笔记
关于 AI 助手记忆的笔记:它如何运作、存在哪里,以及那些取舍真正的代价。
- 2026年10月30日 · 6 min read
Claude memory: what Claude remembers from one chat to the next
Claude has a built-in memory and can search your past chats. What it keeps, on which plan, where to control it, and what it doesn't do.
- 2026年10月30日 · 7 min read
Claude Code memory: keep context between sessions (CLAUDE.md and MCP)
CLAUDE.md and auto memory keep a repository's context on your machine. To find it elsewhere, and in your other AIs, you need an MCP server.
- 2026年10月30日 · 5 min read
Cursor: add a shared memory with a remote MCP server
One JSON block in mcp.json, one OAuth sign-in, one rule: Cursor's agent finds your projects' context and what you decided with your other AIs.
- 2026年10月30日 · 5 min read
VS Code and Copilot: add a persistent memory with an MCP server
One command or a .vscode/mcp.json file, one sign-in in the browser: Copilot finds the context you built with Claude, ChatGPT or Cursor.
- 2026年10月30日 · 4 min read
The 2-minute test: tell Claude, find it in ChatGPT
One fact told to Claude, a new chat in ChatGPT, and no repetition: if ChatGPT gets it right, your AIs really share the same memory.
- 2026年10月30日 · 5 min read
What to store in your AI memory, and what never to put there
An AI memory is only as good as what goes in. Rule of thumb: keep what saves you from repeating yourself in three weeks, and nothing you wouldn't put in a shared notebook.
- 2026年10月30日 · 5 min read
Import your ChatGPT memory into Pryzm (and share it with Claude)
ChatGPT already knows you. One message gets it to reread its memory and pour it into Pryzm, where all your connected AIs can use it.
- 2026年10月30日 · 7 min read
Local vs hosted memory MCP server: which one to choose?
A local memory server keeps everything on your computer and only works there. A hosted one is an address every app can reach. What each gives you, what each costs you, and the options that exist.
- 2026年10月30日 · 6 min read
Perplexity memory: what it remembers and how to extend it
Perplexity can remember your preferences and past searches. Where to control it, what its memory doesn't do, and how to give it a memory your other AIs share.
- 2026年10月6日 · 9 min read
Hybrid search in Postgres: BM25, pgvector, RRF and a reranker
No separate search engine. One Postgres database, two rankings, one fusion and a reranker — with the parameters we actually run.
- 2026年10月3日 · 6 min read
Encrypted AI memory: what the word actually covers
Every AI memory service says it is encrypted. The word covers four different guarantees, and only some of them protect you from the provider itself.
- 2026年10月2日 · 6 min read
Connect ChatGPT to an external memory: the MCP server, step by step
A custom MCP server added as a plugin on chatgpt.com, one sign-in, then @ in the chat.
- 2026年10月2日 · 6 min read
Claude MCP: add a custom connector (free plan included)
One URL in Customize → Connectors, one sign-in, and Claude can use an external tool. Even without a subscription.
- 2026年10月1日 · 6 min read
Solo founder: switch from Claude to ChatGPT without re-explaining your project
When you build alone, your AIs are your team. They just need to remember the same things.
- 2026年10月1日 · 6 min read
Freelancers: keep each client's preferences across missions and AIs
Five clients, three AIs, and the same question every time: which one didn't want informal copy again?
- 2026年10月1日 · 6 min read
Team AI memory: no more “what did we decide again?”
In a small team, decisions are made everywhere and written down nowhere. And the AIs know nothing about them.
- 2026年9月23日 · 阅读时间 6 分钟
如何让 ChatGPT 记住你们的对话
ChatGPT 会记住你。前提是功能已开启、你清楚它记了什么,并且你不需要把这些记忆带到别处去。
- 2026年9月23日 · 阅读约 8 分钟
2026 年最佳记忆 MCP 服务器:10 款对比,本地还是托管
没有绝对最好的记忆服务器。有适合让记忆留在你电脑上的,有适合自托管的,有适合构建智能体的,也有不需要你自己运维的托管记忆。
- 2026年9月23日 · 阅读时间 6 分钟
如何把 ChatGPT 的记忆迁移到 Claude
换助手很容易。把你的上下文一起带走,才是没人写文档的那部分。三种方法,从粗糙到持久。
- 2026年9月22日 · 阅读时间 6 分钟
如何让你的各个 AI 助手共用同一份记忆
现在每个助手都有自己的记忆,却没有一个愿意共享。远程 MCP 服务器就是解决这件事的那块拼图。
- 2026年9月22日 · 阅读时间 7 分钟
你的 AI 记忆究竟存在哪里?
记忆层就是一个装着你对助手说过的一切的数据库。这让它成为你技术栈里最敏感、也最缺乏文档的东西。
- 2026年9月22日 · 阅读时间 6 分钟
MCP 服务器到底是什么?
把营销话术剥掉,MCP 是一个小而无聊的协议。这正是它最好的地方。
- 2026年9月22日 · 阅读时间 8 分钟
持久化 AI 记忆究竟是怎么工作的
「AI 记得我」这句话承担了太多含义。模型里什么都没变。变的是它前面那个东西,变得很会取东西了。