Setup Guide

Build a second brain from plain text files

By Dan Donovan, @askdandonovan

About 15 minutes. You'll have an AI second brain that lives as plain files you own. Claude files them, links them, and answers from them. No database, and no new $20-a-month app.

4 Steps
~15 min Setup time
4 Reusable prompts
$0 New subscription

What this actually is: an AI second brain is just a folder of plain text notes on your own computer that Claude reads, files, links, and answers from. No cloud database. No hosted app. Your files, your machine.

What you'll have after this
A ready structure
One folder: a raw drawer, your notes, and an index Claude keeps updated
Auto-filing
Drop in an article and Claude writes the note and links it to the rest
Answers from your notes
Ask a question, get an answer cited from your own files, not the internet
A weekly tidy
Claude flags contradictions, stale notes, and anything left unlinked
Yours to keep
Plain markdown you can move, back up, or open in any app, forever
No new subscription
Replaces the paid AI note apps and runs on the Claude you already have
Before you start: you'll need Claude Code, which comes with a Claude Pro or Max plan (from $20/month, see claude.com/pricing). If you already pay for Claude, this adds nothing. That's the whole point. It's a folder, not another app. Obsidian is optional and free. It just makes the links between notes clickable.
1

Install Claude Code

Claude Code runs in your terminal and can read and write files on your computer. It comes with any paid Claude plan.

On Mac or Linux, open Terminal and paste this. It needs no Node.js and nothing else installed first:

Terminal 
curl -fsSL https://claude.ai/install.sh | bash

Once installed, start it to log in:

Terminal 
claude

Sign in with your Anthropic account the first time and you're set.

Windows? Open PowerShell and run irm https://claude.ai/install.ps1 | iex, or download the desktop app from claude.ai/download for a click-through setup.
2

Make your second-brain folder

One folder with three parts: a raw drawer to dump things into, a notes folder for the filed notes, and an index file that lists everything.

Terminal 
mkdir -p brain/raw brain/notes && touch brain/index.md

That gives you:

📁
brain/raw/
Drop articles, PDFs, notes, and voice-note transcripts here
📁
brain/notes/
Your actual second brain. One file per topic, linked together
📄
brain/index.md
The one file that lists everything. Claude keeps it updated
Prefer clicking? Just make a folder called brain with two folders inside (raw and notes) and an empty index.md file. Same thing.
3

Start Claude in that folder

Point Claude Code at your new folder so it can see your files.

Terminal 
cd brain && claude
4

File your first notes

Drop something into the raw folder. An article you saved, messy meeting notes, a voice-note transcript. Then paste this into Claude and watch it file and link everything.

Claude · ADD 
Read every file in raw/. For each one, pull out the key ideas and write or update a note in notes/, one file per topic. Link related notes to each other. Add anything new to index.md. When a raw file is fully filed, tell me and I will delete it. Keep everything in plain markdown. Do not invent facts.
That's the loop. Every time you have something worth keeping, drop it in raw and run ADD. Claude asks before it deletes a raw file, so you always stay in control.
▶

What to try first

Once you've filed a few things, these are the prompts that make it feel like a real second brain. Copy, paste, and swap in your own topic.

ASK · answer from your own notes 
Using only the files in notes/ and index.md, answer this: [your question]. Quote the note you got each part from. If my notes do not cover it, say so plainly instead of guessing.
DRAFT · write from what you know 
Using only my notes/, write a short brief on everything I know about [topic], in my own words. Keep it tight. Cite the notes you used at the end.
TIDY · the weekly clean-up 
Go through all of notes/. Flag any two notes that contradict each other. Point out anything that has gone stale. Find notes that are not linked to anything and suggest where they belong. Update index.md. Do not delete anything without asking me first.

The rules that keep it good:

Keep index.md short. It's a table of contents, not a copy of everything. A lean index loads fast and keeps Claude accurate.
One topic per note. Small, linked files beat one giant document.
Never mass-delete. Archive instead, and let Claude suggest deletions for you to approve.
This is the same pattern Andrej Karpathy, a founding member of OpenAI, shared for personal knowledge: let an AI compile and maintain a wiki of plain markdown notes, with no vector database behind it. The people running it this way aren't developers. They're writers, consultants, and founders who got tired of re-explaining themselves to a chatbot every morning, and wanted their notes to stay theirs.