mlsanityData to 14 Oct 2026

AutoMem: Automated Learning of Memory as a Cognitive Skill

arXiv 2607.01224 ↗First posted 1 Jul 2026cs.AI cs.CL cs.MAcited by 5

Not on the map. Why?

Abstract (arXiv)

Memory 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 ~2x-4x, 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.

Key facts. Cited by 5 papers in the map (data through 14 Oct 2026). Builds on 37 of its 40 references that are in the map. First posted on arXiv 1 Jul 2026.

Builds on

37 of 40 references.

References · 37

  1. 2021LoRA: Low-Rank Adaptation of Large Language Modelscited by 14,380
  2. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,459
  3. 2020Retrieval-Augmented Generation for Knowledge-Intensive NLP Taskscited by 9,297
  4. 2022ReAct: Synergizing Reasoning and Acting in Language Modelscited by 8,695
  5. 2023Reflexion: Language Agents with Verbal Reinforcement Learningcited by 4,276
  6. 2023Generative Agents: Interactive Simulacra of Human Behaviorcited by 3,568
  7. 2023Voyager: An Open-Ended Embodied Agent with Large Language Modelscited by 2,191
  8. 2023MemGPT: Towards LLMs as Operating Systemscited by 1,389
  9. 2022Inner Monologue: Embodied Reasoning through Planning with Language Modelscited by 1,081
  10. 2025A-MEM: Agentic Memory for LLM Agentscited by 1,063
  11. 2022Large Language Models Are Human-Level Prompt Engineerscited by 1,045
  12. 2023ExpeL: LLM Agents Are Experiential Learnerscited by 876
25 more
  1. 2023DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelinescited by 832
  2. 2024A Survey on the Memory Mechanism of Large Language Model based Agentscited by 634
  3. 2023MemoryBank: Enhancing Large Language Models with Long-Term Memorycited by 556
  4. 2023Promptbreeder: Self-Referential Self-Improvement Via Prompt Evolutioncited by 551
  5. 2024AFlow: Automating Agentic Workflow Generationcited by 432
  6. 2023Cognitive Architectures for Language Agentscited by 429
  7. 2024Automated Design of Agentic Systemscited by 407
  8. 2023Describe, Explain, Plan and Select: Interactive Planning with Large Language Models Enables Open-World Multi-Task Agentscited by 398
  9. 2025Memory-R1: Enhancing Large Language Model Agents to Manage and Utilize Memories via Reinforcement Learningcited by 237
  10. 2021Benchmarking the Spectrum of Agent Capabilitiescited by 200
  11. 2020The NetHack Learning Environmentcited by 198
  12. 2024BALROG: Benchmarking Agentic LLM and VLM Reasoning On Gamescited by 146
  13. 2026MemSkill: Learning and Evolving Memory Skills for Self-Evolving Agentscited by 142
  14. 2026MemRL: Self-Evolving Agents via Runtime Reinforcement Learning on Episodic Memorycited by 129
  15. 2021MiniHack the Planet: A Sandbox for Open-Ended Reinforcement Learning Researchcited by 97
  16. 2025MemEvolve: Meta-Evolution of Agent Memory Systemscited by 95
  17. 2026Agentic Memory: Learning Unified Long-Term and Short-Term Memory Management for Large Language Model Agentscited by 78
  18. 2025Memory as Action: Autonomous Context Curation for Long-Horizon Agentic Taskscited by 66
  19. 2024MemLLM: Finetuning LLMs to Use An Explicit Read-Write Memorycited by 35
  20. 2023Learning to Reason and Memorize with Self-Notescited by 30
  21. 2025MemSearcher: Training LLMs to Reason, Search and Manage Memory via End-to-End Reinforcement Learningcited by 27
  22. 2024Playing NetHack with LLMs: Potential & Limitations as Zero-Shot Agentscited by 17
  23. 2026EvolveMem:Self-Evolving Memory Architecture via AutoResearch for LLM Agentscited by 15
  24. 2026MetaMem: Evolving Meta-Memory for Knowledge Utilization through Self-Reflective Symbolic Optimizationcited by 7
  25. 2026MeMo: Memory as a Modelcited by 4

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 5.

5 papers.

  1. 2026RRSI: Regularized Recursive Self-Improvement of Agent Harnessescited by 13
  2. 2026Context Language Modelscited by 1
  3. 2026Why Git Is the Memory Solution for the Agentic Development Lifecyclecited by 0
  4. 2026Knowing the Store: What a Memory Backend Must Write Down Before an Agent Can Read Itcited by 0
  5. 2026Harness Evolution Hits a Ceiling: When Weight Training Should Begincited by 0