How to Give Your AI a Memory (Second Brain Setup)

I asked my AI whose birthday is coming up. It answered with the name, the date, and the fact that I'd already picked up the gift. No app, no database. A folder of plain text files.

Fable 5 is back (I covered the whole suspension mess here), and the first thing I did was hand it my entire life. Money, health, tasks, people. It all lives in one folder, and the AI never forgets any of it. In this post I'll show you what a second brain actually is, walk through a working one with eight weeks of life inside, and then build one from scratch. At the end, one prompt turns the whole thing into a Jarvis dashboard.

The problem: goldfish memory

Your life is scattered across five apps. Notes in one, tasks in another, money somewhere else. Stuff goes in all day and nothing useful ever comes back out.

And the AI you talk to every day? Every new chat starts from zero. It forgets you the second you close the tab.

What a second brain actually is

One folder. Plain text files. One file for each part of your life: your money, your health, the people you know, your plans.

The whole system is one habit. Before the AI answers you, it reads your files. When it's done, it writes back whatever changed. Read, then write. That's the entire thing.

No app, no database, no subscription to cancel later. And because it's just text on your computer, you own your data and nobody can take it from you. It's not a smarter AI. It's the same AI, finally allowed to remember.

I know it sounds too simple to work. So let me show you one.

The demo: eight weeks of life in a folder

For the video I built a demo second brain called Alex, loaded with eight weeks of template data (not my real bank statements - my actual life runs in a different one). Inside there are folders for calendar, finance, habits, health, people, projects, even recipes.

The best way to see it is Obsidian. Open the folder as a vault and you get the graph view: the budget links to subscriptions, subscriptions link to the emergency fund, the emergency fund links to a meal-prep note. As you use it, the connections keep building, and that's exactly what gives the AI context when you ask it something.

Then I opened Claude Code inside the folder and just asked questions:

"Where did my money go this month?" It read the finance logs and came back with the answer: 1,910 spent out of a 2,350 budget, a full breakdown table, and an offer to rebuild the dashboard.
"How is my running going?" Distance climbing, what's driving it, current status against the 10K goal, and an offer to sketch a week-by-week long-run plan.
"What does my week look like?" Calendar events and open tasks, pulled from two different folders and connected.
"Whose birthday is coming up?" Tiago, this Sunday, gift already sorted - plus related context it found on its own: a friend's 10K, a flight still on the backlog.

Every answer is contextualized because everything is linked. One question, and it pulls from the calendar, the tasks, and the people files at once.

You can also build dashboards that feed straight from the files, with small automations running in the background to keep them fresh. Mine rebuild themselves on a schedule while I do nothing.

Build your own in a few minutes

You don't want to copy Alex. You want your own. I packaged the exact setup as a skill - it's in the Second Brain Setup Kit.

Make an empty folder (mine is called my-brain) and drop the skill into a .claude/skills/ folder inside it.
Open Claude Code there and trust the folder.
Pick a strong model for the setup. Run /model and select Fable 5 - you only need the big model to build the structure, day-to-day use works fine on anything. One tip: keep the effort at high or xhigh, not max. On max it overthinks a four-question interview.
Run /setup-second-brain. It checks the folder is empty and asks four questions: your first name and one line about yourself, your top three goals, what you want to track, and what you're tired of forgetting.

I answered as John, a video editor who wants to lose 10 pounds, hit the gym three times a week, track money and weight, and stop forgetting birthdays and old subscriptions. A few seconds later the brain was built: folders for money, people, health, journal - shaped by the answers, not a fixed template - plus a CLAUDE.md file at the root. That file is the map: who you are, what each folder holds, and the session protocol (read first, write back after). It's what keeps the whole thing organized.

The restart test

Then I taught it three facts in plain English: Anna's birthday is August 3rd, I pay for Netflix, Spotify, and a gym I haven't visited since March, and I ran a 5K in 31 minutes on Tuesday. It filed each one into the right place - a new line in the birthdays file, the subscriptions with a note about the unused gym, a running log in health.

Then the real test. Close the session. Open a fresh one. Ask: "What do you know about me?"

It knew everything. The name, the goals, the subscriptions, the run. The memory lives in the files rather than in the chat, so a fresh session starts with your life already loaded.

One prompt to Jarvis

The finale. Open a session and ask:

Build me a dashboard over my data, make it feel like Jarvis.

You get an animated HUD with a greeting ("Good evening, John - your second brain is online and tracking three objectives"), vital signs for the gym visits and weight goal, mission objectives, a proximity radar for upcoming dates, recurring burn for the subscriptions, and a system log. It even lists what it's still missing - baseline weight, subscription prices - which is exactly the context we never gave it.

It looks like magic. It's a folder full of text files wearing an Iron Man suit.

The one rule

If it's not in the text files, it doesn't exist. That's the whole system: the AI reads your folder before answering and writes back after.

You've probably heard that building a second brain is procrastination with extra steps. Sometimes true - when you're the one maintaining it. Here the AI does the maintenance and your only job is to open the folder. Set it up today, ask it tomorrow morning what it knows about you, and see what that feels like.