📄 中文摘要
AriadneMem是一种结构化的记忆系统，旨在解决长期对话中存在的两个主要挑战：一是“断开的证据”，即多跳答案需要链接分散在时间上的事实；二是“状态更新”，即不断变化的信息（如日程变更）与旧的静态日志之间产生冲突。该系统通过解耦的两阶段管道来应对这些问题。在离线构建阶段，AriadneMem采用熵感知门控技术来过滤噪声和低信息量的消息，然后进行冲突感知的粗化处理，以合并相关信息，从而提高记忆的准确性和一致性。
AriadneMem：为长期记忆中的大型语言模型代理穿越迷宫
出处: AriadneMem: Threading the Maze of Lifelong Memory for LLM Agents
发布: 2026年3月5日
📄 English Summary
AriadneMem: Threading the Maze of Lifelong Memory for LLM Agents
AriadneMem is a structured memory system designed to address two primary challenges in long-term dialogue: (i) disconnected evidence, where multi-hop answers necessitate linking facts distributed over time, and (ii) state updates, where evolving information (e.g., schedule changes) conflicts with older static logs. The system employs a decoupled two-phase pipeline to tackle these issues. In the offline construction phase, AriadneMem utilizes entropy-aware gating to filter out noise and low-information messages before LLM extraction, followed by conflict-aware coarsening to merge relevant information, thereby enhancing the accuracy and consistency of memory.