🧠

Your Second Brain Full In-Depth Setup Guide

Everything you need to build your personal knowledge base — based on Andrej Karpathy's system

Why This Changes Everything

Most people use AI like a search engine with amnesia. You ask a question, get an answer, close the tab. Tomorrow you start from scratch. Nothing accumulates. Nothing compounds.

This system flips that completely. The AI builds and maintains a structured wiki from everything you feed it. Every question you ask gets saved back in. The next question builds on the last.

The dumbest it will ever be is today.

The Architecture: 3 Folders

Create one project folder anywhere on your computer. Inside it, create three subfolders:

my-second-brain/
├── raw/        ← your junk drawer. dump everything here
├── wiki/       ← the AI writes and maintains this entirely
└── outputs/    ← answers and reports the AI generates

That's the entire architecture. No apps. No plugins. No database. No code.

  • raw/ is your intake shelf. The AI reads from here but never modifies it. This is your source of truth.
  • wiki/ is where the AI compiles everything into organized, interlinked pages. You read it. The AI writes it.
  • outputs/ is where answers to your questions live. File the good ones back into wiki/ to compound them.

The Schema File (CLAUDE.md)

Create a file called CLAUDE.md in the root of your project folder. This is the training manual for your AI librarian. Without it, the AI guesses at what matters. With it, every output is structured exactly how you want.

Copy and paste this exactly — just fill in your topic:

# Knowledge Base Schema

## What This Is
A personal knowledge base about [YOUR TOPIC — e.g. "my business and clients",
"AI research", "health and fitness", "content strategy"].

## How It's Organized
- raw/ contains unprocessed source material. Never modify these files.
- wiki/ contains the organized wiki. AI maintains this entirely.
- outputs/ contains generated reports, answers, and analyses.

## Wiki Rules
- Every topic gets its own .md file in wiki/
- Every wiki file starts with a one-paragraph summary
- Link related topics using [[topic-name]] format
- Maintain an INDEX.md that lists every topic with a one-line description
- Maintain a LOG.md as an append-only record of every operation
- When new raw sources are added, update the relevant wiki articles
- Flag any contradictions between articles with ⚠️
- Never modify files in raw/

## My Focus Areas
[List 3-5 things you want this knowledge base to focus on — be specific]

## Operations
When I say INGEST: read new files in raw/, create summary pages in wiki/sources/,
update concept and entity pages, add wikilinks, update INDEX.md and LOG.md.

When I say QUERY: read INDEX.md, find relevant pages, synthesize an answer
with citations, save to outputs/.

When I say LINT: audit the entire wiki for contradictions, orphan pages,
broken links, missing pages, stale content. Fix what you can. Flag the rest.

What to Dump Into raw/

Anything you want to stop losing or forgetting. Don't organize it. Don't rename files. Just dump it. That's the AI's job.

  • Articles and bookmarks you saved but never re-read
  • Meeting notes and call transcripts (Fathom, Fireflies, etc.)
  • Screenshots of ideas you want to keep
  • Old notes exported from Notion, Apple Notes, Google Docs
  • Research papers and PDFs
  • YouTube video transcripts
  • Your own written ideas and drafts
  • Competitor breakdowns and market research
  • Personal journal entries and reflections

The fastest way to fill raw/: Install the Obsidian Web Clipper Chrome extension. One click on any webpage saves it as a clean markdown file directly into your raw/ folder. Change the default save location in the extension settings from "clippings" to "raw".

The Four Prompts

Copy these and use them with any AI that can read your files (Claude Code, ChatGPT, Gemini, Cursor).

1. Build Your Wiki (First Time)

Run this after you've dumped files into raw/ for the first time.

Read CLAUDE.md for the project rules. Then read everything in raw/.
Compile a wiki in wiki/ following the rules in CLAUDE.md.

Create INDEX.md first with every topic listed and a one-line description.
Then create one .md file per major topic in wiki/. Link related topics
using [[wikilinks]]. Summarize every source. Create LOG.md and record
what you did.

2. Add New Sources (Ongoing)

Run this every time you add something new to raw/.

A new source has been added to raw/. Read it, then read wiki/INDEX.md
to understand what already exists.

1. Write a summary page in wiki/sources/
2. Update existing wiki pages with new information — append, don't rewrite
3. Create new concept or entity pages if needed
4. Add [[wikilinks]] to connect it to existing topics
5. Update INDEX.md and append to LOG.md
6. Flag any contradictions with ⚠️

3. Ask It Anything (Ongoing)

Use this to query your knowledge base.

Read wiki/INDEX.md. I want to understand: [YOUR QUESTION]

Research the answer across the wiki. Read the relevant pages and
synthesize an answer with citations back to specific wiki pages.
Save the answer to outputs/[topic].md and update INDEX.md and LOG.md.

4. Monthly Health Check

Run this once a month to keep your wiki clean and accurate.

Review the entire wiki/ directory and run a full lint. Report on:

1. Contradictions between articles — list both sources
2. Orphan pages with no inbound links — suggest where to add links
3. Concepts mentioned but never explained — create stubs for the top 5
4. Broken [[wikilinks]] pointing to non-existent pages
5. Claims not backed by a source in raw/
6. Stale content that may be outdated
7. Three new questions worth researching next

Fix what you can automatically. Save a report to outputs/lint-[date].md

What You Can Build This For

Content creators — track which topics, hooks, and formats actually perform. Query your own analytics data instead of guessing what to post next.

Business owners — every meeting, decision, and client detail in one place. Ask "what did we agree on across the last 5 calls with this client?" and get an actual answer.

Researchers — 50 papers synthesized into one wiki. Ask questions across all of them at once instead of re-reading every time.

Students — build the wiki a textbook never gave you. Add each chapter as you go, build out pages for concepts and how they connect.

Anyone learning something hard — your knowledge compounds instead of disappearing after the conversation ends.

The Compounding Loop

Every answer you get from your wiki — save it back in. Drop it into outputs/ or have the AI update the relevant wiki article.

The next question builds on the last. The 50th source doesn't just add a page — it updates every related entity, strengthens cross-references, and surfaces patterns you couldn't see in the first 10.

One X user turned 383 scattered files and 100+ meeting transcripts into a compact wiki and dropped their token usage by 95% when querying with Claude. That's the compounding effect in action.

Your wiki isn't a destination you fill up. It's a machine that gets smarter every time you touch it.

Tools That Speed This Up (Optional)

Obsidian (free) — download from obsidian.md. Lets you view your wiki visually with a graph showing all the connections between pages. Not required but makes browsing the wiki significantly more enjoyable.

Obsidian Web Clipper (free Chrome extension) — one-click saves any webpage as markdown directly into your raw/ folder.

Claude Code — Anthropic's command-line AI tool with full filesystem access. Reads, writes, and updates your wiki files directly without any copy-pasting. The most powerful way to run this system.

You Don't Need Obsidian or Fancy Tools

Half the internet is pitching Obsidian plugins and complex setups for this. You don't need them.

A folder of .md files, a good schema file, and an AI that maintains everything. That's the whole system. You could run it from Notepad. The AI doesn't care what app you open the files in.

Stop shopping for the perfect tool. Start building.

The System in 60 Seconds

  1. Three folders — raw, wiki, outputs. That's the architecture.
  1. One CLAUDE.md file tells the AI how to organize everything.
  1. Dump your bookmarks, notes, and articles into raw/. Don't organize them.
  1. One prompt: "Compile a wiki from raw/ following CLAUDE.md."
  1. Ask questions against your wiki. Save answers back. It compounds.
  1. Monthly health check catches errors before they stack.

Built by @artem.novitckii

Based on Andrej Karpathy's LLM Knowledge Base system