The DOI matches its Crossref record; 32 of the 32 references it cites were found in a registry or an index; 11 of the 17 sources its claims cite were checked for support.
A-Mem: Agentic Memory for LLM Agents
Open accessAbstract
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.
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- 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.
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- 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 301–400 of 1,030
HiGMem: A Hierarchical and LLM-Guided Memory System for Long-Term Conversational AgentsPreprint
Cao, S., He, J., & Tan, F. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.18349 · Cited by 5
“Memory-augmented LLM agents such as Mem-oryBank (Zhong et al., 2024), MemGPT (Packer et al., 2023), and A-Mem (Xu et al., 2025) extend long-horizon interaction by storing past experiences as external memories and retrieving them at inference time.”
Liu, Z., Ying, Z., Zhang, W., Zou, Q., Zhang, D., Yang, D., Zhang, X., & Peng, H. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.05704 · Cited by 7
“While our retrieval latency is marginally higher than flat RAG structures, it remains within the millisecond range and is comparable to graph-based memory systems like A-Mem.”
InfluentialGoal-Oriented Reasoning for RAG-based Memory in Conversational Agentic LLM SystemsPreprint
Liang, J., Toroghi, A., Liu, Y. S., Kalarde, F. M., Gallagher, L., & Sanner, S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.12213 · Cited by 5
H-Mem: A Novel Memory Mechanism for Evolving and Retrieving Agent Memory via a Hybrid StructurePreprint
Yu, J., Fang, Y., Liu, X., & Ma, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.15701 · Cited by 5
“A-Mem [19] studies agentic memory mechanisms, while multi-granularity memory methods [20] investigate memory association and selection across different abstraction levels.”
Mix-Quant: Quantized Prefilling, Precise Decoding for Agentic LLMsPreprint
Lu, H., Chen, Z., Fang, G., Ma, X., & Wang, X. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.20315 · Cited by 7
“Large language model (LLM) agents have emerged as a powerful paradigm for solving complex real-world tasks involving tool use, memory retrieval, code generation, and multi-step interaction [43, 33, 41, 38].”
Personalize-then-Store: Benchmarking and Learning Personalized Memory for Long-horizon AgentsPreprint
In, Y., Kim, W., Park, S., Yoon, K., & Park, C. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.25535 · Cited by 5
“The second direction focuses on structured memory representations, leveraging clustering, graph, and tree-based methods to model relationships among memory units and improve retrieval accuracy [4, 5, 8, 11, 1].”
InfluentialSimulating Human Memory with Language ModelsPreprint
Wang, Q., Tomlin, N., Hu, M., Dillon, B., & Linzen, T. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.25680 · Cited by 5
“Beyond extending the context window, recent systems introduce model-updatable parameters [34], sparse embedding lookups [35, 36], and structured memory stores for persistent experience reuse [37, 38, 39].”
ElasticMem: Latent Memory as a Learnable Resource for LLM AgentsPreprint
Feng, T., Ye, C., Yu, F., Luo, T., Xu, X., Xu, X., Zhang, H., Zhang, W., Lei, Z., & You, J. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.30690 · Cited by 6
“RepresentativeMethods MemorySpace TokenEfficient Reasoner-StateAdaptiveRetrieval ElasticCapacity Textuallong-termmemory MemoryBank[56]/Mem0[5]/MemoryOS[19] Text ✗ ✗ ✗ Textualagentmemory A-MEM[45]/LightMem[9]/MemP[10] Text ✗ ✗ ✗ Fixedlatentcompression AutoCompressor[4] Latent ✓ ✗ ✗ Latentmemoryretrieval MemGen[50]/M+[4…”
InfluentialTemporal Order Matters for Agentic Memory: Segment Trees for Long-Horizon AgentsPreprint
Liu, Y. S., Gallagher, L., Kalarde, F. M., Liang, J., Toroghi, A., & Sanner, S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.04555 · Cited by 5
“More recent agentic memory systems such as A-MEM and Mem0 emphasize explicit memory operations, including extraction, updating, linking, and consolidation [5, 41].”
InfluentialRecalling Too Well: Sycophancy Evaluation and Mitigation in Memory-Augmented ModelsPreprint
Bensal, S., Magnuson, A., Balagopalan, A., & Bikel, D. M. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.10949 · Cited by 6
“MIST draws its samples from the datasets listed in Table 1, selected on the basis of diversity and reputation.”
PROJECTMEM: A Local-First, Event-Sourced Memory and Judgment Layer for AI Coding AgentsPreprint
Malo, R. C., & Qiu, T. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.12329 · Cited by 6
“…Mem0g adding only marginal accuracy [5]; A-MEM organizes memory as a Zettelkasten knowledge network of LLM-generated, dynamically-linked notes [28]; Zep/Graphiti is a temporally-aware knowledge-graph engine retrieved via embeddings and reranking [18]; and MemMachine combines a vector database…”
Yan, J., Shen, Y., Zhang, W., Huang, J. W., Chen, P., Liu, Z., Wang, K., Wang, K., & Lian, S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.16285 · Cited by 5
“Retrieval-based systems, such as MemGPT (Packer et al., 2023), MemoryBank (Zhong et al., 2024), Mem0 (Chhikara et al., 2025) and A-MEM (Xu et al., 2025), store past interactions in external memory and retrieve relevant fragments when needed.”
A-TMA: Decoupling State-Aware Memory Failures in Long-Term Agent MemoryPreprint
Shi, Z., Tang, Y., & Tung, A. K. H. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.01935 · Cited by 5
“The LTP table includes Mem0 [6], InsideOut [37], AtomMem [18], AriadneMem [16], Graphiti/Zep [8, 19], SeCom [13], A-Mem [7], MemoryLLM [11], M+ [15], and completed A-TMA variants.”
OneDayAgent: Towards a Long-Horizon Harness for Autonomous AgentsPreprint
Zheng, J., Fang, X., Zhang, J., Gui, Z., Chen, H., & Zhang, N. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.05013 · Cited by 5
Harness the Memory: A Holistic Evaluation of Memory Substrates in Memory AgentsPreprint
Huang, W.-C., Zhang, W., Wu, Y., Chen, Y., Jiang, E. H., Yang, W., Yang, Y., Zou, H. P., Zhang, H., Wu, Y. N., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.15008 · Cited by 5
“Structural introduces relational links: M4 Evolving Notes [23] writes interlinked notes with evolution-triggered rewrites, and M5 Dual-Level Graph [24] builds a dual-level entity-relation graph.”
Evolutionary Safety of Recursive Self-Improving AI: Taxonomy, Risk Discovery, and EvaluationPreprint
Gong, C., Bi, J., Yao, D., Cheng, R., Yang, Y., Wang, K., Xinjian, L., Xiang, C. J., & Ruijie, G. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.31186 · Cited by 6
“ToolEvo extends this adaptation to tools revised through feedback [16], while MemGPT, Mem0, A-MEM, and Agent Workflow Memory support the organization, retrieval, and revision of stored context [24, 94, 127, 129].”
Conversational Agents: From RAG to LTM
Zhang, D., Feng, Y., Liu, H., Sun, C., Luo, J., Chen, X., & Li, X. (2025) · Proceedings of the 2025 Annual International ACM SIGIR Conference on Research and Development in Information Retrieval in the Asia Pacific Region · doi:10.1145/3767695.3769671 · Cited by 4
Memory Management and Contextual Consistency for Long-Running Low-Code Agents
Xu, J., & Lu, Q. (2026) · Lecture notes in networks and systems · doi:10.1007/978-3-032-20746-3_32 · Cited by 6
The Hitchhiker’s Guide to Autonomous Research: A Survey of Scientific Agents
Wang, X., Xu, J., Lian, S., Chen, Y., Guo, H., Zhu, F., Shao, Y., Ren, M., Yi, H., Feng, A. H., et al. (2026) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2026.3721500 · Cited by 12
H-MEM: Hierarchical Memory for High-Efficiency Long-Term Reasoning in LLM Agents
Sun, H., Zeng, S., & Zhang, B. (2026) · Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/2026.eacl-long.15 · Cited by 4
Yu, Y., Yao, L., Xie, Y., Tan, Q., Feng, J., Li, Y., & Wu, L. (2026) · Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/2026.acl-long.981 · Cited by 4
How Memory Management Impacts LLM Agents: An Empirical Study of Experience-Following Behavior
Xiong, Z., Lin, Y., Xie, W., He, P., Liu, Z., Tang, J., Lakkaraju, H., & Xiang, Z. (2026) · Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/2026.acl-long.27 · Cited by 4
Feng, T., Wang, X. E., & Zhu, W. (2026) · National Science Review · doi:10.1093/nsr/nwag662 · Cited by 6
Lumer, E., Gulati, A., Subbiah, V. K., Basavaraju, P. H., & Burke, J. A. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-21300-6_15 · Cited by 4
Xu, M., Wu, J., Song, S., Srirama, S. N., Javad, B., Ranjan, R., Jha, D. N., Wang, S., Tian, W., Xu, H., et al. (2026) · Software Practice and Experience · doi:10.1002/spe.70121 · Cited by 4
Memory Beyond Recall: A Dual-Process Cognitive Memory System for Self-Evolving LLM Agents
Fei, T., Song, M., Zheng, M., & Yu, X. (2026) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/2026.findings-emnlp.1918 · Cited by 4
LLM-MemCluster: Empowering Large Language Models with Dynamic Memory for Text Clustering
Zhu, Y., Yang, L., Xu, K., Zhang, W., Song, Z., Wang, J., & Yu, P. S. (2026) · Workshop on Towards Knowledgeable Foundation Models (KnowFM) · doi:10.18653/v1/2026.knowfm-1.7 · Cited by 4
Beyond the Context Window: Scaling Agentic RL via End-to-end Optimized Context Compression
Lu, M., Sun, W., Du, W., Ling, Z., Yao, X., Liu, K., & Chen, J. (2026) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2026.acl-long.966 · Cited by 10
Text2Mem: A Unified Memory Operation Language for Memory Operating System
Wang, L., Yang, L., Chen, B., Xu, K., Zou, G., Tang, B., Xiong, F., Chen, S., & li, Z. (2026) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2026.findings-acl.100 · Cited by 4
Hierarchical Reinforcement Learning with Augmented Step-Level Transitions for LLM Agents
Zhen, S., Yu, Y., Guo, R., Cheng, N., & Deng, Y. (2026) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2026.acl-long.318 · Cited by 6
ReContext: Recursive Evidence Replay as LLM Harness for Long-Context Reasoning
Zhao, Y., Qiu, R., Wei, T., Bei, Y., Liu, Z., Chen, L., Lourentzou, I., Tong, H., & He, J. (2026) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/2026.findings-emnlp.1255 · Cited by 5
Liu, R., Zhang, J., Schön, A., Müller, K., Zheng, J., Yang, K., Guo, A., Gerling, K., & Stiefelhagen, R. (2026) · International Journal of Human-Computer Interaction · doi:10.1080/10447318.2026.2695932 · Cited by 8
“In terms of software, once the infrastructure is upgraded to deploy all models on a remote server, SLAM (Matsuki et al., 2024; Yan et al., 2024; Zhu et al., 2024) and AI agents (LLMs with memory) (Xu et al., 2025; Yang, Yang, et al., 2025, Anthropic, 2025) can support holistic scene understanding.”
Organizational Memory for Agentic Business Process Execution
Kirchdorfer, L., Rebmann, A., Warmuth, C., Kampik, T., Heilker, T., & Berg, G. (2026) · Lecture notes in business information processing · doi:10.1007/978-3-032-36710-5_14 · Cited by 4
Conflict-Aware Self-Evolving Memory for Long-Horizon Agents
Wang, Y., Peng, C.-C., Shu, S., Wang, Z., & Lai, I.-H. (2026) · doi:10.1109/eecct71142.2026.11746545 · Cited by 4
“More recently, Mem0 [42] provides a production-ready memory layer with automatic fact extraction and vector-based retrieval, while A-Mem [43] proposes agentic memory with self-organizing link generation and memory evolution capabilities.”
When Users Don’t Ask: Benchmarking Context-Driven Memory Retrieval in Conversational Agents
Chang, W.-Y., & Chen, Y.-N. (2026) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/2026.findings-emnlp.1289 · Cited by 5
A Scenario-Driven Cognitive Approach to Next-Generation AI MemoryPreprint
Cai, L., Cheng, Y., Shao, X., Wang, H., Zhao, Y. B., Zhang, W., & Kang Li, B. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2509.13235 · Cited by 4
“To quantitatively assess the advantages of our proposed COLMA framework, we conducted a systematic evaluation against six prominent memory architectures for AI systems: A-Mem [109], MemO [110], MemOg [110], MEM1 [111], MIRIX [112], and Mem p [113].”
PISA: A Pragmatic Psych-Inspired Unified Memory System for Enhanced AI AgencyPreprint
Jia, S., Huang, Z., Wang, X., Zhang, H., & Song, M. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2510.15966 · Cited by 4
“Autonomous evolution strategies driven by LLMs generate new links and maintain balanced states without explicit rules, improving adaptability to long-term interactions (Xu et al., 2025).”
InfluentialNode-as-Agent: Graph Agentic NetworkPreprint
Guo, M., Zhu, X., Jiao, Q., Liu, X., Xue, H., Zhang, C., Lin, S., Huang, J., Ye, Z., & Zhang, Y. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2508.00429 · Cited by 4
“Inspired by general agent-based systems and LLM-based simulation environments [4, 11, 19, 27, 28, 32, 36, 39, 40], our agentic framework endows each node with four core components: Planning , which decides the next operations based on its current objective and context by a frozen LLM; Actions ,…”
Self-citationImproving Language Agents through BREW: Bootstrapping expeRientially-learned Environmental knoWledgePreprint
Kirtania, S., Biyani, P., Gupta, P., Bajpai, Y., Iyer, R. G., Gulwani, S., & Soares, G. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2511.20297 · Cited by 5
“…classified into either ones which (i) store only transient trajectory contexts that vanish between episodes like Mem0 (Chhikara et al., 2025; Xu et al., 2025b), or (ii) embed high-level notes directly in the prompt such as MetaRe-flection (Gupta et al., 2024b) and GEPA(Agrawal et al., 2025).”
Goal-Directed Search Outperforms Goal-Agnostic Memory Compression in Long-Context Memory TasksPreprint
Zheng, Y., McKee, K. L., Miconi, T., Bugaud, Z., van Gelderen, M., & McCaleb, J. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2511.21726 · Cited by 4
An, T. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2601.00821 · Cited by 4
“…salient facts (in its own harness the extracted memory, 66.9, trails its 72.9 full-context ceiling—the within-harness ordering we report); A-Mem (Xu et al., 2025) maintains an evolving note graph whose entries link and rewrite one another; SeCom (Pan et al., 2025) keeps verbatim segments but…”
InfluentialLLMs Can't Play Hangman: On the Necessity of a Private Working Memory for Language AgentsPreprint
Baldelli, D., Parviz, A., Zouaq, A., & Chandar, S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2601.06973 · Cited by 4
Liu, S., Yuan, H., Li, X., ZHU, Z., Cao, Y., & Jiang, Y.-G. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2601.09503 · Cited by 4
HyMem: Hybrid Memory Architecture with Dynamic Retrieval SchedulingPreprint
Zhao, X., Wang, K., Zhang, X., Yao, C., & Wang, A. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.13933 · Cited by 6
“In contrast, our HyMem dynamically allocates memory resources based on task demands, achieving dual optimization of performance and efficiency. et al., 2024; Tseng et al., 2024).”
Shi, Z., Ma, M., Yao, Z., Yang, F., Zhang, J., Han, D., Rühle, V., Lin, Q., Rajmohan, S., & Zhang, D. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.13830 · Cited by 4
MemSifter: Offloading LLM Memory Retrieval via Outcome-Driven Proxy ReasoningPreprint
Tan, J., Dou, Z., Zhang, L., Hu, Y., Cheng, Y., & Wen, J.-R. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2603.03379 · Cited by 6
“To address this, research has pursued two main directions: (1) Structural Enhancement during indexing, using extra computation to build richer indexes such as graphs [29, 34, 47] or hierarchies; and (2) Contextual Expansion during inference, increasing the working LLM’s input tokens.”
Self-citationMeta-TTL: Meta-Learning Self-Improvement Policies for Language AgentsPreprint
Lou, Z., Chen, H., Li, Y., Wang, Q., & Hooi, B. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.00830 · Cited by 4
Hierarchical Memory Orchestration for Personalized Persistent AgentsPreprint
Liu, J., Sun, Y., Cheng, W., Lei, H., Li, Y., Chen, Y., & Wang, D. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.01670 · Cited by 4
“Recent management strategies rely on flat vector databases or graph structures to store dialogue relations [21, 29, 34], often leading to bloated search spaces and high retrieval latency.”
Opal: Private Memory for Personal AIPreprint
Kaviani, D., Ozdarendeli, A. E., Zhu, J., Ding, Y., & Popa, R. A. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.02522 · Cited by 4
“However, agentic memory systems achieve high accuracy precisely because they augment semantic search with query-dependent traversals, using knowledge graphs (KGs), temporal filters, and multi-hop lookups to strategically narrow results [51, 87, 122, 155, 166, 195, 247].”
Yang, B., Xu, L., Li, Y., Liu, K., Jiang, X., & Yan, Z. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.06204 · Cited by 4
“Most existing studies derive personas for agent personalization primarily from chat histories and conversational interactions [10, 15, 53].”
Zhang, X. F., & Wang, L. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.07894 · Cited by 4
“…techniques range from static summarization with decay of older, less relevant memories (Zhong et al., 2024; Salama et al., 2025) to truly dynamic approaches executing autonomous, incremental updates (Kim et al., 2025; Nan et al., 2025; Xu et al., 2025; Rasmussen et al., 2025; Wang et al., 2025b).”
InfluentialMemGround: Long-Term Memory Evaluation Kit for Large Language Models in Gamified ScenariosPreprint
Ding, Y., Xia, W., Zhao, Y., Su, J., Yang, J., Zhang, Z., Wang, K., & Yang, W. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.14158 · Cited by 4
“…et al., 2025; Jimenez Gutierrez et al., 2024; Su et al., 2025); (2) Dynamic memory exploration, which refers to design dynamic memory update mechanisms, enabling LLMs to track information changes in various scenarios (Shinn et al., 2023; Xu et al., 2025; Wang et al., 2025c; Chhikara et al., 2025).”
Training LLM Agents for Spontaneous, Reward-Free Self-Evolution via World Knowledge ExplorationPreprint
Zhang, Q., Ma, D., Fang, T., Li, J., Tang, J., Chen, N., Mi, H., & Wang, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.18131 · Cited by 4
ContextWeaver: Selective and Dependency-Structured Memory Construction for LLM AgentsPreprint
Wu, Y., Zhang, Y., Ghosh, S., Basu, S., Deoras, A., Huan, J., & Gupta, G. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.23069 · Cited by 4
“Recent work on graph-based memory organization primarily focuses on graph construction and retrieval (Xu et al., 2025), with less attention to how these structures can be simplified or optimized over time.”
Self-citationLin, J., Ling, Z., Zhou, C., Xu, T., Jiang, R., Wang, Z., & Ge, D. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.25847 · Cited by 4
MemRouter: Memory-as-Embedding Routing for Long-Term Conversational AgentsPreprint
Hu, T., Lin, W., Zhang, W., Ma, J., & Wang, S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.00356 · Cited by 4
From History to State: Constant-Context Skill Learning for LLM AgentsPreprint
Xie, H., Wang, X., Wang, Y., Zhao, P., & Ju, F. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.05413 · Cited by 5
“A large line of memory-augmented agents reduces this burden by storing, summarizing, retrieving, or compressing past interactions [44, 37, 30, 3, 41, 13, 15].”
The Trap of Trajectory: Towards Understanding and Mitigating Spurious Correlations in Agentic MemoryPreprint
Tang, L., Vaje, R. R., 孟宇桥, Narkar, S. S., Ma, W., Ding, Z., Zhang, D., & Xi, Z. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.09330 · Cited by 4
“Embedding-based (vector-based) memory : A-Mem [57] and Mem0 [8].”
Formalize, Don't Optimize: The Heuristic Trap in LLM-Generated Combinatorial SolversPreprint
Wang, H., Song, Y., Li, T., Deng, Z., Wang, Y., Ramachandran, D., Cohen, E., & Roth, D. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.12421 · Cited by 4
“Python + OR-Tools attains the highest correctness at the 256 s tail on every LLM; the dotted–solid gap is largest on Python and smallest on MiniZinc + OR-Tools ; the heuristic prompt fails to reliably dominate the baseline, previewing the heuristic trap . working memory [Xu et al., 2025].”
MedMemoryBench: Benchmarking Agent Memory in Personalized HealthcarePreprint
Wang, Y., Xu, H., Gu, R., Ye, Y., Chen, X., Mu, X., Gao, Y., Guo, C., Wei, P., Gu, J., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.11814 · Cited by 5
Structured Belief State and the First Precision-Aware Benchmark for LLM Memory RetrievalPreprint
Flynt, J. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.11325 · Cited by 4
“A-MEM [7] treats memory as a self-organizing network of Zettelkasten-inspired notes where new memories trigger updates to the contextual representations of existing ones.”
RecMem: Recurrence-based Memory Consolidation for Efficient and Effective Long-Running LLM AgentsPreprint
Dai, Z., Deng, S., Guan, S., Tian, Y., Yao, X., Yan, X., & Cheng, J. S.-C. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.16045 · Cited by 4
“Most existing memory systems construct long-term memory by incrementally transforming in-coming interactions (or short windows thereof) into retrievable memory units, such as summaries (Kang et al., 2025; Packer et al., 2024; Zhong et al., 2023), atomic facts (Chhikara et al., 2025; Wang and Chen, 2025), or structured…”
InfluentialMem-$π$: Adaptive Memory through Learning When and What to GeneratePreprint
Wang, X., Wang, C., Nekoei, H., Pal, C., Lacoste, A., Gella, S., Liu, B., & Taslakian, P. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.21463 · Cited by 4
“…reinforcement learning, training controllers to manage storage, retrieval, update, and deletion (Yu et al., 2026, 2025a; Zhang et al., 2026a,b; Zhan et al., 2025), or co-evolving memory architectures with the agent’s policy (Xu et al., 2025; Zhou et al., 2025; Xiao et al., 2026; Wang et al., 2026).”
Personal Visual Memory from Explicit and Implicit EvidencePreprint
Nguyen, V., Nguyen, T., Patel, V. M., & Li, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.28806 · Cited by 6
“A broad line of work augments LLMs with external memory through retrieval-augmented generation (RAG) [16, 17, 18] or explicit memory modules [4, 5, 11, 19, 20, 21, 22, 23, 24].”
FlowBank: Query-Adaptive Agentic Workflows Optimization through Precompute-and-ReusePreprint
Yuan, L., Deng, C., Yu, F., Chakraborty, S., Rostami, M., & Huang, F. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.11290 · Cited by 4
“One promising direction is to connect FlowBank with recent memory-based agent frameworks such as A-Mem (Xu et al., 2025), ArcMemo (Ho et al., 2025), MemoryBank (Zhong et al., 2024), and ReasoningBank (Ouyang et al., 2026), so that it stores not only reusable workflows but also structured experience…”
Jin, Y., Xu, Y., Zhu, J., & Yang, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.17162 · Cited by 4
ECHO: Prune To Act, Trace To Learn With Selective Turn Memory In Agentic RLPreprint
Xie, Z., Zheng, B., Gong, E., Liu, J., You, Y., Liu, L., Tang, J., Zhao, G., Hu, A., & Chen, Z. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.31650 · Cited by 5
“…maintain external stores that can be written to and retrieved across interactions (Packer et al., 2023; Zhong et al., 2024; Chhikara et al., 2025; Xu et al., 2025; Shan et al., 2025; Wang & Chen, 2025), with recent systems structuring memory through dependency graphs, subgoal-oriented working…”
Shortening agent skills leaves LLM agent cost unchangedPreprint
Xu, X. X., & Wu, W. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.03048 · Cited by 4
MANTA: Multi-Agent Network Topology Adaptation for Self-Evolving Multi-Agent SystemsPreprint
Huang, M.-X., Wang, J., Lai, Y.-C., Zhang, Z., Cardie, C., & Huang, H.-H. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.28527 · Cited by 4
The Horizon Gap: Planning, Memory, Execution, Training, and Evaluation for Long-Horizon LLM AgentsPreprint
Chen, M., Wang, L., & Qu, B. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.06663 · Cited by 6
Mitigating Over-Personalization in LLMs via Structured MemoryPreprint
Hannoon, H., Zhao, A., Narayan, M., Goyal, S., & Sheth, I. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.08300 · Cited by 4
“…graphs, or links: G ENERATIVE A GENTS derives reflective memories from an observation stream (Park et al., 2023); Zep/Graphiti organizes memory as temporally aware episode, entity, and community subgraphs (Rasmussen et al., 2025); and A-MEM stores memories as linked, tagged notes (Xu et al., 2025).”
Li, R., Zhang, L., Xu, B., Du, M., Fu, Z., & Chen, W. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.12888 · Cited by 5
“…2020) retrieves external evidence at inference time; Mem0 (Chhikara et al., 2025) extracts structured entries and stores them as embeddings; A-Mem (Xu et al., 2025) organizes memories as linked note cards; GraphRAG (Edge et al., 2024), HippoRAG (Gutiérrez et al., 2024, 2025), and Zep (Rasmussen…”
InfluentialRippleMem: From Isolated Retrieval to Associative Recollection for Long-Term Agent MemoryPreprint
Ji, J., Li, L., Cheng, X., Zhou, Y., Zhang, W., Tan, Y., & Qin, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.13334 · Cited by 4
“A complementary line introduces explicit structure or association, using graphs, temporal links, or recollection-style retrieval to let memory access move beyond independent record lookup (Xu et al., 2025b; Rasmussen et al., 2025; Zhang et al., 2026).”
Agentic ESOpt: Fine-Tuning Long-Horizon LLM Agents with Minimal GPU RequirementsPreprint
Zheng, Z., Chen, R., 巴云鹏, Wang, Z., Teh, Y. W., & Lee, W. S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.17310 · Cited by 4
NavHarness: Towards Lifelong Embodied NavigationPreprint
Zhao, X., Zhou, J., Lin, S., Zhou, G., Li, Z., Yan, X., Liu, J., van den Hengel, A., & Wu, Q. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.34276 · Cited by 4
Agent Plasticity: Measuring Self-Improvement Through ExperiencePreprint
Singh, H., Bakhtin, A., Shao, R., Synnaeve, G., Kulikov, I., Fergus, R., Arora, S., Keutzer, K., Weston, J., Mahajan, A., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2610.08902 · Cited by 6
“Dynamic Cheatsheet and ReasoningBank accumulate strategies across interactions (Suzgun et al., 2026; Ouyang et al., 2026), while memory architectures organize persistent information for future retrieval (Park et al., 2023; Xu et al., 2025).”
Is Agent Memory a Database? Rethinking Data Foundations for Long-Term AI Agent Memory
Orogat, A., & Mansour, E. (2026) · arXiv · doi:10.48550/arxiv.2605.26252 · Cited by 4
From Human Memory to AI Memory: A Survey on Memory Mechanisms in the Era of LLMs
Wu, Y., Liang, S., Zhang, C., Wang, Y., Zhang, Y., Guo, H., Tang, R., & Liu, Y. (2025) · arXiv · doi:10.48550/arxiv.2504.15965 · Cited by 3
“…Conditional Memory [47], SECOM [48], Memory 3 [49], MemInsight [50] Management MemoChat [51], MemoryBank [17], RMM [52], LD-Agent [53], A-MEM [54], Generative Agents [55], EMG-RAG [56], KGT [46], LLM-Rsum [57], COMEDY [58] Retrieval RET-LLM [44], ChatDB [59], Human-like Memory [60], HippoRAG…”
Self-citationFrom Storage to Experience: A Survey on the Evolution of LLM Agent Memory Mechanisms
Luo, J., Tian, Y., Cao, C., Luo, Z., Lin, H., Li, K., Kong, C., Yang, R., & Ma, J. (2026) · Findings of the Association for Computational Linguistics: ACL 2026 · doi:10.18653/v1/2026.findings-acl.2069 · Cited by 3
“More importantly, the persistent retrieval of irrelevant or obsolete memories can disrupt the coherence in reasoning of the LLM agent (Xu et al., 2025b; Tan et al., 2025).”
LEGOMem: Modular Procedural Memory for Multi-agent LLM Systems for Workflow Automation
Han, D., Couturier, C., Madrigal Diaz, D., Zhang, X., Rühle, V., & Rajmohan, S. (2026) · Proceedings of the 25th International Conference on Autonomous Agents and Multiagent Systems · doi:10.65109/vlua1303 · Cited by 3
“…for single-agent systems and are often episodic/semantic, replaying information from dialogue histories [18, 21, 24, 31, 38], such as A-MEM [34] which captures interactions as a network of interconnected notes that form an evolving memory structure, and Mem0 [6], which focus on managing…”
Gao, H.-A., Geng, J., Hua, W., Hu, M., Juan, X., Liu, H., Liu, S., Qiu, J., Qi, X., Wu, Y., et al. (2025) · arXiv · doi:10.48550/arxiv.2507.21046 · Cited by 3
Self-citationYan, D., Zhang, H., Huzhang, G., (7082), Y. L., Wang, Y., Chen, Q.-G., Xu, Z., Luo, W., (38224), Y. L., Dong, W., et al. (2026) · IEEE Transactions on Multimedia · doi:10.1109/tmm.2026.3742640 · Cited by 4
Chen, X., Manjunatha, V., Wang, X., & Siu, A. F. (2026) · CHI Conference on Human Factors in Computing Systems · doi:10.1145/3772318.3791669 · Cited by 5
“Second, recent memory systems, from ChatGPT’s memory 1 to agent memory architectures [11, 42, 58], capture observations and preferences but rarely persist process-oriented reasoning.”
S3Mem: Structured Spatiotemporal Scene-Event Memory for Long-Horizon Interactive Question Answering
Su, E., Wu, J., Zhang, J., Yu, Q., Tang, C., Li, P., Wang, L., Wang, A., Ma, X., Tang, S., et al. (2026) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/2026.emnlp-main.992 · Cited by 3
Role-Agent: Bootstrapping LLM Agents via Dual-Role Evolution
Wang, X., Ma, Z., Yang, S., Huang, T., Wang, P., Wang, Y., & Chu, X. (2026) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/2026.findings-emnlp.208 · Cited by 3
Jiang, Z. (2026) · Design and Artificial Intelligence · doi:10.1016/j.daai.2026.100087 · Cited by 3
DMV-Bench: Diagnosing Long-Horizon Multimodal Agents’ Visual Memory with Incidental Cue Injection
Tang, Y., Shang, C., Xu, R., & Singh, N. (2026) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/2026.emnlp-main.1892 · Cited by 3
CodeMEM: AST-Guided Adaptive Memory for Repository-Level Iterative Code Generation
Wang, P., Zhang, L., Liu, F., Tao, C., & Zhu, Y. (2026) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2026.findings-acl.834 · Cited by 4
Van, H. P., Hieu, N. M., Tuan, K. P. T., Hai, N. L., Van, L. N., Diep, N. T. N., & Le, T. (2026) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2026.findings-acl.1408 · Cited by 13
“To balance coherence and relevance, Pan et al. (2025) proposed SECOM , which segments dialogue into coherent topical units and applies compression-based denoising, while Xu et al. (2025) developed A-MEM , constructing dynamic “atomic notes” linked by shared attributes.”
MemPO: Self-Memory Policy Optimization for Long-Horizon Agents
Li, R., Zhang, X., Yu, H., Duan, S., Li, X. L., Xiang, W., Liao, C., Guo, X., Li, Y., & Suo, J. (2026) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2026.findings-acl.1166 · Cited by 3
Hybrid Self-evolving Structured Memory for Computer-Use Agents
Zhu, S., Wenyi, W. U., Zhou, K., Wang, S., & Huang, B. (2026) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2026.findings-acl.549 · Cited by 5
“To improve organization, graph-based systems (Xu et al., 2025; Zhang et al., 2025a) structure knowledge as connected nodes, enabling agents to perform multi-hop retrieval and reasoning beyond similarity matching (Shen et al., 2025).”
Learning How and What to Memorize: Cognition-Inspired Two-Stage Optimization for Evolving Memory
Xu, D., Liu, S., Luo, P., Jia, P., Zhang, Y., Wen, Y., Deng, Y., Zhang, W., Chen, E., Zhao, X., et al. (2026) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2026.acl-long.2084 · Cited by 8
Self-citationSafe and Scalable Collaboration in Multiagent LLM Systems: A Comprehensive Review
Hossain, E., Nipu, M. M. H., Mahmood, M. S., Hossen, M. J., & Mridha, M. F. (2026) · IEEE Transactions on Systems Man and Cybernetics Systems · doi:10.1109/tsmc.2026.3704902 · Cited by 4
“They exhibit capabilities such as multiagent orchestration [22], memory generation [23], recursive self-prompting [24], and tool use [25].”
UniMem: Complementary Episodic-to-Parametric Memory for Boundary-Agnostic Task Streams
Xia, S., Zhang, C., Wu, Y., Li, H., Chai, J., Wang, X., Yin, G., Lin, W., Lin, Z., Zhang, H., et al. (2026) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/2026.emnlp-main.1490 · Cited by 3
CleanS2S: Single-file Framework for Proactive Speech-to-Speech InteractionPreprint
Lu, Y.-D., Niu, Y., Hu, S., & Wang, H. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2506.01268 · Cited by 5
From Aggregation to Selection: User-Validated Distributed Social RecommendationPreprint
Huang, J., Luo, D., Ye, Z., Chen, W., Guo, M., & Zhang, Y. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2505.21388 · Cited by 3
“Furthermore, the storage and computational capabilities of each node are limited, making it infeasible to deploy large-scale models such as LLM-based encoders [28, 53, 20, 49] or even invoke external LLM APIs [73, 33, 36, 74, 89, 86, 22, 73, 23, 21].”
Self-citationMAPLE: Multi-Agent Adaptive Planning with Long-Term Memory for Table ReasoningPreprint
Bai, Y., Wang, M., & Vu, T.-T. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2506.05813 · Cited by 4
“Inspired by frameworks like A-MEM (Xu et al., 2025), we adapt their approaches specifically for table-based reasoning challenges.”
InfluentialAlla, C. V. K., Gaddam, H. N., Kommi, M., & Ahmed, S. N. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2511.04919 · Cited by 4
“A-Mem [35] implements selective top-k retrieval for agents with 85-93% token reduction.”
CubeBench: Diagnosing Interactive, Long-Horizon Spatial Reasoning Under Partial ObservationsPreprint
Gao, H.-A., Zhang, Z., Luo, T., Yang, K., Juan, X., Qiu, J., Chen, T., He, B., Zhao, H., Zhou, H., et al. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2512.23328 · Cited by 3
Self-citationLearning Context: A Unified Framework and Roadmap for Context-Aware AI in EducationPreprint
Liu, N., Bradford, B. C., Hatchett, J., Ramos, G. D., Luzi, L., Wang, Z., Mallick, D. B., & Baraniuk, R. G. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2512.24362 · Cited by 4
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.