[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-81801-en":3,"doc-seo-81801-105":29,"detail-sidebar-cat-0-en-105":90},{"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":4,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},81801,137441390410,"Hazel","https://ap-avatar.wpscdn.com/avatar/2000252f4ab5702993?_k=1776741390130283984",8,"Research & Report","AUTOMEM: Automated Learning of Memory as a Cognitive Skill","Memory expertise is a learned capability—deciding what to encode, when to retrieve, and how to organize knowledge—known as metamemory in cognitive science. The work applies this perspective to LLMs by treating memory management as a trainable skill: file-system operations become first-class memory actions alongside task actions, enabling the model to decide how to store, search, and update memory. AUTOMEM automates both memory structure and proficiency via two meta-LLM-driven outer loops, improving performance by ~2×–4× on long-horizon games without altering task-action behavior.","arXiv :2607 .0 1224v 1 [ cs .AI] 1 Jul 2026  \nAUTOMEM: Automated Learning of Memory as a  \nCognitive Skill  \nShengguang Wu Hao Zhu Yuhui Zhang Xiaohan Wang Serena Yeung-Levy  \nStanford University  \n[https://autolearnmem.github.io/](https://autolearnmem.github.io/)  \nAbstract  \nMemory expertise is a learned skill: knowing what to encode, when to retrieve, and how to organize knowledge—a capacity known in cognitive science as metamemory. We bring this perspective to LLMs by treating memory management as a trainable skill. We promote file-system operations to first-class memory actions alongside task actions, letting the model itself decide how to manage its memory. This memory skill improves along two axes: the structure that supports it (prompts, file schemas, action vocabulary), and the proficiency of the model exercising it. Both axes resist manual optimization: episodes in long-horizon tasks run for thousands of steps, and a single memory mistake can hide long before it surfaces, making human review of full trajectories impractical. We introduce AUTOMEM, a framework that automates both axes. In the first loop, a strong LLM reviews complete agent trajectories and iteratively revises the memory structure that shapes how the agent interacts with its memory files. In the second loop, the agent’s own good memory decisions are identified from many episodes and used as training signal to sharpen the model’s memory proficiency directly. Across three procedurally generated long-horizon games (Crafter, MiniHack, and NetHack), optimizing memory alone—without modifying the model’s task-action behavior—improved the base agent’s performance ∼2×–4×, bringing a 32B open-weight model competitive with frontier systems such as Claude Opus 4.5 and Gemini 3.1 Pro Thinking. Our results show that memory management is an independently learnable skill, and a high-leverage objective yielding large gains on long-horizon tasks.  \nFigure 1: Memory skill optimization with Qwen2 .5-32B-Instruct. Starting from a base agent equipped with file-system memory (v0), AUTOMEM progressively improves performance through memory scaffold optimization (v0–v5/v4/v2), followed by memory proficiency training (+train) that yields further gains on top of the optimized scaffold.  \nPreprint.  \n1 Introduction  \nHumans routinely manage information beyond what can be held in mind at any one moment. Cognitive scientists call this capacity metamemory: the learned skill of deciding what is worth remembering, when to retrieve it, and how to organize what is known [Flavell, 1979, Nelson, 1990] . Metamemory develops with practice, and skilled use of external aids—notes, indices, files—is part of how people extend cognition beyond working memory [Clark and Chalmers, 1998] .  \nLLMs face an analogous bottleneck. Their context window plays the role of working memory, i.e., a fixed-size buffer that bounds what the model can attend to at once. Long-horizon tasks routinely exceed this capacity, and external memory has been explored in various forms, including retrieval databases, vector stores, scratchpads, and summary buffers [Lewis et al., 2020, Packer et al., 2023, Park et al., 2023, Xu et al., 2025, Sumers et al., 2023, Zhang et al., 2024b, Zhong et al., 2024] . These approaches typically treat memory as an architectural module: a fixed mechanism designed into the system. We take a different view, one inspired by metamemory: memory management is an active, trainable skill, and the model itself decides what to store, what to look up, and how to structure its records.  \nConcretely, we promote file-system operations (read, write, search, append, create) as first-class memory actions in the model’s action space, on equal footing with the actions it uses to act on the world [Yao et al., 2022] . The same forward pass that picks a task action can also select memory file operations (e.g., \u003C|APPEND| > or \u003C|SEARCH|>) . This minimal design gives the model full control over its external memory whil","cbCaikxy0KDQK7GU","https://ap.wps.com/l/cbCaikxy0KDQK7GU","pdf",2613199,1,16,"English","en",105,"# Abstract\n# Introduction\n## Metamemory and the external memory bottleneck for LLMs\n## File-system operations as first-class actions\n## Two-axis improvement: structure and proficiency\n## AUTOMEM framework with two outer loops","[{\"question\":\"What is AUTOMEM and what problem does it address?\",\"answer\":\"AUTOMEM proposes learning memory management in LLM agents as an independently trainable cognitive skill, addressing the long-horizon failure modes where context limits make it hard to store and retrieve information reliably.\"},{\"question\":\"How does AUTOMEM represent memory inside the model?\",\"answer\":\"It promotes file-system operations such as read, write, search, append, and create to first-class actions in the model’s action space, so the same forward pass can select both task actions and memory file operations.\"},{\"question\":\"How are memory structure and memory proficiency improved in AUTOMEM?\",\"answer\":\"The first outer loop uses a meta-LLM to review full agent trajectories and revise the memory scaffold, including prompts and file schemas; the second outer loop selects good memory decisions across many episodes as training signal to sharpen a dedicated memory model’s proficiency.\"}]",1784176242,40,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":85,"head_meta":87,"extra_data":89,"updated_unix":27},"automem-automated-learning-of-memory-as-a-cognitive-skill","",{"@graph":35,"@context":84},[36,53,67],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/automem-automated-learning-of-memory-as-a-cognitive-skill/81801/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":61,"encodingFormat":60,"isAccessibleForFree":62,"interactionStatistic":63},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-16",true,{"@type":64,"interactionType":65,"userInteractionCount":4},"InteractionCounter",{"@type":66},"ViewAction",{"@type":68,"mainEntity":69},"FAQPage",[70,76,80],{"name":71,"@type":72,"acceptedAnswer":73},"What is AUTOMEM and what problem does it address?","Question",{"text":74,"@type":75},"AUTOMEM proposes learning memory management in LLM agents as an independently trainable cognitive skill, addressing the long-horizon failure modes where context limits make it hard to store and retrieve information reliably.","Answer",{"name":77,"@type":72,"acceptedAnswer":78},"How does AUTOMEM represent memory inside the model?",{"text":79,"@type":75},"It promotes file-system operations such as read, write, search, append, and create to first-class actions in the model’s action space, so the same forward pass can select both task actions and memory file operations.",{"name":81,"@type":72,"acceptedAnswer":82},"How are memory structure and memory proficiency improved in AUTOMEM?",{"text":83,"@type":75},"The first outer loop uses a meta-LLM to review full agent trajectories and revise the memory scaffold, including prompts and file schemas; the second outer loop selects good memory decisions across many episodes as training signal to sharpen a dedicated memory model’s proficiency.","https://schema.org",{"og:url":51,"og:type":86,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":88,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":91},[92,96,100,104,109,114,118,121,126,129,133],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":93,"show_sort_weight":94,"slug":95},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":97,"show_sort_weight":98,"slug":99},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":101,"show_sort_weight":102,"slug":103},"Exam",70,"exam",{"id":105,"doc_module":4,"doc_module_name":45,"category_name":106,"show_sort_weight":107,"slug":108},5,"Comic",60,"comic",{"id":110,"doc_module":4,"doc_module_name":45,"category_name":111,"show_sort_weight":112,"slug":113},6,"Technology",50,"technology",{"id":115,"doc_module":4,"doc_module_name":45,"category_name":116,"show_sort_weight":28,"slug":117},7,"Healthcare","healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":119,"slug":120},30,"research-report",{"id":122,"doc_module":4,"doc_module_name":45,"category_name":123,"show_sort_weight":124,"slug":125},9,"Religion & Spirituality",20,"religion-spirituality",{"id":124,"doc_module":4,"doc_module_name":45,"category_name":127,"show_sort_weight":124,"slug":128},"World Cup","world-cup",{"id":130,"doc_module":4,"doc_module_name":45,"category_name":131,"show_sort_weight":130,"slug":132},10,"Lifestyle","lifestyle",{"id":134,"doc_module":4,"doc_module_name":45,"category_name":135,"show_sort_weight":105,"slug":136},19,"General","general"]