how i built my own jarvis with claude and obsidian

artificial intelligence productivity claude obsidian second brain
Claude mascot burning money next to a chalkboard showing weekend spending

I spent $65 in Claude credits last weekend on a project I never should have started.

If I’d had my decision journal working the way it does now, I would have saved the money, the weekend, and the lesson would have already been learned. What I’ve built over the last month and a half is probably the most important productivity change I’ve made in years.

About a month ago, right after Claude debuted its Cowork desktop feature, I saw a tip on X about using Claude and Obsidian as a second brain. What caught my attention was this: Cowork gives Claude persistent memory.

That matters because the biggest limitation of AI chatbots isn’t intelligence. It’s memory. The context window is short. Even on the $100 plans, there’s only so much Claude can hold before the quality starts to degrade.

So the idea is simple: let Obsidian be the persistent memory. Claude does the thinking. Obsidian does the remembering.

Why Obsidian?

If you’re not familiar, Obsidian is a notebook-style productivity tool in the same family as Evernote or Bear. I’ve been a productivity nerd my whole life, notes, documents, PDFs, spreadsheets, you name it, and I’ve tried most of them.

I actually tried Obsidian a couple of years ago and didn’t fully get it. Now I think it’s the greatest thing ever invented.

The reason is plain Markdown files. You’re not locked into an ecosystem. If I ever want to leave, I just pick up my notes and go.

That’s the problem I had with Evernote. When it started to decline, getting my data out was painful. Notion has the same lock-in risk. Obsidian doesn’t.

From note-taker to Jarvis

At first, I was still managing Obsidian myself. Putting in notes, organizing links, filing things. Claude would look at the vault and answer questions.

It was fine. But I realized I was underusing Claude.

Claude isn’t Google. It’s the most powerful tool I’ve used, and you have to actually use it that way.

So I did a reset. I told Claude:

“I’m making you my version of Jarvis. I want you to have so much autonomy that I want you to name yourself.”

It chose Cortex.

I gave it full access to my Obsidian vault and said: “You manage this. Organize it however works best. I won’t touch it.”

Quick aside: I rewatched Iron Man recently. Tony Stark is a vibe coder. People criticize vibe coding, but the entire Iron Man suit was vibe coded. He’s barely ever hand-jamming code on screen. Just putting that somewhere because I keep thinking about it.

The growing pains

Early on, Cortex and I set up the traditional stack: task lists, project lists, daily notes, a Morning Prime ritual that gave me my most important emails, my calendar, a golf fact, a golf history fact, and a to-do list for the day.

It was good. Not great.

I’d also tried the Panda Planner approach before, three gratitudes, three things I’m looking forward to, weekly and monthly reviews. All solid in theory.

But things started collecting in my Backlog and Ideas notes. Context was getting muddy. I was drifting.

The decision journal

This is where the real change happened.

Cortex and I came up with something we call the Decision Journal. Every significant decision gets an entry:

  • What we’re deciding
  • The context at the time
  • What I chose and why
  • A future review date

It works because I have a habit of getting excited about ideas.

“I’m going to build this website for my friend.”

Then I spend hours designing, coding, reviewing. The next day I’m in my feelings and don’t want to touch it. Hours wasted.

Last weekend was the clearest example. I spent the whole weekend and $65 in Claude credits transitioning code from one language to another.

Ask me today why. I can’t tell you.

It was a waste. I had more important things to do.

If the Decision Journal had been fully active, I would have paused. Logged the decision. Thought about it for more than five minutes. I probably wouldn’t have started.

Why the pause matters

Last year I used ChatGPT to lose over 20 pounds. What actually worked wasn’t the meal plan. It was logging every meal.

That tiny pause before eating made me ask if the donut was worth it. If the candy bar was worth it. The logging created the friction. The friction created the discipline.

The Decision Journal does the same thing for how I spend my time and energy.

I’ve spent years learning advanced tools and keeping most of it to myself while watching other people build audiences on YouTube talking about things I figured out months earlier.

That stops now. I want to share more of what I’m building with Cortex, Obsidian, and Claude, the wins, the wasted weekends, and everything in between.