AI Second Brain

A practical walkthrough of the Claude Code + Obsidian setup I'm actually using, and the one change that made it work.

Screenshot of Obsidian and Claudian systems.

I Built My Second Brain. Here's What I Learned.

About a month ago I started going down the AI rabbit hole. One of the first things I wanted to figure out was how to build an AI-powered second brain to manage my life and my projects. Then I came across KJ Rainey's YouTube video covering exactly that, using Obsidian and Claude Code. So I watched it a couple times, built it, and then made one change that actually mattered: I split its brain into separate files instead of stuffing everything into one.

This post is a brief overview of the process. What I did, and what I changed.

This system isn't perfect. But it's perfect for me, for now. I'll give an honest update around the end of Q3.

Part One: Installation and Setup

The stack is three things:

Obsidian is the structure. The vault where everything lives. It's a local-first, Markdown-based note-taking app, which means your notes are plain text files on your computer. No lock-in, no cloud dependency, no subscription required to own your own thoughts.
Claude Code is the motor. Anthropic's agentic tool that runs in your terminal. It reads files, edits them, runs commands, and generally acts like a capable collaborator who lives inside your project directory. It's the part that makes the whole thing intelligent.
The Claudian Plugin is the assistant. An Obsidian plugin that embeds Claude Code directly into your vault. Instead of switching between your terminal and your notes, Claudian lets you talk to Claude from inside Obsidian. It reads, writes, edits, and searches your vault files in place.

The Install Order

Download and install Obsidian.
Install Claude Code (full install guide here). On Windows, I used the Command Prompt:

curl -fsSL https://claude.ai/install.cmd -o install.cmd && install.cmd && del install.cmd

On Mac or Linux:

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

Grab KJ Rainey's second brain template from his Useful AI community on Skool. It's free to join, and the Building Your AI Second Brain 2.0 course walks you through the full setup.
Drop the template into Obsidian as your vault folder.
Use Claude Code from the terminal to install the Claudian plugin, then enable it inside Obsidian (Settings, then Community Plugins, then search "Claudian" and toggle it on).

That's the foundation. Five steps, maybe an hour or two if you're moving slowly.

If you want the full, click-by-click rundown, watch KJ Rainey's video or join his Useful AI community on Skool. He walks through every step in detail. I'm keeping it short here so we can get to the part I actually changed.

Part Two: Why I Didn't Stick With One Giant File

Here's where I made my first real change to the template.

The walkthrough I followed generates a single CLAUDE.md file. One document that tells Claude everything it needs to know about you, all in one place:

Your goals
Your rules
Your personality
Your history
etc.

It works, and it's the standard way to do it.

This didn't sit right with how I think. Cramming everything into one file felt like running the whole system out of a junk drawer. So I went digging through my "Watch Later" tab to follow up on my hunch.

I came across a video by Simon at Better Creating where he walks through his own setup. In it, Simon explains splitting everything into separate files that each have one job: about-me, claude, memory. That matched how I was already thinking and gave me a cleaner way to organize my setup.

The core problem with one big file is that you're asking a single document to do too many things. Every time anything changes, you're editing the whole thing. And eventually you stop editing it at all, because it's too big and too tangled to touch. Even with AI doing the work, keeping it current turns into a chore.

So I split it up. Here's why that works better:

Each file has one job. Identity, memory, and personality stop competing for the same document. Your rules file defines how Claude should behave. Your memory file tracks what's been learned. Your about-me file describes who you are. One job, one file.
You only update what changes. Memory updates often. Weekly, sometimes daily. Core rules barely move. If everything lives in one file, every memory update becomes a chance to accidentally break a rule. Separate files mean you touch the thing that changed and leave the rest alone.
It grows with you. My About Me file evolves through journaling, Bible reading, and notes on what I'm learning about myself. My rules file doesn't need to know any of that. The structure I set up on day one stays intact while the living parts of the system keep growing.

The honest drawback: more files to manage. I don't really see this as a downside. If you're already in Obsidian, you're used to a library of linked files. And even though I'm new to Obsidian, it still made sense to me right away. One more isn't a burden. It's the system working the way it's supposed to.

Part Three: My Custom Architecture
Here's how I actually structured it.

CLAUDE.md, Structure Not Storage

My CLAUDE.md does one thing. It tells Claude who it is, what its main objective is, and what the guardrails are. That's all.

It doesn't hold my life story. It doesn't keep a running list of project updates. It's the frame that holds everything else together. The picture goes elsewhere.

About Me, My Personality in Its Own File

There's a separate file for who I am, how I think, what I value, and how I like to communicate. This one evolves slowly but meaningfully over time, as I learn more about how I actually work.

Keeping it separate means I can update it thoughtfully instead of squeezing a change into whatever white space is left in a bloated main file.

Memory.md, The Living Document

This is the file that earns its keep.

Weekly updates go here. Monthly and quarterly reviews go here. The running list of active projects lives here. When Claude needs to know what I'm working on, this is where it looks.

Because it's separate, it can change constantly without touching the rules or the identity. It's meant to be messy and current. That's the point.

Part Four: Personal Touches

I added a few things that aren't in the original template.

A blog and article section. I'm actively trying to write more, and I wanted a dedicated space inside the vault for capturing ideas, working on drafts, and tracking what I've published. The post you're reading right now lives there.

Note templates and journal templates. Still in progress. The goal is a consistent structure for daily notes and project notes, so that when Claude reads them, it's reading something predictable instead of a different format every time.

The larger point is that this system scales with me. I can add structure where I need it and leave space where I don't. That's hard to do with a rigid system you didn't build yourself.

Closing: Come Back at the End of Q3

I'm not going to tell you this is a finished system. It isn't. It's a working one.

What I know right now: the split-file structure is worth it. The Claudian plugin makes the workflow feel native instead of bolted-on. And having a real structure, even an imperfect one, beats keeping everything in your head or scattered across fifty different apps.

I'll be back at the end of Q3 with a follow-up. I'll cover what changed, what I got wrong the first time, and whether the 90-day version of this thing looks anything like what I built on day one.

You're not there yet. Neither is the system. That's the whole point of building one.