为大模型代理设计可编程记忆治理层，解决记忆过时与冲突问题
MemArchitect: A Policy Driven Memory Governance Layer
- 通过规则驱动机制分离记忆生命周期管理与模型权重
- 治理后记忆在代理任务中表现更优，减少过时信息干扰
- 适合需要安全可靠自主决策的AI系统开发者
持久化大型语言模型（LLM）代理暴露了内存管理中的关键治理空白。标准检索增强生成（RAG）框架将记忆视为被动存储，缺乏解决矛盾、保障隐私或防止过时信息（“僵尸记忆”）污染上下文窗口的机制。我们提出MemArchitect，一个将记忆生命周期管理与模型权重解耦的治理层。该层执行显式的规则化策略，包括记忆衰减、冲突消解和隐私控制。实验表明，在代理场景中，受控记忆的表现持续优于无管理记忆，凸显了结构化记忆治理对可靠且安全自主系统的重要性。
原文摘要 · Abstract (English)
Persistent Large Language Model (LLM) agents expose a critical governance gap in memory management. Standard Retrieval-Augmented Generation (RAG) frameworks treat memory as passive storage, lacking mechanisms to resolve contradictions, enforce privacy, or prevent outdated information ("zombie memories") from contaminating the context window. We introduce MemArchitect, a governance layer that decouples memory lifecycle management from model weights. MemArchitect enforces explicit, rule-based policies, including memory decay, conflict resolution, and privacy controls. We demonstrate that governed memory consistently outperforms unmanaged memory in agentic settings, highlighting the necessity of structured memory governance for reliable and safe autonomous systems.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。