., Yu, T., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.26637 · Cited by 7
Self-citationTakahashi, K. (2026) · Zenodo (CERN European Organization for Nuclear Research) · doi:10.5281/zenodo.18905241 · Cited by 6
Hindsight: Structured Agent Memory that Retains, Recalls, and Reflects
Latimer, C., Boschi, N., Neeser, A., Bartholomew, C., Srivastava, G., Wang, X., & Ramakrishnan, N. (2026) · Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations) · doi:10.18653/v1/2026.acl-demo.27 · Cited by 5
“…INDSIGHT with an open-source 20B model achieves 83.6% and 83.2% overall accuracy, outperforming full-context GPT-4o and prior memory systems such as Zep (Rasmussen et al., 2025), Mem0 (Chhikara et al., 2025), and A-Mem (Xu et al., 2025); scaling to Gemini-3 Pro pushes LongMemEval accuracy to 91.4%.”
InfluentialLLM-based Agents Suffer from Hallucinations: A Survey of Taxonomy, Methods, and Directions
Lin, X., Ning, Y., Zhang, J., Dong, Y., Liu, Y., Wu, Y., Qi, X., Sun, N., Shang, Y., Wang, K., et al. (2025) · arXiv · doi:10.48550/arxiv.2509.18970 · Cited by 5
“However, this process is susceptible to Information Compression issues, where the generated summaries may be overly general, omit crucial details [198, 223, 284, 320], or introduce distortions due to imperfect abstraction [101, 241, 315, 336].”
L3M+P: Lifelong Planning with Large Language Models
Agarwal, K., Jiang, Y., Hu, J., Liu, B., & Stone, P. (2025) · IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) · doi:10.1109/iros60139.2025.11247529 · Cited by 7
“A-MEM [27] uses a flexible graph structure to organize information, populating nodes with atomic notes and assigned tags to dynamically insert inter-node links based on similarity.”
KEEP: A KV-Cache-Centric Memory Management System for Efficient Embodied Planning
Yang, Z., Xie, T., Lu, B., Liu, S., Yu, B., & Li, M. (2026) · ACM/IEEE Design Automation Conference (DAC) · doi:10.1145/3770743.3804093 · Cited by 7
“The design of memory is pivotal for enabling agents to learn from past experiences and improve planning performance [13, 30].”
LARA-HPC: Validation-driven agentic supercomputer workflows for atomistic modeling
Dawson, W., Beal, L., Curé, Y., Fisicaro, G., Rolland, D., & Genovese, L. (2026) · PRX Intelligence · doi:10.1103/cp6v-t93q · Cited by 5
AMARIS: A Memory-Augmented Rubric Improvement System for Rubric-Based Reinforcement Learning
Wu, P., Zhang, X., Wan, K., Zhao, W., Wu, G., Du, X., & Chen, Z. (2026) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/2026.findings-emnlp.2152 · Cited by 5
PyraVid: Hierarchical Multimodal Memory for Long-Horizon Video Reasoning
Yan, S., Dong, S., Wang, H., Nie, E., Liu, Y., Bi, J., Xu, Y., Schwarzmann, S., Trivisonno, R., Tresp, V., et al. (2026) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/2026.emnlp-main.777 · Cited by 5
MemPro: Agentic Memory Systems as Evolvable Programs
Liu, Q., Wang, G., Wu, W., Huang, J., Tao, X., Song, D., Zhou, J., & He, L. (2026) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/2026.emnlp-main.1364 · Cited by 6
TiMem: Temporal-Hierarchical Memory Consolidation for Long-Horizon Conversational Agents
Li, K., Yu, X., Ni, Z., Zeng, Y., Xu, Y., Zhang, Z., Li, X., Sang, J., Duan, X., Wang, X., et al. (2026) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2026.findings-acl.1091 · Cited by 18
Self-citationLi, X., Li, H., Zhou, Y., Liu, Y., & Ai, Q. (2026) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2026.acl-long.1831 · Cited by 6
Evaluating Memory Capability in Continuous Lifelog Scenario
Zheng, J., Liu, Z., Shen, Z., Qu, J., Chen, G., Wang, Y., Xu, Y., Liu, Y. J., & Cheng, S. (2026) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2026.findings-acl.351 · Cited by 7
Personalizing Agent Privacy Decisions via Logical Entailment
Flemings, J., Yi, R., Suciu, O., Fawaz, K., Annavaram, M., & Gruteser, M. (2026) · Proceedings on Privacy Enhancing Technologies · doi:10.56553/popets-2026-0087 · Cited by 6
ADO: Automatic Data Optimization for Inputs in LLM Prompts
Lin, S., Hua, W., Li, L., Wang, Z., & Zhang, Y. (2025) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2025.findings-acl.1340 · Cited by 6
Self-citationText2Mem: A Unified Memory Operation Language for Memory Operating SystemPreprint
Wang, Y., Yang, L., Chen, B., Zou, G., Xu, K., Tang, B., Xiong, F., Chen, S., & Li, Z. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2509.11145 · Cited by 5
“M EM GPT proposes modularizing context into dynamic pages [7], A-MEM introduces agentic memory abstractions for LLM agents [17], and M EM OS presents a more comprehensive operating system view with richer primitives and scheduling capabilities [6].”
Co-Investigator AI: The Rise of Agentic AI for Smarter, Trustworthy AML Compliance NarrativesPreprint
Naik, P. V., Dintakurthi, N. K., Hu, Z., Wang, Y., & Qiu, R. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2509.08380 · Cited by 7
Securing Generative AI Agentic Workflows: Risks, Mitigation, and a Proposed Firewall ArchitecturePreprint
Bahadur, S. K. J., & Dhar, G. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2506.17266 · Cited by 9
Training-Free Multimodal Large Language Model OrchestrationPreprint
Xie, T., Ma, Y., Wu, Y., Chen, W., Ji, J., Chua, T.-S., Zheng, X., & Ji, R. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2508.10016 · Cited by 6
EvoMem: Improving Multi-Agent Planning with Dual-Evolving MemoryPreprint
Fan, W., Yan, N., & Mortazavi, M. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2511.01912 · Cited by 6
“– Maximize the number of friends met during the day by considering various schedules.”
Experience-Driven Multi-Agent Systems Are Training-free Context-aware Earth ObserversPreprint
Dai, P., Xuan, W., Wang, J., Chen, H., Song, J., Ou, Y., & Yokoya, N. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.02559 · Cited by 6
“Recent advances in agent memory and retrieval-augmented reasoning indicate that storing distilled experience as retrievable in-context demonstrations can effectively shape future behavior (Zhang et al., 2025c; Ouyang et al., 2025; Xu et al., 2025; Zhao et al., 2024; Cai et al., 2025).”
Pancake: Hierarchical Memory System for Multi-Agent LLM ServingPreprint
Hu, Z., Pan, Z., Kaur, P., Murthy, V., Yu, Z., Guan, Y., Wang, Z., Swanson, S., & Ding, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.21477 · Cited by 8
“The baselines include: A-Mem [73]: backend for semantically evolving memory in long-term conversational retrieval.”
InfluentialEvoking User Memory: Personalizing LLM via Recollection-Familiarity Adaptive Retrieval
Zhang, Y., Li, J. J., Zhang, W., Jia, P., Li, X., Wang, Y., Xu, D., Wen, Y., Guo, H., Liu, Y. U., et al. (2026) · International Conference on Learning Representations · doi:10.48550/arxiv.2603.09250 · Cited by 18
“…(Li et al., 2025a; Zhao et al., 2025; Chen et al., 2025; Shen et al., 2024; Salama et al., 2025), index construction (Zhong et al., 2024; Pan et al., 2025; Xu et al., 2025c; Tan et al., 2025b; Xu et al., 2025d; Ong et al., 2024; Chhikara et al., 2025), and retrieval strategy (Xu et al., 2021).”
Self-citation
Metadata
- Authors
- Xu, W., Liang, Z., Mei, K., Gao, H., Tan, J., & Zhang, Y.
- Affiliations
- Rutgers, The State University of New Jersey (US)
- Ant Group (China) (CN)
- Salesforce (United States) (US)
- Venue
- Advances in Neural Information Processing Systems 38
- Published
- 2025
- Pages
- 20004-20031
- ISSN
- 1049-5258 (ISSN-L)
- OpenAlex
- W4407759090
- Cited by
- 51
- References
- 0
- Citations by year
- 2026: 48 · 2025: 3
- FWCI
- 48.02
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- neural information processing systems (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › AI-based Problem Solving and Planning
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Physical Sciences › Computer Science › Information Systems › Information Retrieval and Search Behavior
- Keywords
- agentic memory, LLM agents, knowledge networks, memory organization, dynamic indexing, evolution of memory, knowledge representation, memory management
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-09-28, under version 3 of the claim definition (prompt extract-v3).
A second model labelled each claim on 2026-09-28 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.