适应性记忆接纳控制用于大语言模型代理

LLM-based agents increasingly rely on long-term memory to support multi-session reasoning and interaction, yet current systems provide limited control over retained information. In practice, agents either accumulate large volumes of conversational content, including hallucinated or obsolete facts, or depend on opaque, fully LLM-driven memory policies that are costly and difficult to audit. Consequently, memory admission remains a poorly specified and weakly controlled component in agent architectures. To address this issue, Adaptive Memory Admission Control (A-MAC) is proposed, framing memory admission as a structured decision problem. A-MAC decomposes memory value into five complementary and interpretable factors, aiming to enhance the efficiency and transparency of memory management.