Elamur Research⌕
☰
home / papers / 2025 / MemInsight: Autonomous Memory Augmentation for LLM Agents

MemInsight: Autonomous Memory Augmentation for LLM Agents

Paper · 2025 · Rana Salama, Jason Cai, Michelle Yuan et al.

Summary

LLM agents rely on long-term memory to give coherent, personalized answers. As stored interactions pile up, raw and unstructured memory becomes noisy and harder to search. MemInsight addresses this by having an LLM backbone enrich each memory entry automatically with attribute–value annotations, without a hand-designed schema.

The framework has three modules:

  • Attribute Mining chooses a perspective (entity-centric or conversation-centric) and a granularity (turn-level or session-level).
  • Annotation attaches the attributes to the memory entry, either unordered (Basic) or ranked by relevance (Priority).
  • Memory Retrieval either returns all related entries with their annotations, or uses the annotations for refined retrieval through attribute filtering or embeddings.

The authors test the approach on three tasks: conversational recommendation (LLM-REDIAL), question answering and event summarization (LoCoMo). Backbones include Claude-3-Sonnet, Claude-3-Haiku, Llama v3 and Mistral v1. MemInsight produces more persuasive recommendations and substantially higher retrieval recall than a RAG baseline. Its annotations also proved factually reliable.

The work matters because it shows that automatically structured memory can make agent retrieval more precise and context-aware. It also complements standard RAG pipelines rather than replacing them.

Key points

  • On LLM-REDIAL, recommendation persuasiveness rises by as much as 14%. On LoCoMo, recall improves by 34% over a RAG baseline.
  • Priority Augmentation, which orders attributes by relevance, improves similarity search and retrieval accuracy compared with unordered Basic Augmentation.
  • In event summarization, generating the annotations with Claude-3-Sonnet and summarizing with Llama v3 raised G-Eval Relevance from 2.03 to 3.15 over the non-augmented baseline. Turn-level annotations generally beat session-level ones.
  • A DeepEval hallucination check found that 99.14% of Claude-3-Sonnet annotations were grounded in the source dialogues. The remaining 0.86% were mostly vague attributes rather than factual errors.
  • The method is text-only and its quality depends on the backbone LLM. The code is published at github.com/amazon-science/MemInsight.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

Cite

Cite as Rana Salama, Jason Cai, Michelle Yuan, Anna Currey, Monica Sunkara, Yi Zhang and Yassine Benajiba. 2025. MemInsight: Autonomous Memory Augmentation for LLM Agents. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2025.emnlp-main.1683

@inproceedings{salama2025meminsight,
  title = {{MemInsight: Autonomous Memory Augmentation for LLM Agents}},
  author = {Rana Salama, Jason Cai, Michelle Yuan, Anna Currey, Monica Sunkara, Yi Zhang and Yassine Benajiba},
  booktitle = {Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing},
  year = {2025},
  doi = {10.18653/v1/2025.emnlp-main.1683},
  url = {https://doi.org/10.18653/v1/2025.emnlp-main.1683}
}

References

32 references, 10 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Petr Anokhin, Nikita Semenov, Artyom Sorokin, Dmitry Evseev, Mikhail Burtsev, and Evgeny Burnaev. 2024. Arigraph: Learning knowledge graph world models with episodic memory for llm agents. Preprint, arXiv:2407.04363. — [arXiv:2407.04363](https://arxiv.org/abs/2407.04363) find
  2. Prateek Chhikara, Dev Khant, Saket Aryan, Taranjeet Singh, and Deshraj Yadav. 2025. Mem0: Building production-ready ai agents with scalable long-term memory. Preprint, arXiv:2504.19413.
  3. Yubin Ge, Salvatore Romeo, Jason Cai, Raphael Shu, Monica Sunkara, Yassine Benajiba, and Yi Zhang. 2025. Tremu: Towards neuro-symbolic temporal reasoning for llm-agents with memory in multi-session dialogues. arXiv preprint arXiv:2502.01630. find
  4. Yuki Hou, Haruki Tamoto, and Homei Miyashita. 2024. “my agent understands me better”: Integrating dynamic human-like memory recall and consolidation in llm-based agents. In Extended Abstracts of the CHI Conference on Human Factors in Computing Systems, page 1–7. ACM. find
  5. Chenxu Hu, Jie Fu, Chenzhuang Du, Simian Luo, Junbo Zhao, and Hang Zhao. 2023a. Chatdb: Augmenting llms with databases as their symbolic memory. arXiv preprint arXiv:2306.03901. — [arXiv:2306.03901](https://arxiv.org/abs/2306.03901) find
  6. Chenxu Hu, Jie Fu, Chenzhuang Du, Simian Luo, Junbo Zhao, and Hang Zhao. 2023b. Chatdb: Augmenting llms with databases as their symbolic memory. Preprint, arXiv:2306.03901. — [arXiv:2306.03901](https://arxiv.org/abs/2306.03901) find
  7. Mengkang Hu, Tianxing Chen, Qiguang Chen, Yao Mu, Wenqi Shao, and Ping Luo. 2024. Hiagent: Hierarchical working memory management for solving long-horizon agent tasks with large language model. Preprint, arXiv:2408.09559.
  8. Jeff Johnson, Matthijs Douze, and Hervé Jégou. 2017. Billion-scale similarity search with gpus. Preprint, arXiv:1702.08734.
  9. Vladimir Karpukhin, Barlas Oğuz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen tau Yih. 2020. Dense passage retrieval for open-domain question answering. Preprint, arXiv:2004.04906.
  10. Kuang-Huei Lee, Xinyun Chen, Hiroki Furuta, John Canny, and Ian Fischer. 2024. A human-inspired reading agent with gist memory of very long contexts. Preprint, arXiv:2402.09727.
  11. Tingting Liang, Chenxin Jin, Lingzhi Wang, Wenqi Fan, Congying Xia, Kai Chen, and Yuyu Yin. 2024. LLM-REDIAL: A large-scale dataset for conversational recommender systems created from user behaviors with LLMs. In Findings of the Association for Computational Linguistics: ACL 2024, pages 8926–8939, B… find
  12. Lei Liu, Xiaoyan Yang, Yue Shen, Binbin Hu, Zhiqiang Zhang, Jinjie Gu, and Guannan Zhang. 2023a. Think-in-memory: Recalling and post-thinking enable llms with long-term memory. Preprint, arXiv:2311.08719. — [arXiv:2311.08719](https://arxiv.org/abs/2311.08719) find
  13. Zhengyang Tang, Ziniu Li, Zhenyang Xiao, Tian Ding, Ruoyu Sun, Benyou Wang, Dayiheng Liu, Fei Huang, Tianyu Liu, Bowen Yu, and Junyang Lin. 2025. Enabling scalable oversight via self-evolving critic. Preprint, arXiv:2501.05727. — [arXiv:2501.05727](https://arxiv.org/abs/2501.05727) find
  14. Junlin Wang, Jue Wang, Ben Athiwaratkun, Ce Zhang, and James Zou. 2024a. Mixture-of-agents enhances large language model capabilities. Preprint, arXiv:2406.04692. find
  15. Lei Wang, Chen Ma, Xueyang Feng, Zeyu Zhang, Hao Yang, Jingsen Zhang, Zhiyuan Chen, Jiakai Tang, Xu Chen, Yankai Lin, Wayne Xin Zhao, Zhewei Wei, and Jirong Wen. 2024b. A survey on large language model based autonomous agents. Frontiers of Computer Science, 18(6).
  16. Qineng Wang, Zihao Wang, Ying Su, Hanghang Tong, and Yangqiu Song. 2024c. Rethinking the bounds of llm reasoning: Are multi-agent discussions the key? Preprint, arXiv:2402.18272. find
  17. Qineng Wang, Zihao Wang, Ying Su, Hanghang Tong, and Yangqiu Song. 2024d. Rethinking the bounds of llm reasoning: Are multi-agent discussions the key? Preprint, arXiv:2402.18272. find
  18. Qineng Wang, Zihao Wang, Ying Su, Hanghang Tong, and Yangqiu Song. 2024e. Rethinking the bounds of llm reasoning: Are multi-agent discussions the key? Preprint, arXiv:2402.18272. find
  19. Zora Zhiruo Wang, Jiayuan Mao, Daniel Fried, and Graham Neubig. 2024f. Agent workflow memory. Preprint, arXiv:2409.07429.
  20. Wujiang Xu, Kai Mei, Hang Gao, Juntao Tan, Zujie Liang, and Yongfeng Zhang. 2025. A-mem: Agentic memory for llm agents. Preprint, arXiv:2502.12110.
  21. Yang Liu, Dan Iter, Yichong Xu, Shuohang Wang, Ruochen Xu, and Chenguang Zhu. 2023b. G-eval: Nlg evaluation using gpt-4 with better human alignment. Preprint, arXiv:2303.16634.
  22. Yixin Yang, Zheng Li, Qingxiu Dong, Heming Xia, and Zhifang Sui. 2024. Can large multimodal models uncover deep semantics behind images? Preprint, arXiv:2402.11281. find
  23. Adyasha Maharana, Dong-Ho Lee, Sergey Tulyakov, Mohit Bansal, Francesco Barbieri, and Yuwei Fang. 2024. Evaluating very long-term conversational memory of llm agents. Preprint, arXiv:2402.17753. find
  24. Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2023. React: Synergizing reasoning and acting in language models. Preprint, arXiv:2210.03629.
  25. Charles Packer, Sarah Wooders, Kevin Lin, Vivian Fang, Shishir G. Patil, Ion Stoica, and Joseph E. Gonzalez. 2024. Memgpt: Towards llms as operating systems. Preprint, arXiv:2310.08560. find
  26. Zeyu Zhang, Xiaohe Bo, Chen Ma, Rui Li, Xu Chen, Quanyu Dai, Jieming Zhu, Zhenhua Dong, and Ji-Rong Wen. 2024. A survey on the memory mechanism of large language model based agents. Preprint, arXiv:2404.13501. find
  27. Gustavo Penha, Ali Vardasbi, Enrico Palumbo, Marco de Nadai, and Hugues Bouchard. 2024. Bridging search and recommendation in generative retrieval: Does one task help the other? Preprint, arXiv:2410.16823. — [arXiv:2410.16823](https://arxiv.org/abs/2410.16823) find
  28. Jihoon Tack, Jaehyung Kim, Eric Mitchell, Jinwoo Shin, Yee Whye Teh, and Jonathan Richard Schwarz. 2024. Online adaptation of language models with a memory of amortized contexts. arXiv preprint arXiv:2403.04317. find
  29. Andrew Zhao, Daniel Huang, Quentin Xu, Matthieu Lin, Yong-Jin Liu, and Gao Huang. 2024a. Expel: Llm agents are experiential learners. Preprint, arXiv:2308.10144. find
  30. Andrew Zhao, Daniel Huang, Quentin Xu, Matthieu Lin, Yong-Jin Liu, and Gao Huang. 2024b. Expel: Llm agents are experiential learners. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 19632–19642. find
  31. Wanjun Zhong, Lianghong Guo, Qiqi Gao, He Ye, and Yanlin Wang. 2023. Memorybank: Enhancing large language models with long-term memory. Preprint, arXiv:2305.10250. find
  32. Xizhou Zhu, Yuntao Chen, Hao Tian, Chenxin Tao, Weijie Su, Chenyu Yang, Gao Huang, Bin Li, Lewei Lu, Xiaogang Wang, Yu Qiao, Zhaoxiang Zhang, and Jifeng Dai. 2023. Ghost in the minecraft: Generally capable agents for open-world environments via large language models with text-based knowledge and mem… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/salama-2025-meminsight-autonomous-memory-augmentation-llm/