微软发布 Memora：平衡抽象与具体的长期记忆系统

AI 摘要

微软研究院发布了 Memora，一种面向长期任务 AI 代理的可扩展记忆系统，通过解耦存储内容与检索方式，平衡抽象与具体性，显著提升代理效率。Memora 在 LoCoMo 和 LongMemEval 基准上达到最先进性能，相比全上下文推理减少高达 98% 的 token 使用。该论文发表于 ICML 2026，代码已在 GitHub 开源。

正文节选

At a glance - Today’s AI agents don’t remember past interactions. They must repeatedly be fed relevant information or retrieve it from external sources, which becomes less efficient as they handle longer and more complex tasks. To scale agent capabilities, we need a more efficient way to retain and access information over time. - Memora is a scalable memory system that dramatically increases agent productivity on long-horizon tasks by decoupling what is stored (rich memory content) from how it’s

发布时间：2026-06-30 05:14

抓取时间：2026-08-02 00:27

来源机构：Microsoft Research