Article · Method

Cheating: a very simple second brain for all your AI

No vector database, no plugin, no subscription. My second brain is a folder of markdown files with one index, and every AI I use reads the same one. It feels like cheating because it is.

People show me elaborate "second brain" setups. Vector databases, embedding pipelines, a plugin for every tool. Then they show me the AI still not knowing what project they are on. Mine is a folder. Here is the whole architecture.

The shape

One index file, and small memory files.

The index is a single markdown file with one line per memory file: a link and a short hook that says when to read it. "Customer X, the contract terms and who signed" or "How I like my weekly plan formatted". That index is the only thing loaded by default. It is cheap, and it is enough for the AI to know where to look.

Each memory file holds one fact or one topic. A few lines of frontmatter (name, one-line description, type) and then the content. One fact per file, so updating never means rewriting a document.

That is the entire design. It is how Claude Code's own memory works. It is how the Cortex memory inside the Pocket Operator works. I did not invent it, I just refuse to complicate it.

Reading it from everywhere

The reason this beats every plugin is that plain files go anywhere.

  • Claude Code reads the folder as project memory.
  • Codex and ChatGPT projects read it as attached files.
  • The Pocket Operator reads it through /brain and rewrites it when you tell it something new.
  • Any MCP client reads it through three tools: index, read, search. Same files, no copy.

Ask three different AIs the same question about a real project and all three answer from the same brain. Same facts, same names, same tone. Nobody re-explains anything.

Keeping it honest

The folder only works if it stays true. Three habits.

  1. One fact per file. When something changes, update that file. Never add a second file that contradicts the first.
  2. Delete what turns out to be wrong. A wrong memory is worse than no memory, because the AI will use it with confidence.
  3. Keep the index as the only default. If everything loads every time, the brain gets expensive and the AI stops reading. The index says what exists; the AI fetches what it needs.

It is called cheating because it feels like it. The model did not get smarter. It just stopped guessing.

Do this week

  • Create a folder with one index file. Add three memory files for the three things you keep re-explaining to every AI.
  • Attach the folder to whichever AI you use most. Ask it something only the folder knows.
  • When it answers correctly, add the fourth file. Repeat until you stop re-explaining.

Want the same thing done for you?

Join the launch list and you get your personal start link the day it opens.