Mandol: An Agglomerative Agent Memory System for Long-Term Conversations
作者: Yuhan Zhang, Zhiyuan Guo, Ziheng Zeng, Wei Wang, Wentao Wu, Lijie Xu
分类: cs.DB, cs.AI, cs.CL, cs.IR
发布日期: 2026-06-29
备注: 10 pages, 3 figures
💡 一句话要点
提出Mandol以解决长时间对话中的记忆碎片化问题
🎯 匹配领域: 支柱三:空间感知与语义 (Perception & Semantics)
关键词: 长时间对话 记忆系统 聚合架构 信息检索 语义图 量化查询 智能代理
📋 核心要点
- 现有的代理记忆系统存在记忆信息碎片化和高I/O延迟的问题,影响了长时间对话的准确性和效率。
- Mandol通过聚合记忆表示和存储,构建了一个统一的记忆原生架构,提升了信息检索的效率和准确性。
- 在LoCoMo和LongMemEval基准测试中,Mandol实现了最佳准确性,并显著提高了检索和插入速度。
📝 摘要(中文)
长时间对话代理需要记住和查询跨会话的多类型信息及其复杂关联。现有的代理记忆系统依赖异构向量和图数据库,导致记忆信息碎片化并造成高跨数据库I/O延迟。为此,本文提出Mandol,一个聚合记忆系统,将碎片化的记忆表示和存储整合为统一的记忆原生架构。其核心组件包括分层记忆模型、聚合语义数据结构和量化查询机制。实验结果表明,Mandol在LoCoMo和LongMemEval基准测试中表现出最佳的整体准确性,并在10 QPS并发负载下实现了5.4倍的检索加速和4.8倍的插入加速,同时在消费级硬件上保持低延迟。
🔬 方法详解
问题定义:本文旨在解决长时间对话中代理记忆系统的碎片化问题,现有方法依赖异构数据库,导致高I/O延迟和信息丢失。
核心思路:Mandol通过构建一个统一的记忆架构,将碎片化的记忆整合为层次化的结构,提升信息的检索和存储效率。
技术框架:Mandol的整体架构包括三个主要模块:分层记忆模型、聚合语义数据结构和量化查询机制,分别负责记忆的组织、存储和检索。
关键创新:Mandol的核心创新在于其聚合语义数据结构,能够融合多种数据结构,消除跨数据库I/O,提高检索效率。
关键设计:在设计中,采用了层次化的记忆表示,结合了语义图和语义映射,确保信息的高效存储与检索,同时引入了量化查询机制以优化检索过程。
🖼️ 关键图片
📊 实验亮点
实验结果显示,Mandol在LoCoMo和LongMemEval基准测试中实现了最佳的整体准确性,并在10 QPS并发负载下,检索速度提升了5.4倍,插入速度提升了4.8倍,且在消费级硬件上保持低延迟。
🎯 应用场景
Mandol的研究成果可广泛应用于智能客服、虚拟助手和社交机器人等领域,提升其在长时间对话中的记忆和响应能力,具有重要的实际价值和未来影响。
📄 摘要(原文)
Long-term conversational agents need to remember and query cross-session, multi-typed information with complex correlations. Existing agent memory systems rely on heterogeneous vector and graph databases, which fragment memory information and cause high cross-database I/O latency. For retrieval, common RAG-style methods tend to introduce noise, miss correlated clues, and lack token budget control, degrading LLM accuracy and efficiency. We propose Mandol, an agglomerative memory system that consolidates fragmented memory representations and storage into a unified memory-native architecture. Its core components include: (1) a hierarchical memory model that organizes memory into a basic layer representing raw memory information and a high-level abstract layer that agglomerates basic memories into traceable abstract memories, both uniformly represented as structured semantic graphs; (2) an agglomerative semantic data structure combining SemanticMap and SemanticGraph, which natively fuses key-value, vector, and graph structures and provides unified hybrid retrieval operators to eliminate cross-database I/O; and (3) a quantitative query mechanism with query-adaptive routing, quantitative denoising and conflict resolution, and token-constrained context generation, all without involving LLMs during retrieval. Experiments on two widely used long-term conversation benchmarks, LoCoMo and LongMemEval, show that Mandol achieves the best overall accuracy among representative agent memory systems. For performance comparison, Mandol also obtains a 5.4x retrieval speedup and a 4.8x insertion speedup under 10 QPS concurrent load, while still maintaining low latency on consumer-grade hardware.