从上下文到记忆：智能体需要真正的记忆层

核心内容

Anders Swanson 把智能体记忆视为系统设计问题，讨论为何上下文窗口和临时笔记不能替代长期记忆。发布方介绍涉及记忆的形成、召回、演化与评估。

演讲还覆盖记忆类型、混合检索、写入前脱敏、审核历史，以及过时、错误归属、投毒和冲突记忆的处理。

Anders Swanson treats agent memory as a systems problem, examining why context windows and scratch notes do not replace persistent memory. The publisher describes formation, recall, evolution and evaluation.

Other topics include memory types, hybrid retrieval, redaction before writing, audit history and handling stale, mis-scoped, poisoned or conflicting memories.

介绍依据 · 发布方说明 · 完整内容待核读

编辑导览，依据原始来源 ↗，整理于 2026-10-10。中英文均由本站概括整理。内容介绍与下方原话、主张的逐条核对分开标注。

上下文窗口与临时笔记无法替代持久记忆系统：真正的智能体记忆层需要具备严格的写入前脱敏与范围隔离机制，并通过融合向量相似度、全文检索与实体图谱的混合检索实现稳健的上下文重建。

原话

2逐字核对过；核对了说话人和位置

redaction is very important as the first step because before I do embeddings, before I do enrichment, I want to strip out PII. I don't want secrets or irrelevant weird information slipping into the memory layer. That's a recipe for disaster. So, we do redaction first.

自动字幕 · 已核对Anders Swanson▶ 00:08:10

And this this form of retra retrieval can fuse different search methods. So, I can take vector similarity search, I can take a full-text keyword search, I can take relational entities, I can do graph traversal, I can take other feedback signals like importance uh uh and usefulness of the memory to yield more comprehensive results than one specific query branch alone.

自动字幕 · 已核对Anders Swanson▶ 00:10:20