The Complete Guide

What You’re Building (and Why It Works)

The philosophy: your only job is to dump and ask

Here’s the deal. You have two jobs in this system:

1. Dump — Drop raw material into the raw/ folder. Meeting notes, articles, book highlights, ideas, research, anything. Don’t organise it. Don’t tag it. Don’t think about where it goes. Just dump.
2. Ask — Ask the AI questions. “What do I know about X?” “Help me think through Y.” “What patterns do you see?” The AI searches your entire knowledge base, synthesises an answer, and saves it.

That’s it. Everything else — the structuring, the cross-linking, the organising, the indexing — is the AI’s job. You are the collector and the thinker. The AI is the librarian.

Most knowledge systems fail because they ask you to be the librarian andthe thinker. You have to decide which folder, which tag, which link, which format. That’s busywork that adds friction and kills the habit. This system removes all of that friction. You dump, you ask, and the AI handles the rest.

How Karpathy’s method works

Andrej Karpathy (founding team at OpenAI, former head of AI at Tesla) described his personal wiki system: a set of markdown files maintained by an LLM. He dumps information in, the LLM structures it into wiki pages, and he queries it to recall and synthesise what he knows.

Your Second Brain Kit implements this method with a specific architecture:

Three folders:

raw/     Your dump zone. Everything goes here. Never modified by the AI.
wiki/    The AI's structured output. Interlinked markdown pages, like a personal Wikipedia.
outputs/ Saved answers to your questions. Reports, analysis, synthesis.

One instruction file:

CLAUDE.md   The AI reads this file to understand how to behave. It defines the
            folder structure, the ingestion protocol, the wiki conventions, and
            every rule the AI follows. This is the brain of the system.

The cycle:

1. You dump a file into raw/.
2. You tell the AI to ingest it.
3. The AI reads the file, extracts structured information, creates or updates wiki pages with proper formatting, cross-links, and citations back to the original source.
4. You ask questions. The AI searches both wiki/ and raw/, synthesises an answer, and saves it to outputs/.
5. Over time, your wiki grows. Connections form. The AI can see patterns across everything you’ve ever dumped.

Why this beats Notion/Obsidian/everything else

Notionis powerful but you’re the librarian. You design the databases, maintain the views, manually link pages, and the information just sits there unless you go looking. There’s no synthesis layer.

Obsidianis closer — it has backlinks and a graph view. But you’re still manually creating the links, choosing the tags, structuring each note. And it doesn’t synthesise across notes to answer your questions.

Apple Notes / Google Docs / Evernote are dumping grounds with search. Useful, but no structure forms. No connections emerge. No synthesis happens.

This system separates the human job (dumping and asking) from the AI job (structuring, linking, synthesising). You get:

• Zero-friction input. Drop a file. Done. No tagging, no folders, no metadata to fill in.
• Automatic structure. The AI creates properly formatted wiki pages with frontmatter, headings, cross-links, and citations.
• Synthesis on demand. Ask a question and get an answer drawn from everything you’ve ever put in.
• Compounding returns. Every piece of information you add makes every other piece more valuable, because the AI can cross-reference and synthesise across the whole base.

What “compounding knowledge” actually means in practice

In month one, you dump 20 files. The AI creates 15 wiki pages. You ask it some questions. Useful, but not dramatically different from a search tool.

In month three, you have 80 files and 60 wiki pages. The AI starts connecting things you didn’t connect yourself. You ask “What patterns do you see in my decision-making?” and it pulls from your meeting notes, your book highlights, your personal reflections, and your project retrospectives. You get insights you genuinely wouldn’t have reached on your own.

In month six, you have 200 files and 150 wiki pages. You dump a new article about a market trend. The AI connects it to three competitor analyses you did months ago, a book you read about the same topic, and a conversation note where someone mentioned the same dynamic. Your wiki is now smarter than any single person’s memory.

That’s compounding. Each piece of knowledge is more valuable because of every other piece already in the system.

Setup (Step by Step)

What you need

Hardware:

• Mac with Apple Silicon (M1 or later) — recommended
• Or Windows/Linux with 16GB+ RAM
• At least 10GB free disk space (your knowledge base will be mostly text, which is tiny)

Software:

• Claude Code — This is the AI engine that powers your Second Brain. It’s Anthropic’s official CLI tool that reads your CLAUDE.md schema and maintains your entire knowledge base.

That’s it. Claude Code is the only tool you need.

Step 1: Install Claude Code

On Mac:

1. Open Terminal (press Cmd+Space, type “Terminal”, hit Enter).
2. Check if you have Node.js:
   $ node --version

   If you see a version number (v18 or higher), skip to step 3. If you get “command not found”, install Node.js first:

   $ brew install node

   (If you don’t have Homebrew either: go to brew.sh, copy the install command, paste it into Terminal, then run brew install node.)

3. Install Claude Code:
   $ npm install -g @anthropic-ai/claude-code

4. Verify it installed:
   $ claude --version

   You should see a version number. You’re good.

On Windows:

1. Open PowerShell or Command Prompt.
2. Install Node.js from nodejs.org (download the LTS version, run the installer).
3. Restart your terminal, then:
   $ npm install -g @anthropic-ai/claude-code

4. Verify: claude --version

Step 2: Create your Second Brain folder

Choose where you want your Second Brain to live. Your home folder or Documents is fine.

On Mac:

$ mkdir ~/second-brain

On Windows:

$ mkdir %USERPROFILE%\second-brain

Now unzip the kit you downloaded and copy everything into this folder. Your folder should look like:

second-brain/
  CLAUDE.md
  GUIDE.md
  README.md
  .gitignore
  raw/
    README.md
  wiki/
    INDEX.md
  outputs/
    README.md

Step 3: Open Claude Code in your Second Brain folder

This is the key step — you need to open Claude Code inside your Second Brain folder so it reads the CLAUDE.md schema.

On Mac:

$ cd ~/second-brain$ claude

On Windows:

$ cd %USERPROFILE%\second-brain$ claude

The first time you run Claude Code, it will ask you to authenticate. Follow the prompts — you’ll either sign in with your Anthropic account or enter an API key via the browser.

Once Claude Code starts, it automatically reads CLAUDE.md and understands the entire system. You’ll see it ready for your instructions. That’s your Second Brain, active and waiting.

Step 4: Verify it’s working

Type this into Claude Code:

Read CLAUDE.md and confirm you understand the system. What are the key rules you’ll follow?

Claude should respond with a summary of the schema rules — folder structure, ingestion protocol, wiki conventions, etc. If it does, everything is working.

Step 5: Viewing your files (optional)

You can view and browse your markdown files with any text editor:

• VS Code (free): Open the second-brain/ folder with File > Open Folder
• Obsidian (free): Point it at your second-brain/ folder for nice wiki viewing with backlinks
• Any text editor: The files are plain markdown — they work everywhere

But for interacting with the AI, always use Claude Code in the terminal. That’s where the magic happens.

Configuring CLAUDE.md

Open CLAUDE.mdin your editor. At the top, you’ll see the header with placeholders. Replace them:

That’s the only configuration required. The rest of CLAUDE.md is the instruction set the AI follows. You don’t need to read all 800+ lines now — but it’s worth skimming later to understand what the AI does with your material.

Verifying the folder structure

After configuration, your kit should look like this:

second-brain-kit/
  CLAUDE.md              (placeholders filled in)
  GUIDE.md               (this file)
  README.md              (the welcome file)
  .gitignore             (for version control)
  raw/
    README.md
  wiki/
    INDEX.md
  outputs/
    README.md

If anything is missing, something went wrong during download. Email paul@secondbrain.nzand we’ll sort it out.

Your First Session (Do This Now)

Set aside 15 minutes. You’re going to dump some material, trigger ingestion, and ask your first questions.

Step 1: Pick 3–5 files to dump into raw/

Grab a few things from your existing digital life. The more varied, the better — it shows you how the AI handles different types of material.

Good first dumps:

1. A recent meeting note. Copy-paste from wherever you keep it (email, notes app, Google Docs) into a new file: raw/2026-04-08-meeting-notes-example.md
2. A document you reference often. A strategic plan, a product spec, a framework you use, your company’s mission statement. Copy it into raw/.
3. A piece of research. An article you read recently, a competitor analysis, market research. Paste the text into a markdown file in raw/.
4. A book summary or highlights. If you use Kindle, Readwise, or just have notes from a book you read, drop them in.
5. A personal framework or set of principles. Something you’ve written about how you make decisions, what you value, or how you think about a specific topic.

Quick way to create files:

Open your editor, create a new file, paste your content, save it to the raw/ folder with a descriptive name. Or from the terminal:

# Paste your content into a new file$ pbpaste > raw/2026-04-08-my-meeting-notes.md

Step 2: Trigger ingestion

Open the AI chat (Cursor: Cmd+L, Claude Code: just type) and paste this prompt:

Ingest everything new in raw/ and update the wiki. For each file:
1. Read the full content
2. Create or update wiki pages with proper frontmatter, structure, and cross-links
3. Update wiki/INDEX.md
4. Give me a summary of what you created

The AI will work through each file, creating structured wiki pages. This might take a few minutes depending on how much material you dumped.

Step 3: Watch what happens

As the AI processes your files, you’ll see wiki pages appearing in the wiki/ folder. Open a few and notice:

• Frontmatter at the top of each page (title, type, created date, sources, tags, related pages).
• Structured sections with proper headings.
• Cross-links to other wiki pages where topics overlap.
• Citations back to the raw files where information came from.
• INDEX.md updated with every new page and a one-line description.

This is the AI doing the librarian work for you. Every fact traces back to a source. Every related topic links to its page.

Step 4: Ask your first questions

Now for the fun part. Ask the AI 3 questions about the material you just dumped:

Question 1 — Simple retrieval:

What do I know about [topic from your dumped files]?

Question 2 — Synthesis across sources:

Based on everything in the knowledge base, what are the key themes or patterns you can identify?

Question 3 — Applied thinking:

If I had to make a decision about [something relevant to your dumps], what does the knowledge base suggest I should consider?

Notice how the AI draws from multiple wiki pages and raw files to construct answers. It cites its sources. It flags gaps where information is missing.

Step 5: Check outputs/

Open the outputs/folder. You should see saved files from your substantive questions, named with today’s date and a descriptive slug. These are the permanent record of your queries and the AI’s synthesised answers.

Congratulations. Your knowledge base is live. From here, it only gets better.

Daily Practice

The 5-minute daily habit

The system works because of a simple daily habit: dump and ask.

Every day, spend 5 minutes doing two things:

1. Dump something (1–2 minutes). Drop at least one new file into raw/. It doesn’t have to be polished or complete. A few bullet points from a meeting. An article you read. A thought you had. Anything.
2. Ask something (2–3 minutes). Ask the AI a question that matters to you today. It could be about something you just dumped, or something that’s been on your mind. Let the AI synthesise.

That’s the whole habit. Five minutes. The compounding happens automatically.

What to dump (20 examples)

To keep your knowledge base growing and useful, dump a variety of material. Here are 20 types with examples:

Books and reading:

1. Book highlights and margin notes (“Here are my highlights from Thinking, Fast and Slow...”)
2. Article summaries (“I read this piece on the future of remote work...”)
3. Podcast notes (“Key points from the Tim Ferriss episode with...”)
4. Course or workshop notes (“Notes from the leadership workshop I attended...”)

Decisions and reasoning:

5. Decision journals (“I decided to do X because... the alternatives were...”)
6. Post-mortems (“Here’s what happened with the product launch...”)
7. Trade-off analyses (“We chose A over B because...”)
8. Prediction logs (“I predict X will happen by Q3 because...”)

Ideas and creative work:

9. Rough ideas (“I have a half-baked idea about...”)
10. Project briefs (“Here’s what I’m thinking for the next project...”)
11. Content drafts (“Draft of the blog post about...”)
12. Creative concepts (“What if we approached this differently by...”)

Professional knowledge:

13. Meeting notes (“Key points from today’s board meeting...”)
14. Industry research (“Here’s what I found about the market for...”)
15. Competitor analysis (“Looked at what Company X is doing...”)
16. Client or stakeholder notes (“Call with the investor today...”)

Personal knowledge:

17. Lessons learned (“Things I learned this week/month/quarter...”)
18. Frameworks and models (“My framework for evaluating opportunities...”)
19. Reflections (“Thinking about how I handled the situation with...”)
20. Relationship context (“Met someone interesting at the conference...”)

How to ask good questions

Not all questions are equal. Here’s how to get the most from your knowledge base:

How answers compound over time

In week 1, the AI has 5 files to work with. Its answers are thin.

In month 1, it has 30 files and 20 wiki pages. Its answers start connecting things.

In month 3, it has 100 files and 70 wiki pages. You start getting answers that surprise you — connections you didn’t see, patterns you didn’t notice, forgotten insights resurfaced at exactly the right moment.

In month 6, you have a genuine second brain. You can ask “What do I think about X?” and get a rich, nuanced answer drawn from everything you’ve read, written, discussed, and decided over half a year.

This is what compounding knowledge looks like. The most valuable answers come from the intersections between things you dumped months apart.

Weekly and Monthly Rhythms

Weekly review (15 minutes)

Set aside 15 minutes once a week. Friday afternoon or Sunday evening work well.

Dump (5 min):

• Go through your week. What did you learn? What did you decide? What did you read?
• Dump anything you haven’t already: meeting notes, articles, reflections.

Query (5 min):

• Ask: “What are the most important themes from what I’ve added this week?”
• Ask one question that’s been on your mind.

Quick check (5 min):

• Glance at wiki/INDEX.md — is it growing? Are new sections forming?
• Open one or two recent wiki pages. Do they look right? Is the AI structuring things well?
• If something looks off, tell the AI: “The wiki page about X doesn’t capture the key point. The main insight was Y.”

Monthly health check (30 minutes)

Once a month, run a full health check. Paste this prompt:

Run a full health check on my knowledge base. Check for:
1. Contradictions between wiki pages
2. Knowledge gaps (what's missing?)
3. Stale pages that need updating
4. Broken cross-links
5. Orphan pages with no connections
6. How well raw/ material has been ingested

Save the report to outputs/health-checks/

What it finds:

• Pages that haven’t been updated in a while.
• Information that contradicts across different pages (you said one thing in January, something different in March).
• Topics that are referenced but don’t have their own page yet.
• Areas where your knowledge is thin and could use more material.

How to act on it:

• Look at the gaps. Do you have raw material that could fill them? Dump it.
• Look at contradictions. Which is correct? Tell the AI.
• Look at stale pages. Has anything changed? Dump an update.
• Look at orphans. Should they connect to something? Tell the AI to link them.

Quarterly reflection

Every three months, ask the AI for a big-picture review:

Generate a quarterly review of my knowledge base. I want to understand:
1. How the knowledge base has grown (new pages, new sections, volume)
2. What topics I've focused on most
3. What topics I've neglected
4. What patterns or themes have emerged across my knowledge
5. What the most valuable insights have been
6. What I should focus on learning or capturing next quarter

Save this to outputs/reports/

This is one of the most valuable things the system does. It holds up a mirror to your thinking and shows you where your attention has been going.

Customisation

How to modify CLAUDE.md for your domains

CLAUDE.md is not sacred. It’s designed to be modified. The AI follows whatever instructions are in there, so if you want to change its behaviour, change the file.

Common modifications:

Change the wiki sections. The default sections (personal knowledge, professional knowledge, reference) are a starting point. If you work in a specific domain, add sections for it:

Open CLAUDE.md, find the wiki section definitions, and add your own following the existing pattern:

### wiki/investments/

**Purpose:** Track investment research, portfolio decisions, and market analysis.

**Required pages:**
- investments-overview.md -- Portfolio summary, strategy, allocation.
- asset-[slug].md -- One page per investment or asset class.
- thesis-[slug].md -- Investment theses with evidence.
- performance.md -- Returns, benchmarks, attribution.

Adjust ingestion priorities.If certain types of material are more important to you, tell the AI in the ingestion section of CLAUDE.md. For example: “Prioritise book notes and decision journals over meeting notes.”

Change output format. If you want reports in a specific format, modify the output formatting section. The AI will follow whatever structure you define.

Adding new wiki sections

1. Open CLAUDE.md.
2. Find the Wiki Section Definitions section.
3. Add a new section following the pattern: purpose, required pages, compilation rules.
4. Tell the AI: “I’ve updated CLAUDE.md with a new wiki section. Please create the folder and initial pages.”

Adjusting anonymisation rules

By default, the AI anonymises personal information (names, emails, phone numbers) in wiki pages. This is a privacy protection.

• To turn off anonymisation entirely: Find the anonymisation section in CLAUDE.md and add: “Anonymisation is disabled for this knowledge base. Use real names and details.”
• To turn it off for specific people: Tell the AI: “Use real names for [Person Name].”
• To increase anonymisation: Add categories to the anonymisation list. For example, if you want company names anonymised too, add that rule.

Creating custom prompts

Beyond the 12 prompt cards included in the kit, you can create your own. The pattern is:

1. Write a clear instruction for the AI.
2. Reference the knowledge base explicitly (“using everything in the wiki” or “search raw/ and wiki/”).
3. Specify the output format you want.
4. Specify where to save it (outputs/).

Example custom prompt:

Search my entire knowledge base for everything related to [client name].
Synthesise a comprehensive profile including: what we've discussed,
what decisions have been made, what's pending, and what I should
prepare for our next meeting. Save to outputs/queries/

Industry-specific tips

Optional Enhancements

These are nice-to-haves. The system works without them, but they add polish.

Using with Obsidian (setup in 5 minutes)

Obsidian is a markdown editor with a graph view that makes your wiki beautiful. You can use it alongside Cursor or Claude Code as a viewer.

1. Download Obsidian from obsidian.md (free).
2. Open Obsidian. Click “Open folder as vault.”
3. Select your second-brain-kit/ folder.
4. Your wiki pages are now browsable in Obsidian with backlinks and a graph view.

Git version control (for non-technical users)

Git tracks changes to your files over time, so you can always go back to a previous version. It’s like Time Machine for your knowledge base.

Setup (one time):

$ cd /path/to/second-brain-kit$ git init$ git add -A$ git commit -m "Initial knowledge base setup"

Daily use (optional but recommended):

After each session, save your progress:

$ cd /path/to/second-brain-kit$ git add -A$ git commit -m "Knowledge base update - $(date +%Y-%m-%d)"

That’s it. If you ever need to undo something or see how a page changed over time, git has your back.

For extra safety, push to a private GitHub repository:

1. Create a private repository on github.com.
2. Follow GitHub’s instructions to connect your local folder.
3. After each commit: git push

Now your knowledge base is backed up in the cloud.

Sharing with a team member

The system is designed for individual use, but you can share it:

1. Shared folder approach: Put the kit folder in a shared location (Dropbox, Google Drive, or a shared server). Multiple people can dump into raw/.
2. Git collaboration: Use a shared private repository. Each person commits their changes. The AI handles merging the wiki.
3. Limitations: The AI processes one person’s session at a time. If two people trigger ingestion simultaneously, results may conflict. Best practice: coordinate who runs ingestion, or have one person own the ingestion process.

For teams larger than 2–3 people, consider the Done-For-You build, which includes team setup and coordination.

Upgrading to Done-For-You

What the full build adds

The self-serve kit gives you the system. The Done-For-You build gives you the system customised, seeded, and running from day one:

• Custom CLAUDE.md — We interview you, understand your domains, and write a CLAUDE.md tailored to your specific knowledge areas, workflow, and thinking style. Much more detailed than the template you have now.
• Initial knowledge seeding — We ingest your existing material (documents, notes, articles, whatever you have) and build out your initial wiki. You start with a populated knowledge base, not an empty one.
• Domain-specific wiki sections — Custom sections designed for your specific industry and role.
• Advanced ingestion rules — Custom rules for the types of material you work with most.
• Workflow integration — Setup for your specific tools (Obsidian, git, cloud sync, etc.).
• Onboarding session — 60-minute call where we walk through your running system together.

Pricing

Your kit purchase is credited toward a Done-For-You build:

How to book

Visit secondbrain.nz/book or email paul@secondbrain.nz.

Mention that you’re a kit customer and we’ll apply your purchase credit automatically. We’ll schedule a 30-minute discovery call to understand your needs, then get started.

Troubleshooting

Keyboard Shortcuts (Cursor)

Common Prompts Quick Reference

These are in the prompts/folder as full prompt cards, but here’s a quick reference: