Build a second brain your AI can actually read

The problem this solves: you can't connect the dots looking forward — only looking back. But you can only look back if you wrote it down. Most people can't connect their dots because they don't have the dots. And an AI knows you exactly as well as you've written yourself down.

1Install Obsidian

Free, and your notes stay on your machine as ordinary Markdown files — no lock-in, no export needed, readable in ten years. Download at obsidian.md. Create one vault. Name it anything.

MyBrain/
├─ CLAUDE.md      ← the rules
├─ MEMORY.md      ← running state
├─ index.md       ← the map
├─ 10-Projects/   ← has an end date
├─ 20-Areas/      ← ongoing, no end date
├─ 30-Resources/  ← reference
└─ 40-Archive/    ← done, never deleted

---
id: short-slug
type: concept | project | decision
tags: [topic]
created: 2026-08-01
updated: 2026-08-01
---

# Title
One sentence: the single idea.

Body. Link things with [[wikilinks]].

4Write the rules file — the part everyone skips

A second brain without rules becomes a junk drawer in about three weeks. Put these four in CLAUDE.md at the root. They are the whole system:

• Frontmatter always. Every note carries the header above. No exceptions.
• Wikilinks always. Every note links to at least one neighbour. An orphan note is a lost note.
• Never strip context. Only add or update — never delete. Bump updated: every edit.
• Mark what expires. Every fact is timeless (write it plainly), dated (tag it (as of 2026-08)), or a pointer (link the live source instead of copying a number). This one is why the vault is still true in five years.

5Connect the AI

Install VS Code, then Claude Code. Open your vault folder as the workspace. The AI now reads the same plain files Obsidian does — one folder, two front doors. Paste this as your first message:

You are the agent for my second brain in this folder.

Before answering anything, read CLAUDE.md, then MEMORY.md, then index.md.

Rules you follow on every edit:
1. Every note starts with YAML frontmatter (id, type, tags, created, updated)
   and contains at least one [[wikilink]] to a related note.
2. Never delete existing content. Only add or update. Bump `updated:` every time.
3. Mark each fact as timeless, dated "(as of YYYY-MM)", or a pointer to a live source.
4. Before you state a fact about me, find the file it came from. If it isn't in
   the vault, say so instead of guessing.

At the end of every session, append what changed to MEMORY.md — decisions made,
questions still open, what I should pick up next time.

Now: [your question]

Rule 4 is the important one. It's the difference between an assistant that knows you and one that invents you.

6Index it, or it doesn't work

Two files carry the whole system, and they're the ones people forget:

• index.md — one line per project or area, with a wikilink and a short description. This is the map the AI reads first, so it knows what exists before it starts guessing. Refresh it whenever you add something top-level.
• MEMORY.md — current state, decisions made, open questions. Updated at the end of every session. This is what makes tomorrow start where today ended instead of from zero.

Then run one habit: at the end of each working day, write what you decided and what you got wrong. Two lines is enough. In six months those two lines a day are the only reason you'll be able to look back and see the dots at all.