mlsanityData to 7 Oct 2026

Compiled Memory: Not More Information, but More Precise Instructions for Language Agents

arXiv 2603.15666 ↗First posted 12 Mar 2026cs.AIcited by 0

On the map: LLM agents and tool use → Agent memory systems

Abstract (arXiv)

Existing memory systems for language agents address memory management: how to retrieve and page more information within a context budget. We address a complementary problem -- memory utility: what experience is worth keeping, and how it should change agent behavior. We present Atlas, a memory kernel that compiles accumulated task experience into an agent's instruction structure -- without fine-tuning, RAG, or human intervention. Memory is distillation, not storage; delivery is instruction rewriting, not context injection. Facts extracted from agent failures and successes are verified through a three-step promotion gate and delivered by rewriting the agent's system prompt with learned sub-bullets. On CUAD contract analysis, the evolved prompt improves GPT-4o token-level F1 by +8.7pp and precision by +12.5pp. On HotpotQA multi-hop QA, joint F1 improves +3.16pp. An ablation isolates the mechanism's defining property -- the training signal constraint: the evolved prompt learns exactly what it is taught, and nothing more. Applied to Claude Sonnet~4.5 using the same evolved prompt -- compiled from GPT-4o errors, unchanged -- joint F1 improves +2.31pp, with gains concentrating where Claude's stronger baseline leaves the most room -- confirming that the compiled knowledge is task-shaped, not model-shaped.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Agent memory systems, one of 16 directions in LLM agents and tool use. Builds on 11 of its 14 references that are in the map. First posted on arXiv 12 Mar 2026.

Builds on

11 of 14 references.

Closest ancestors · 1

Same direction, newest first

  1. 2025How Memory Management Impacts LLM Agents: An Empirical Study of Experience-Following Behavior21 May

Other references · 10

  1. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,393
  2. 2023Reflexion: Language Agents with Verbal Reinforcement Learningcited by 4,064
  3. 2018HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answeringcited by 3,686
  4. 2022Large Language Models Are Human-Level Prompt Engineerscited by 1,020
  5. 2023Large Language Models as Optimizerscited by 830
  6. 2023ExpeL: LLM Agents Are Experiential Learnerscited by 814
  7. 2023DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelinescited by 803
  8. 2021CUAD: An Expert-Annotated NLP Dataset for Legal Contract Reviewcited by 232
  9. 2024Better Call GPT, Comparing Large Language Models Against Lawyerscited by 25
  10. 2025ContractEval: Benchmarking LLMs for Clause-Level Legal Risk Identification in Commercial Contractscited by 9

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026MemSkill: Learning and Evolving Memory Skills for Self-Evolving Agents
  2. 2026M^star: Every Task Deserves Its Own Memory Harness
  3. 2026Agent Memory Distillation: Empowering Small LLM Agents with Hierarchical Teacher Memory
  4. 2026What to Keep, What to Forget: A Rate--Distortion View of Memory Compaction in LLMs and Agents
  5. 2026Memory as a Controlled Process: Learned Adaptive Memory Management for LLM Agents
  6. 2026Muscle Memory for Agents: Compile not Merely Retrieve
  7. 2026EvoMemBench: Benchmarking Agent Memory from a Self-Evolving Perspective
  8. 2026Auto-Dreamer: Learning Offline Memory Consolidation for Language Agents