43 lines
1.4 KiB
Markdown
43 lines
1.4 KiB
Markdown
---
|
||
title: "Cumulative Memory"
|
||
type: concept
|
||
tags: [memory, agent, persistence, accumulation]
|
||
sources: [second-brain]
|
||
last_updated: 2026-04-22
|
||
---
|
||
|
||
## Definition
|
||
|
||
累积记忆（Cumulative Memory）是一种 AI Agent 记忆系统特性：Agent 永久保留用户告知的所有内容，随时间推移积累越来越丰富的个人上下文，使 Agent 变得越来越个性化、越来越有价值。与传统会话式 AI 的"每次会话从零开始"形成鲜明对比。
|
||
|
||
## Aliases
|
||
|
||
- Cumulative Memory
|
||
- 累积记忆
|
||
- Persistent Memory
|
||
- 永久记忆
|
||
- Long-term Memory
|
||
- 长期记忆
|
||
|
||
## Mechanism
|
||
|
||
- 用户通过对话/短信告知 AI 的任何信息（想法、链接、书目、餐厅推荐）都会被永久存储
|
||
- 每次新对话都基于历史上下文展开
|
||
- 随时间推移，Agent 对用户的了解从"陌生人"进化到"贴身助理"
|
||
|
||
## Key Insight
|
||
|
||
> "OpenClaw's memory system is cumulative — it remembers *everything* you've ever told it, making it more powerful over time."
|
||
|
||
系统价值不是线性增长，而是**复利增长**：记忆越多 → 理解越深 → 建议越精准 → 用户越依赖 → 记忆越多。
|
||
|
||
## In System
|
||
|
||
- [[Second Brain]] 的核心价值主张
|
||
- [[OpenClaw]] Workspace/MEMORY.md 文件承载累积记忆
|
||
|
||
## Relationships
|
||
|
||
- [[Zero-Friction Capture]] — 零摩擦捕获是积累大量记忆的前提（摩擦越低，记录越多）
|
||
- [[Agent Memory]] — 技术层面的 Agent 记忆实现
|