Second Brain Mastery: How to Build an Obsidian + AI System That Actually Remembers You

Stop re-explaining yourself to ChatGPT. Here's the folder structure, the rules, and the workflow I use every day to make AI feel like a partner instead of an intern with amnesia.

If you've ever caught yourself thinking, "I hired this AI assistant, but it acts like an intern with amnesia," you're not alone. That phrase has come up in my DMs, on client calls, and in coffee shops at least a dozen times in the last three months. You're not the problem, and your AI tool isn't really the problem either.

The problem is context. The lack of a place for your AI to actually remember anything between conversations.

I've been wrestling with this since ChatGPT launched. Two-plus years of trial, error, and YouTube rabbit holes. What I landed on is a system I call the second brain. (Or in my case, the main brain, because I named the folder before I thought about the marketing. The architecture matters, not the name.)

This article walks you through what it is, why a prompt library will never get you there, and the four folders that make the whole thing work. It's the 101 version. The full iceberg is what I'm covering in the 10-week AI Content System course launching in July.

Want to watch me walk through this live? The full webinar recording is embedded above.

What Is a Second Brain for AI?

A second brain for AI is a local folder structure on your computer that holds your source material, your atomic knowledge pages, your daily journal, and your inbox of half-thoughts. It's organized so any AI assistant you point at it (Claude, ChatGPT, Gemini, whichever) can read it and write to parts of it. The payoff: full context about your work, your voice, and the humans you serve.

It's not an app. It's not a SaaS subscription. It's folders with files with rules.

I'll say that again because it's the whole game: folders with files with rules. If you can embrace those five words, the rest is execution.

Why a Prompt Library Doesn't Fix the Amnesia Problem

Here's the trap most of us fall into. AI disappoints us, so we build a prompt library. Maybe a Notion page with 47 favorite prompts. Maybe a Skool community with prompt swaps. Maybe one of the umpteen YouTubers selling prompt courses.

A prompt library is a copy and paste prosthetic. It's not solving the actual problem. A bigger context window is the same thing with more padding. The prompt still has to live somewhere, and you're still copying and pasting every time. The intern still has amnesia. You've just given them a slightly larger notepad.

The real shift: stop thinking of AI as a conversation, and start thinking of AI as a memory layer.

The conversation is the surface. The memory is the substrate. The conversation is what you see. The memory is what makes the conversation actually worth having.

When your AI has a place to remember you between sessions, when it already knows your stories, your voice, your client list, your last 60 podcast transcripts, your January pillar article, the game changes. You stop re-explaining. You start activating. You don't ask the AI to write a LinkedIn post. You ask your assistant to write a LinkedIn post that draws on three meetings you had with Chris last week and the pillar article you shipped in January. And it can. Because the memory is right there.

The Four Folders That Run the Whole System

The entire architecture fits on one slide. Four folders. That's the base. Everything else is what you decide to layer on top.

Folder 1: Raw

Raw is the source material. Every transcript, every article you save, every meeting recording, every voice sample, every Claude conversation you exported. Anything from the outside world, or from your own past, lands here.

Raw is sacred. It is never edited. Only appended.

Here's why that rule matters. Everything downstream depends on raw being trustworthy. When your AI builds a wiki page about a concept it pulled from a podcast episode, it needs to cite the exact line where that concept came from. If you've been silently editing raw files, that citation is a lie. Trust collapses. The system collapses. Now you've built something with the same problem as the default version: hallucination.

Treat raw like a research library. You can read from it all day. You don't get to rewrite the books and put them back on the shelf. Just keep adding. Never subtract.

Folder 2: Wiki

Wiki is where the synthesized knowledge lives that your AI reads from.

It's a set of atomic markdown pages. One concept per page. One person per page. One tool per page. Each page has a small YAML header with type, tags, and sources. The body is the description. The bottom has links to related wiki pages and back to the raw source files.

Here's why atomic matters. When you ask your AI "what do you know about pillar pages," it doesn't have to read every transcript you've saved. It reads the index, follows the links into the relevant wiki pages, pulls the atomic notes, and synthesizes the answer with citations.

That's also how you and I learn. Somebody asks you about pillar pages, your brain doesn't replay every conversation you've had on the topic. It fires the relevant connections and the answer comes out. We're building a digital version of how the brain actually works.

Folder 3: Intake

Intake is your inbox. The frictionless capture zone.

You see a URL you want to remember? Save it. You're brushing your teeth and a thought hits? Save it. You hear a quote on a podcast? Save it. The point is that nothing dies because you didn't have a place to put it in the moment. You have a place. It takes two seconds. Drop it in, move on.

Then, once a day or once a week, you run a slash command (mine is `/process-intake`) and your AI triages everything. Some items get promoted to raw because they're real sources worth keeping. Some get expanded into wiki pages. Some get archived. Intake is the holding tank with a clear exit ramp: capture, then commit to the brain.

Folder 4: Journal

Journal is the daily log. One file per day, organized into monthly folders.

What did you work on? What did you decide? What broke? What did you learn? What's still open?

This isn't a "dear diary" journal. It's a work log, so that the future you (and your AI) can answer the question, "What was George up to on the 2nd of May?"

The magic in my setup is that I don't actually type most of this. There's a hook in Claude Code that appends a one-paragraph session summary every time I close a session. The journal mostly writes itself. The machine handles the bookkeeping. (I hate bookkeeping. That's why I hire somebody for that.)

When I ask the AI later, "Hey, I was talking to somebody last week about that thing, can you figure out what day that was and who I was with?" the journal answers. I pull the thread. I run with it.

The Data Flow

Day to day, the four folders work together:

1. You drop a source into raw.
2. The AI reads raw and writes wiki pages from it, with backlinks.
3. You capture half-thoughts, URLs, and clips into intake throughout the day.
4. You run `/process-intake`, and the AI routes everything to raw, wiki, or the archive.
5. The hook writes the journal as you work.
6. You ask questions, and the brain answers using raw + wiki + journal as memory.

The reason this works isn't the folders. It's the rules attached to each one. Raw is sacred. Wiki is atomic. Intake is frictionless. Journal is automatic. Honor the rules and the system compounds.

How Content Actually Gets Into the Brain

You don't have to type all this stuff yourself. The ingestion paths I use:

• Meeting transcripts: ClickUp's note taker runs on every client call. When the call ends, I hit a "Pull" button in my Obsidian command center, the transcript drops into intake, and the system routes it. One button push.
• Local video and audio: Whisper running locally transcribes any video or audio file for free. (Not WhisperFlow. The actual OpenAI Whisper model that you can run on your own machine.) Drop the file, run the script, the transcript lands in raw.
• Web articles: The Obsidian Web Clipper with a custom template sends content (plus core idea, why it matters, hook options, and a LinkedIn draft) straight to intake. Like having an assistant watch YouTube videos with me.
• Whole websites: Firecrawl pulls every page of a client's site into their folder. Cheap API, fast, a research library I can audit against.

Notice what's not in that list: manual file management. I push buttons. The system routes.

The Obsidian Command Center I Built (Echo)

I'm running this inside Obsidian, with a custom plugin I call the Main Brain Command Center. The voice running inside it is Echo. (Echo's job description is in the name: echo my work forward, farther and louder.)

Here's the wild part. I'm not a developer. I built the plugin by opening Claude Code and asking, "Are you able to help me build an Obsidian plugin?" Claude said yes. I told it what I wanted. Tab for today's brief. Tab for clients with per-client pull buttons. Tab for focus items. Tab for an Echo chat window with voice-in and voice-out. We built it together over a couple sessions.

Earlier today, before this webinar, a client emailed: "Anything you're waiting on from us before I head out for the long weekend?" Historically I would have opened ClickUp, navigated to their folder, scrolled meeting notes, and pieced together a response. Today, I opened Echo, asked "What are we waiting on from this client?", got the list in five seconds, pasted it into the email, and hit send. Five minutes of work compressed to thirty seconds. The time I got back went into being more human with the humans I work with.

Why This Beats Every Notes App You've Tried

Notes apps are built for your eyeballs. The second brain is built for two readers: future you, and your AI. Only one of them is going to be doing the heavy lifting for you over the next decade. Optimize accordingly.

When the data is in plain markdown files on your hard drive, every AI tool that comes out next year, the year after, and the year after that can read it. Switch from Claude to Gemini? Same brain. Switch from Cursor to whatever comes next? Same brain. You aren't locked into a vendor. You aren't paying for a subscription that disappears.

That's portability. That's permanence. That's worth way more than fancy formatting.

The Pre-Flight Checklist

1. Install the folder structure first. Don't optimize. Just raw, wiki, intake, journal. Get the bones.
2. Capture before you curate. Spend a week dumping things into raw and intake. Feed the brain. The wiki happens later.
3. Get the slash commands. Once your system is built, ask your AI: "Can you give me a user guide for the slash commands I can use with the system we just built?" The system knows itself. Let it teach you.
4. Activate Claude Code. This is where the deeper magic happens. Hooks, custom commands, automated journaling. Don't worry about it on day one. Get there on day fifteen.
5. Build a command center when you're ready. You can run the whole system without it for the first month and still get most of the value.

A few warnings. Don't over-engineer the folder structure. Never skip raw, ever; the temptation to summarize on the way in kills the whole system. Don't draft directly into wiki; new ideas go into intake first. And don't treat this like a notes app. If you treat it like Evernote with extra steps, you'll bounce off.

What You Actually Get

You don't re-explain yourself to AI anymore. The brain has the context.

You write content that sounds like you. I asked Echo to write a 650-word article on hustle versus healthy living for agency owners, pulling a quote from my Beyond Your Default podcast and a personal story from my story bank. It opened with the line, "Agency owners, your hustle is lying to you." Then it told the stretcher story. The moment paramedics wheeled me out of my office on a stretcher and I told them, "I don't have time for this. I have a meeting in 10 minutes." That was 95% ready to ship. The AI didn't invent that story. It pulled it from the brain. From the raw materials I'd fed in.

You stop losing things. The intake folder catches everything before it disappears into the ether.

You compound. Every transcript, every wiki page, every journal entry, every clip adds to the substrate. Month six is more useful than month one. Month twelve is a different machine.

Frequently Asked Questions

Is this local on my computer, or in the cloud?

Everything runs locally on my machine. I push the brain to a private GitHub repo as a backup. For now, it's my second brain, just for me.

Which AI model do I have to use?

Any of them. The architecture is model-agnostic. Claude, ChatGPT, Gemini, or whatever comes out next quarter. Switch when something better lands. The brain stays put.

How much does this cost?

The folders cost nothing. Obsidian is free. Whisper running locally is free. The cost is in whichever AI you point at it. If you're already paying for ChatGPT Plus or a Claude subscription, you can run the basics for the same cost.

Do I need to know how to code?

No. I'm not a developer. I built my command center plugin by asking Claude Code to help me. The barrier is patience, not technical skill.

How long until it's actually useful?

You'll feel the difference within the first week of consistent feeding. By month three, the system is a different brain than what you started with. Feed it every day for 90 days and see for yourself.

The Takeaway

Stop trying to make your AI smarter with better prompts. Build it a memory layer instead. Folders with files with rules. That's the whole game.

Your AI doesn't have amnesia because it's broken. It has amnesia because nobody gave it a place to remember you. Once you do, everything changes. The conversations get sharper. The content sounds like you. The intern grows up. You get back the hours you used to spend re-explaining yourself.

If you want to go deeper than the 101 version, the full 10-week AI Content System course launches in July. It walks you through every layer of the iceberg: wiki conventions, slash commands, identity profiles, Claude Code hooks, the command center build, and the workflows that make this a daily operating system. If that's where you want to go next, DM me on LinkedIn and I'll get you on the early list.

In the meantime, here's the homework. Pick one folder. Just raw. Spend a weekend dumping every transcript, every saved article, every meeting note, every voice memo into one folder. Don't organize it. Don't summarize it. Just feed it. By Monday, you'll have something the rest of the system can be built on top of.

The brain be braining. But only if you give it something to remember.