[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-82996-en":3,"doc-seo-82996-105":30,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},82996,7971461740886,"Theodore","https://ap-avatar.wpscdn.com/davatar_3d24733baf745e90a7e4bdd5f77d97b2",8,"Research & Report","From Passive Retrieval to Active Memory Navigation Learning to Use Memory as a Structured Action Space","Long-term user memory is crucial for personalized conversational agents, yet many systems rely on passive retrieval that feeds pre-selected context and limits how thoroughly models can inspect user evidence. NapMem introduces a framework that treats long-term memory as a structured action space, organizing interaction histories into a linked multi-granularity memory pyramid. Memory tools expose different abstraction levels, and reinforcement learning trains the agent to navigate and decide which evidence granularity to consult, improving performance on memory-intensive tasks.","arXiv :2607 .05794v 1 [ cs .AI ] 7 Jul 2026  \nFrom Passive Retrieval to Active Memory Navigation: Learning to Use Memory as a Structured Action Space  \nYue Xu* 1,2 , Yutao Sun* 3 , Yihao Liu*4 , Mengyu Zhou†1, Jiayi Qiao* 5 , Lu Ma1 , Kai Tang1 , Wenjie Wang†2, Xiaoxi Jiang1 and Guanjun Jiang1  \n1 Qwen Large Model Application Team, Alibaba, 2 ShanghaiTech University, 3 Zhejiang University, 4Peking University, 5National University of Singapore  \n*Work done during an internship at Alibaba. †Corresponding author.  \nLong-term user memory is essential for personalized conversational agents, yet many memory systems still expose memory through passive retrieval interfaces, making the model a consumer of pre-selected evidence. We introduce NapMem, a framework for learning to use long-term user memory as a structured action space rather than passively retrieved context. NapMem organizes user history into a linked multigranularity memory pyramid, where raw conversations, typed memory records, topic tracks, and user profiles are connected through provenance relations, and exposes these levels through memory tools. The agent is trained to select memory according to the query and intermediate evidence, allowing it to inspect different memory granularities before answering. Experiments on PersonaMem-v2, LongMemEval, and LoCoMo show that a NapMem agent trained with memory-tool reinforcement learning is competitive across diverse memory-intensive tasks, while evaluations on non-memory tasks suggest that the learned policy largely preserves general reasoning and tool-use abilities. Additional analyses examine storage, inference cost, tool-use behavior, and ablations over navigation, memory granularity, and RL training. Our results suggest that long-term user memory benefits from coupling structured storage with a learned policy for using memory at the appropriate granularity.  \n1. Introduction  \nAs Large Language Model (LLM) agents are deployed in personal assistance, education, workplace productivity, and other long-running interactive settings, they are expected to maintain continuity and personalization across sessions, requiring agents to build a persistent understanding of the user. Long-term user memory is therefore becoming an essential component for personalized conversational agents (Hu et al., 2025; Packer et al., 2024; Xu et al., 2026b) . However, building an effective user memory system is challenging since user information is multifaceted and future queries impose heterogeneous memory requirements (Zhong et al., 2023) . Different queries may require different types of user memory, making it difficult for a fixed memory representation or access strategy to generalize across future interactions.  \nExisting efforts improve user memory from two directions: memory construction and memory retrieval. On the construction side, recent systems build richer memory infrastructures by designing finer-grained memory categories, introducing vector, graph, hybrid, or hierarchical  \nFigure 1 | Motivating example for active memory navigation. Passive retrieval leaves the agent with partial evidence, while NapMem exposes longterm user memory as an action interface, enabling the agent to search for sufficient evidence before answering.  \nCorresponding author(s): [wangwj1@shanghaitech.edu.cn](wangwj1@shanghaitech.edu.cn), [zhoumengyu.zmy@alibaba-inc.com](zhoumengyu.zmy@alibaba-inc.com)  \nFigure 2 | Overview of NapMem. The agent actively navigates a multi-granularity long-term memory pyramid through memory tools, selecting appropriate abstraction levels and refining actions based on intermediate evidence.  \nstorage structures, and using agentic methods to manage memory entries (Chhikara et al., 2025; Rasmussen et al., 2025; Li et al., 2025; Kang et al., 2025) . On the retrieval side, another line of work makes memory access more adaptive by introducing reflective retrieval, personalizing retrieval queries, or optimizing retrieval reranking (Tan et al.,","cbCaiqOFhxQ2muz3","https://ap.wps.com/l/cbCaiqOFhxQ2muz3","pdf",1712732,4,1,20,"English","en",105,"# Introduction\n## Background and motivation\n## Related work: memory construction and retrieval\n# Method: NapMem\n## Multi-granularity memory pyramid\n## Memory tools and provenance relations\n## Reinforcement learning for navigation\n# Experiments and analysis\n## Memory-intensive evaluations\n## Non-memory task generalization\n## Cost, tool-use behavior, and ablations","[{\"question\":\"What problem does NapMem address in current user memory systems?\",\"answer\":\"Many systems expose user memory through passive retrieval, which delivers pre-selected context and restricts the agent’s ability to inspect additional evidence when it is incomplete. NapMem aims to improve memory use by enabling active navigation across evidence levels.\"},{\"question\":\"How does NapMem represent long-term user memory?\",\"answer\":\"NapMem organizes user interaction histories into a linked multi-granularity memory pyramid. It connects raw conversations, typed memory records, topic tracks, and user profiles via provenance relations to support navigation across abstraction levels.\"},{\"question\":\"How is the NapMem agent trained to use memory?\",\"answer\":\"The agent is trained with reinforcement learning to select memory through memory tools based on the query and intermediate evidence. This makes memory usage an explicit, optimizable part of the agent’s decision process.\"}]",1784184548,50,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"from-passive-retrieval-to-active-memory-navigation-learning-to-use-memory-as-a-structured-action-space","",{"@graph":36,"@context":85},[37,53,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,51],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":50},"https://docshare.wps.com/document/research-report/",3,{"item":52,"name":13,"@type":43,"position":20},"https://docshare.wps.com/document/from-passive-retrieval-to-active-memory-navigation-learning-to-use-memory-as-a-structured-action-space/82996/",{"url":52,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":24,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":41,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-23","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What problem does NapMem address in current user memory systems?","Question",{"text":75,"@type":76},"Many systems expose user memory through passive retrieval, which delivers pre-selected context and restricts the agent’s ability to inspect additional evidence when it is incomplete. NapMem aims to improve memory use by enabling active navigation across evidence levels.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does NapMem represent long-term user memory?",{"text":80,"@type":76},"NapMem organizes user interaction histories into a linked multi-granularity memory pyramid. It connects raw conversations, typed memory records, topic tracks, and user profiles via provenance relations to support navigation across abstraction levels.",{"name":82,"@type":73,"acceptedAnswer":83},"How is the NapMem agent trained to use memory?",{"text":84,"@type":76},"The agent is trained with reinforcement learning to select memory through memory tools based on the query and intermediate evidence. This makes memory usage an explicit, optimizable part of the agent’s decision process.","https://schema.org",{"og:url":52,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":52},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":92},[93,97,101,105,110,114,119,122,126,129,133],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":20,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":29,"slug":113},6,"Technology","technology",{"id":115,"doc_module":4,"doc_module_name":46,"category_name":116,"show_sort_weight":117,"slug":118},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":120,"slug":121},30,"research-report",{"id":123,"doc_module":4,"doc_module_name":46,"category_name":124,"show_sort_weight":22,"slug":125},9,"Religion & Spirituality","religion-spirituality",{"id":22,"doc_module":4,"doc_module_name":46,"category_name":127,"show_sort_weight":22,"slug":128},"World Cup","world-cup",{"id":130,"doc_module":4,"doc_module_name":46,"category_name":131,"show_sort_weight":130,"slug":132},10,"Lifestyle","lifestyle",{"id":134,"doc_module":4,"doc_module_name":46,"category_name":135,"show_sort_weight":106,"slug":136},19,"General","general"]