📄 中文摘要
构建了一个本地 AI 代理（Androi），旨在跨会话记住用户信息。当用户提到“我的名字是 Namhyun”或“我的爱好是远足”时，代理需要在未来的会话中回忆并使用这些信息。初始实现较为简单，用户消息直接提取关键信息并存储在长期记忆（LTM）中。然而，在53个端到端测试中，有8个测试失败，显示出记忆系统存在问题。为了解决这一问题，研究者们模拟了人类记忆巩固过程，构建了一个强大的记忆管道，最终在测试套件中取得了100%的通过率。
为本地 AI 代理设计三层记忆系统 — STM / MTM / LTM
出处: Designing a 3-Tier Memory System for a Local AI Agent — STM / MTM / LTM
发布: 2026年2月24日
📄 English Summary
Designing a 3-Tier Memory System for a Local AI Agent — STM / MTM / LTM
A local AI agent (Androi) was developed to retain user information across conversations. When a user states facts such as 'My name is Namhyun' or 'My hobby is hiking,' the agent must recall and utilize this information in future sessions. The initial implementation was straightforward, where user messages were directly processed to extract key-value pairs and stored in long-term memory (LTM). However, out of 53 end-to-end tests, 8 tests failed, indicating issues within the memory system. To address these challenges, human memory consolidation was modeled to create a robust memory pipeline, achieving a perfect score on the test suite.