Adnan KhanAdnanProduct & AI
San Francisco Bay Area
Get in touch

Source note3 min read

Memory as a Controlled Process: Learned Adaptive Memory Management for LLM Agents

Source

arXiv

arxiv.org/abs/2607.13591v1

Read the source

Source

  • Source name: arXiv
  • Source kind: paper
  • Authors / organization: Eric Hanchen Jiang, Zhi Zhang, Yuchen Wu, Levina Li, Dong Liu, Xiao Liang, Rui Sun, Yubei Li, Edward Sun, Haozheng Luo, Zhaolu Kang, Aylin Caliskan, Kai-Wei Chang, Ying Nian Wu
  • Venue: arXiv: cs.CL
  • Published: 2026-07-15
  • Collected: 2026-07-16
  • URL: http://arxiv.org/abs/2607.13591v1
  • PDF: https://arxiv.org/pdf/2607.13591v1
  • DOI: UNKNOWN

Plain-Language Brief

The paper presents a new framework called Memory as a Controlled Process (MemCon) for large language model (LLM) agents. MemCon aims to address the limitation of existing memory systems that rely on fixed, hand-designed heuristics by learning an adaptive policy for deciding when, what, and how much to retrieve from memory, inject a distilled plan, and consolidate or forget. The framework is backend-agnostic, lightweight, and improves task success by up to 15.2 points while reducing token consumption.

What the Document Says

  • The paper introduces MemCon, a framework that models memory operations as a Markov Decision Process.
  • MemCon learns an online policy for deciding when, what, and how much to retrieve from memory.
  • It also learns when to inject a distilled plan and when to consolidate or forget.
  • The framework is designed to work with any existing memory implementation.
  • It learns from task-by-task binary feedback with no pretraining and no additional LLM calls.
  • The authors use a lightweight tabular contextual bandit with UCB exploration.
  • MemCon outperforms multiple memory baselines across 6 benchmarks, 3 agent frameworks, and 3 LLM backbones.
  • The improvement in task success is up to 15.2 points.
  • Token consumption is reduced by 5-20%.

Key Concepts Mentioned

  • Large Language Model (LLM) agents
  • Memory management
  • Markov Decision Process
  • Adaptive policy
  • Distilled plan
  • Consolidation
  • Forgetting
  • Backend-agnostic
  • Task success
  • Token consumption

Claims Made by the Document

  • Existing memory systems for LLM agents rely on fixed, hand-designed heuristics.
  • This static view of memory is a core bottleneck for agentic learning because optimal memory behavior is context-dependent.
  • MemCon outperforms multiple memory baselines across various benchmarks.

Evidence Used

  • The text mentions that existing memory systems for LLM agents rely on fixed, hand-designed heuristics.
  • The authors argue that this static view of memory is a core bottleneck for agentic learning.
  • Across 6 benchmarks, 3 agent frameworks, and 3 LLM backbones, MemCon outperforms multiple memory baselines.

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

Large Language Model (LLM) agents increasingly rely on external memory systems to accumulate experience across tasks. Yet nearly all existing approaches, from graph-structured memories to reflective insight stores, access memory through fixed, hand-designed heuristics. We argue that this static view of memory is a core bottleneck for agentic learning because optimal memory behavior is fundamentally context-dependent. The early stages of the tasks, benefit from minimal retrieval because memory is sparse; recurring goal types benefit from plan reuse rather than generic nearest-neighbor lookup; stuck agents benefit from re-retrieval with alternative queries; and across long task streams, the memory store itself must be consolidated and pruned to remain useful. We present Memory as a Controlled Process (MemCon), a framework that models memory operations as a Markov Decision Process and learns an online policy that adaptively decides when, what, and how much to retrieve, when to inject a distilled plan, and when to consolidate or forget. MemCon is backend-agnostic: it wraps any existing memory implementation, learns from task-by-task binary feedback with no pretraining and no additional LLM calls, and uses a lightweight tabular contextual bandit with UCB exploration that converges within tens of tasks. Across 6 benchmarks, 3 agent frameworks, and 3 LLM backbones, MemCon consistently outperforms multiple memory baselines by up to 15.2 points in task success while reducing token consumption by 5--20%.