Artificial Corner

Artificial Corner

I Made Claude Smarter With an Obsidian Second Brain

Build a second brain that Claude can read, write and keep tidy for you

Frank Andrade's avatar
Frank Andrade
Oct 11, 2026
∙ Paid

Click here to get my FREE Claude course (20+ lessons)

If you use Claude a lot, you’ve probably built a setup around it: skills, scheduled tasks, prompts you reuse every week. They tell Claude how to do your work.

But none of them have room for the rest of your job: the decision from last week’s meeting, the fix you spent a whole afternoon finding, the idea you had on a walk. So it ends up in a notes app, and that’s where it gets messy:

  • Apple Notes: Claude can read your notes, but it’s clumsy at editing them.

  • Notion: it connects to Claude, but the app gets complicated fast.

And in both, your notes live inside someone else’s app. There’s no folder you can open and move your notes somewhere else.

That’s the gap Obsidian fills. It’s a free note-taking app, but you barely take notes in it.

Claude does: it writes them, edits them and files each one in the right place. Your job is to drop things in and ask questions.

Today, when someone asks why a decision was made, you dig through old emails, or you guess. With Obsidian, Claude keeps all of that for you: a second brain for your work that stays organized and up to date.

It works like your own Wikipedia. The more you add, the more it knows.

In this guide, you’ll learn:

  • How to set up your Obsidian vault in a few minutes

  • The one prompt that turns your files into a wiki Claude writes and keeps up to date

  • How to keep your wiki growing with your new files and your chats with Claude

  • The health check that catches outdated docs before you act on them

  • A skill that reviews your wiki every week and flags what needs your attention, before it turns into a problem

Two things before we start:

1. Save this guide, and set up your vault today
2. Send it to a friend who still keeps their best ideas in a pile of Apple Notes.

PS: This newsletter grows mostly from people like you sharing it. Every share helps me keep writing guides like this one.

Share

What Obsidian is

Obsidian is a note-taking app, but under the hood it’s just a folder on your computer.

Every note you write is a plain text file that ends in .md (a markdown file). Obsidian calls the folder that holds them a “vault”. That’s it.

This is what makes it different from Notion or Apple Notes. There, your notes are locked inside someone else’s app. In Obsidian, every note is a file on your own computer, and any app can open and edit it (TextEdit, Word, Claude, etc).

I said it when Anthropic moved scheduled tasks to the cloud: what you build with AI is only yours if it lives in files you own. Those files are assets.

Repeat after me: own your files, rent the AI.

Obsidian itself is free. The only paid part is Obsidian Sync ($5/month), which puts the same vault on your phone. You most likely won’t need it. I pay for it, but I rarely use it, because Claude does most of the writing and editing on my Mac.

Set up your vault

Here’s how to go from zero to a vault you can use.

Create a vault

  1. Download Obsidian from obsidian.md and open it

  2. Next to “Create new vault”, click “Create”

  1. Give your vault a name, and click “Browse” to pick where it lives on your computer

  2. Click “Create”

Pick a place you’ll remember. You’ll need this folder again when you connect Claude.

Write your first note

Click the new note icon (or press Cmd+N on a Mac, Ctrl+N on Windows) and start typing. Obsidian saves as you go, so there’s no save button.

Obsidian notes are written in markdown, which uses a few simple symbols for formatting:

  • # at the start of a line makes a heading

  • ## makes a subheading

  • - makes a bullet

Link your notes

This is the feature Obsidian is famous for.

Type two square brackets [[ and start typing the name of another note. Obsidian suggests matching notes, and you pick one. Now the two notes are linked, and you can click from one to the other.

See your notes as a graph

Open the command palette (Cmd+P on a Mac, Ctrl+P on Windows), type “graph” and pick “Open graph view”. Every note is a dot, and every link is a line between two dots.

Here’s my work vault. The clusters are the projects I keep notes on: my reels, my Substack, my notes on Claude. You see which ideas connect without opening a single note:

Click here to get my FREE Claude course (20+ lessons)

Get a deep-dive guide every week, plus all my video courses.

Connect Claude to your vault

A quick note on names: Anthropic recently merged chat and Cowork. The Claude app now only has two sides, which you switch with the icons at the top left: regular Claude (the speech bubble), which is basically Claude chat + Cowork, and Claude Code (the </> icon).

Here’s how to give regular Claude your vault:

  1. Open the Claude desktop app and click the speech bubble at the top left, so you’re in regular Claude (it works the same with Claude Code though)

  2. Under the message box, click “Project or folder”, then “Add folder”

  1. Choose your vault’s folder and click “Open”

  2. Claude asks if it can change files in that folder. Click “Always allow”, so it doesn’t ask every time

Now Claude can read every note in your vault, edit them and create new ones. You don’t upload anything.

Karpathy’s viral method: a second brain Claude writes for you

Andrej Karpathy, one of the best-known AI researchers, shared an idea earlier this year that went viral: let the AI keep your notes for you. He calls it an LLM wiki.

It works with two Obsidian folders:

  • raw: what you collect. Articles, PDFs, meeting notes, transcripts. You add files here, and nobody edits them.

  • wiki: the notes Claude writes from them. One note per idea, person or tool, all linked to each other.

Every time you add a file and tell Claude, it reads the file and updates every note it touches. A normal chat forgets everything when it ends. This wiki keeps growing.

Say you save your work notes there for a few months: meeting notes, project docs, the emails that matter. Claude builds a note for each project, each person you work with and each decision. When a June meeting reverses what the team agreed in March, it flags it. When your performance review comes, you ask “What did I get done this year, and where’s the proof?” and Claude lists your wins with links to the notes that back them up.

It works the same when you’re learning a skill for your next career move. Every lesson, article and video transcript goes in raw, and Claude turns them into one note per concept, linked to everything it connects to.

Here’s mine after my first run. Nobody wrote these notes by hand:

The rest of this guide is for paid subscribers. It’s every step I followed, with every prompt I pasted, so you can copy it instead of building it from scratch.

Below the line:

  • A wiki that writes itself: Meeting notes, important emails, docs and course notes pile up, and nobody opens them again. Now you drop them in a folder, and Claude files every idea into linked notes. Plus the one prompt that builds it for you, and how to save every important chat to it, so a lesson you learned the hard way never has to be learned twice.

  • A health check that finds what’s out of date: Old docs keep getting shared long after they’re wrong. I created a prompt that finds every outdated doc and what replaced it, while skipping the changes that don’t matter.

  • How to use your second brain for your career: The questions to ask your wiki before a performance review, a one-on-one, a raise conversation and more, so you walk in with the facts and the proof.

  • A skill that reviews your wiki every week: What slips at work usually shows up when it’s too late to fix. So I created a skill that reviews my wiki every Monday and shows me what needs my attention, before it turns into a problem.

Past this line, the full setup is waiting:

Build your wiki with one prompt

I built mine on my own work: the guides I publish in this newsletter. Yours could be your meeting notes and project docs, or everything from a course you’re taking.

Here’s how to set it up:

  1. In your vault, create a folder called raw

  2. Put the files you want in it. Markdown works best, but PDFs and text files work too

  3. Open a new chat in regular Claude, and add your vault as the folder (the same steps as before)

  4. Paste the prompt below and send it

Here’s the template, fill in the part in brackets:

Read Andrej Karpathy’s LLM Wiki idea here

Set it up in this folder. My [what your files are, like saved articles or meeting notes] are in raw/. Create the wiki folder, the index, the log and a CLAUDE.md with the rules, then add every file in raw/ to the wiki. Keep it simple.

The last part of the prompt (”add every file in raw/ to the wiki”) turns the files you already have into the first batch of notes. You only do this once. The next section covers the files you add after that.

Here’s mine, with my newsletter guides:

Right after you send it, a second window asks if Claude can use the folder. Tick “Don’t ask again for this folder on this device”, then click “Allow”:

When it finished, Claude listed what it made:

  • CLAUDE.md: the rules Claude follows when it works on the wiki

  • index.md: every note with a one-line summary. Claude reads it first when you ask a question

  • log.md: what was added and when

  • sources: one summary note per file in raw

  • topics: one note per idea, tool or person that shows up in more than one file

CLAUDE.md is the most important file of the five. It’s the manual for your wiki: what goes in raw, how notes are named and linked, how to add a new file, how to flag a contradiction. Without it, every new chat starts from zero, and Claude files things its own way each time. With it, the wiki stays consistent no matter which chat you’re in.

Read the CLAUDE.md that Claude writes. It’s yours to change, and you’ll add rules to it as you use the wiki.

There’s one catch. Claude Code reads CLAUDE.md on its own every time it starts in a folder. In regular Claude, that’s not guaranteed: in one of my tests, a new chat ignored it until I asked. So if you work in regular Claude, start each chat with one line:

Read CLAUDE.md first and follow it.

Keep feeding your wiki

Setting up the wiki takes one prompt. Keeping it useful is a habit: everything new goes in raw, and Claude adds it to the wiki. Here’s what to feed it:

New files

When you have something new, drop it in raw and tell Claude:

I added [file name] to raw/. Add it to the wiki.

Claude reads it, writes its summary note, and updates every note it touches. When I added a report, it created a new topic note and updated three others:

Say you just had a client call. You drop your notes from it in raw and ask Claude to add them. Claude writes a summary note for the call, adds the client’s new request to the project’s note, and adds what each person promised to their own note: a coworker sends the pricing by Friday, you send the draft on Monday, etc.

A week later, before the next call, you ask Claude “What has this client asked for so far, and what did we promise?” and get the answer from every call in seconds.

Tip: For web articles, the free Obsidian Web Clipper extension saves any page to raw in one click (set the folder in its settings).

Your chats with Claude

Some of your best thinking happens in chats with Claude: you weigh two options, pick one and decide what to do next. Then the chat ends, and it’s gone. It’s like a meeting with a coworker where nobody took notes.

So when a chat ends with a decision, treat it like a meeting: save a summary in raw, and let Claude add it to the wiki like any other source.

I do this every time I go down a rabbit hole. When I spend an hour with Claude figuring out why something broke, I ask it to write down what went wrong and how we fixed it, so I never go through it twice. Hard-earned lessons deserve to be written down.

Think of the afternoon you spent working out why the numbers in the monthly report didn’t match, or the chat where you planned how to ask for a raise (the wins you’ll mention, what you’ll say if your manager says “not this year”). Save it once, and the next time you ask, Claude has the answer in seconds.

Here’s how you can do it:

Option 1: paste a prompt at the end of the chat

Copy this and paste it as your last message:

Before we finish, save this conversation to the wiki:

  1. Write a short summary in raw/, named with today’s date and the topic. Include what we decided and why, the options we dropped, and any open questions or next steps. Skip the back-and-forth.

  2. Add it to the wiki like any other source: update the notes it touches, and flag anything that changes an earlier decision.

  3. Tell me which notes you created or changed.

It works every time, but you have to keep the prompt somewhere and paste it.

Option 2: teach it once, then say “save this chat to the wiki”

This is what I do. Ask Claude to add the rule to your wiki’s CLAUDE.md, once:

Add this rule to CLAUDE.md: when I say “save this chat to the wiki”, write a short summary of this chat in raw/ (what we decided, why, and what we dropped), named with today’s date and the topic. Then add it to the wiki like any other source, and tell me which notes you changed.

From then on, end every chat that matters with one line:

save this chat to the wiki

Claude reads the rule and does the rest.

Ask your wiki anything

Start a new chat with your vault as the folder, and ask the way you’d ask a coworker to remind you of something. No special wording.

Here are some questions I asked my own vault over the past days:

  • Tell me what my latest Substack report found

  • Does my Obsidian vault say anything about a test I’m running on Threads?

  • Can you read the latest report on my Instagram reels?

Claude found the notes and answered from them, the same way a coworker who kept your notes would.

Here’s how I’d use it for your career:

  • What did I get done this year, and where’s the proof?

  • I’m asking for a raise next week. Which wins and numbers should I bring up?

  • What’s still open on [project], and who’s waiting on me? (every Monday)

  • What did we decide about the budget in the last meeting?

  • Who was supposed to send the pricing to the client, and did they?

  • What did my manager say last time we talked about my promotion?

  • Did I already write down the feedback from my one-on-one with my manager? Where?

  • What did the client complain about last time, and did we fix it?

  • I’m meeting [name] tomorrow. What did we talk about last time?

  • Which decisions on [project] are still waiting on someone?

When the answer pulls together many notes, like a year of meetings or a comparison, add one sentence at the end:

Save the answer as a new note.

A quick lookup doesn’t need it, because the answer is already in your notes. A combined answer isn’t, and saving it means you don’t redo the work next time.

Run a health check

The more files you add, the more notes repeat each other, contradict each other or go out of date.

Once a week, ask Claude:

Do a health check of the wiki. Find notes that contradict each other, claims a newer source has replaced, and topics mentioned often that have no note yet. Search the web for any tool or vendor a note relies on. Only report what would mislead me if I followed it today. Don’t change anything. Give me a short list, and I’ll decide.

The last line is the one that matters most. Without it, the list fills up with changes that don’t affect anything.

Picture the docs your team keeps passing around: the onboarding guide, the steps to get an expense approved, the pricing sheet. Add them to raw along with your meeting notes. When a meeting changes a process, the health check points to the doc that still describes the old way, and to the meeting that changed it, before a new hire follows the wrong steps. A renamed folder or a new logo won’t show up, since neither changes what anyone has to do.

Running this every week is easy to forget, so I made a skill that does it for you, and more.

A skill that reviews your wiki every week

Your wiki grows every week, and so do the things that slip through it: a decision reversed in a meeting you missed, a deadline that moved, a promise nobody kept. You’re not going to reread every note to find them.

This skill does it for you. Every time it runs, it writes one review note into your wiki with two parts:

This post is for paid subscribers

Already a paid subscriber? Sign in
© 2026 Frank Andrade · Privacy ∙ Terms ∙ Collection notice
Start your SubstackGet the app
Substack is the home for writing, podcasting, and media founders.