I Connected Claude to My Obsidian Vault. 2 Months Later It Knows My Thinking Better Than I Do.
Eight weeks ago Claude surfaced a connection I had completely missed.
I had a note from six weeks prior about how attention scarcity was driving premium pricing in crypto. Sitting three folders away was a capture from a podcast about how scarcity mechanics in gaming drove in-app purchase behaviour. I had never linked these two ideas. They lived in completely separate parts of my brain.
Claude found them in 11 seconds and wrote a one-paragraph synthesis that became one of my best-performing posts.
That moment changed how I think about what a second brain is actually supposed to do.
The Problem With Every Obsidian Setup You Have Read About
Most Obsidian guides teach you how to organise information. Better folder structures. Smarter tagging systems. Prettier dashboards.
None of that is the point.
Organisation is just a faster way to find something you already know exists. It does not create new knowledge. It does not surface what you forgot. It does not connect ideas that live in different parts of your vault.
After two months of running Claude as the intelligence layer on top of Obsidian, I now have a vault that does all three automatically. Every morning before I open anything else, Claude has already been through my recent captures and produced a synthesis of what it noticed. Not a summary. An actual output: connections I missed, patterns forming across weeks of notes, questions worth thinking about that day.
The difference between an organised vault and an intelligent one is the difference between a library and a researcher.
The Four Layers
Everything runs on four components. Remove any one of them and the system degrades into a slightly fancier note-taking app.
Layer 1: Capture (Zero Friction)
The system dies if capture requires effort. Every time you have to think about where something goes, you are burning willpower that should go toward thinking about the content itself.
Readwise handles everything I read. Every highlight from articles, books, PDFs, and newsletters flows automatically into Obsidian via the Readwise Official plugin. For voice notes, Whisper transcribes them on my phone before they hit the vault. For X and web content, a Telegram bot catches anything I forward to it.
The rule: if saving something takes more than three seconds, the friction is too high and the system will fail over time.
Layer 2: Automation (N8N as the Router)
N8N runs in the background connecting everything together. It takes daily Readwise output and formats it into structured notes. It runs a nightly sweep categorising inbox items. It triggers the Claude synthesis every morning at 6am.
N8N is free to self-host. The entire automation layer costs nothing to run once it is set up.
Layer 3: Memory (Obsidian as the Context Layer)
Obsidian is not the intelligence in this system. It is the memory. The distinction matters.
Everything is organised by note type, not by topic. Six subfolders inside Captures: observations, reactions, patterns, questions, numbers, and references. Every note goes into one of these based on what kind of thinking it represents, not what it is about.
When you organise by topic, a note about attention economics in crypto and a note about attention mechanics in gaming live in completely separate folders and never encounter each other. When you organise by type, they both land in the patterns folder and Claude can find the connection between them.
The CLAUDE.md file lives at the top of the vault and tells Claude everything it needs to know about who I am and how to reason about my notes.
Layer 4: Intelligence (Claude as the Cognitive Partner)
Most people use Claude with no context. They open a new chat, ask a question, get an answer, close the tab. The output is as good as whatever they typed in the prompt box.
With the vault connected, Claude is not answering a question. It is reasoning across months of accumulated thinking. It knows which ideas I keep returning to. It knows which questions I have not resolved. It knows which claims I have saved that contradict each other.
The CLAUDE.md File
This is the most important file in the entire system.
CLAUDE.md teaches Claude how to think alongside you specifically. Without it, every synthesis Claude produces is generic. With it, the outputs are calibrated to your thinking, your projects, your voice, and your unresolved questions.
Key sections: Who I Am / What I Am Building / How This Vault Works / My Thinking Style / What I Want From You / Hard Rules.
Hard Rules example: “Do not produce generic insight. Every output must contain at least one connection traceable to a specific note. If you are pattern-matching to general knowledge rather than my notes, say so.”
The Daily Synthesis Prompt
Runs every morning via N8N at 6am. Passes the last 7 days of captures to Claude. Returns four sections:
- Connections — non-obvious links between separately captured notes
- Patterns — themes appearing across 3+ notes
- Contradictions — notes where stated positions conflict
- Open Questions — questions appearing repeatedly without resolution
“Do not produce summaries. Every item must trace back to a specific note I actually captured.”
What Changes After Eight Weeks
At two weeks: connections catch maybe once every three sessions. At four weeks: genuinely surprising connections. At eight weeks: it knows which arguments I keep returning to, which questions haven’t resolved, which domains I’m pulling from simultaneously.
The most useful feature at eight weeks: the contradictions section. I have updated my thinking on three significant positions because Claude surfaced notes that directly conflicted.
The accumulated context of someone who started six months ago — that is the actual moat in the AI era. Not the tools. Not the prompts.