MemHarness: Memory Is Reconstructed, Not Replayed

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Rong Wu, Daocheng Fu, Licheng Wen, Xuemeng Yang, Shu Zou, Jianbiao Mei, Yuxin Wang, Hairong Zhang, Yu Yang, Tao Hu, Cong Zhang, Botian Shi, Pinlong Cai
• Venue: arXiv: cs.AI
• Published: 2026-07-30
• Collected: 2026-07-31
• URL: http://arxiv.org/abs/2607.28272v1
• PDF: https://arxiv.org/pdf/2607.28272v1
• DOI: UNKNOWN

Plain-Language Brief

Researchers propose MemHarness, a framework for enhancing large language models that allows them to adapt and reconstruct past experiences based on the current context, rather than simply replaying them verbatim as most existing memory-augmented agents do. This approach aims to improve the robustness of the agent in out-of-distribution scenarios and enhance its intrinsic reasoning capabilities.

What the Document Says

• The paper discusses MemHarness, a framework for large language models that allows them to reconstruct past experiences based on the current context.
• MemHarness equips LLM agents with a unified policy model that critiques and reorganizes retrieved memories at each decision step.
• Experiments on ALFWorld and WebShop show that MemHarness outperforms pure RL and static memory-augmented baselines, particularly in out-of-distribution scenarios.
• The reconstruction objective of MemHarness not only prevents negative transfer but also improves the agent's intrinsic reasoning capabilities.

Key Concepts Mentioned

• MemHarness
• large language models (LLM)
• reconstruct past experiences
• unified policy model
• out-of-distribution scenarios
• negative transfer
• intrinsic reasoning capabilities

Claims Made by the Document

• MemHarness outperforms pure RL and static memory-augmented baselines.
• The reconstruction objective of MemHarness improves the agent's intrinsic reasoning capabilities.

Evidence Used

• Experiments on ALFWorld and WebShop show that MemHarness substantially outperforms...
• Our analyses reveal that this reconstruction objective not only prevents negative transfer but also serves as latent guidance during training...

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Retrieving past experiences has become a common strategy to enhance large language model agents. However, most existing memory-augmented agents treat retrieved experiences as static records to be replayed verbatim, injecting them into the context regardless of whether they align with the agent's current situation. This ``replay'' paradigm ignores the gap between the abstract, general nature of stored experience and the concrete, ever-changing states encountered at decision time, frequently causing negative transfer. In contrast, humans rarely recall past experiences verbatim; instead, they reorganize and adapt retrieved memories to fit the present context. Inspired by this, we propose MemHarness, a framework that equips LLM agents to actively harness and reconstruct past experiences based on the present context. At each decision step, a unified policy model critiques and reconstructs the retrieved experience conditioned on the current state, producing context-grounded guidance before acting. This reconstructive ability emerges naturally through end-to-end training with GRPO. Experiments on ALFWorld and WebShop show that MemHarness substantially outperforms pure RL and static memory-augmented baselines, demonstrating strong robustness in out-of-distribution (OOD) scenarios. Furthermore, our analyses reveal that this reconstruction objective not only prevents negative transfer but also serves as latent guidance during training, fundamentally improving the agent's intrinsic reasoning capabilities.