Obsidian Is Not a Note-Taking App. Here Is What It Actually Is.

I spent the first year using Obsidian comparing it to Notion.

Which one organises better. Which one has nicer templates. Which one syncs faster across devices. By every metric that matters for a note-taking app, the comparison was close, and most days Notion looked like it was winning.

I was asking the wrong question for over a year.

Obsidian is not competing with Notion. It is not a better or worse version of the same category. Judging Obsidian as a note-taking app is like judging a server as a filing cabinet. Technically you can store files in both. You are missing what the second one actually does.

Here is what Obsidian actually is, and the exact three things you need to build to make it that instead of a glorified notepad.

The Category Error

A note-taking app is built around one job: let a human write something down and find it again later. Search, folders, tags, templates. Every feature serves retrieval.

Obsidian has all of those features, which is exactly what makes the category error easy to fall into. It looks like a note-taking app if you only ever use the built-in editor.

But the foundational decision underneath Obsidian is about file format, not note-taking. Every note is a plain text markdown file, stored locally, with no proprietary structure underneath it.

That decision is unremarkable until you ask what else can read a plain text file. The answer is everything. Every AI model. Every script. No export, no API integration, no conversion step.

Notion stores your notes behind an API with a rate limit. Obsidian stores them as files an AI can read the instant you point it at the folder.

To be fair to Notion: it is genuinely better at relational databases, team collaboration, and structured views. This is not an argument that Notion is worse at everything. It is that for becoming a reasoning substrate, the architecture works against it.

The obvious objection: Notion supports markdown export, so could you not just export and get the same result? For a one-time snapshot, yes. The problem is a reasoning substrate needs to be live. A synthesis automation reading your vault every morning needs what you wrote yesterday, not what you exported three weeks ago through an API with rate limits. Native plain text removes a maintenance burden that export reintroduces.

One more precision: the plain text foundation is not unique to Obsidian. A folder of .md files with no app would technically work too. Obsidian’s actual advantage is the editing experience and plugin ecosystem that make it practical to sustain a vault large enough to be worth reasoning across.

What Obsidian Actually Is

Obsidian is a context layer for a reasoning system. Not storage. Not retrieval. The substrate an AI uses to reason in your specific voice, about your specific open questions.

A note-taking app, queried, returns what you wrote. Search “DeFi custody” and it shows three notes containing that phrase.

A reasoning substrate, queried, returns what your accumulated notes imply when read together. You ask what your vault reveals about the custody bottleneck, and the system reads every note touching the topic and produces an answer none of the individual notes contained alone.

This only works because the notes are plain text Claude can read directly without friction.

The Three Things That Make It a Reasoning Substrate

These are not concepts. These are the exact builds. Do all three and the category error fixes itself, because the vault will start behaving like something other than a note-taking app.

1. The CLAUDE.md File

This file tells Claude not what you have saved, but how you think. Without it, every session starts cold. With it, Claude already knows your reasoning patterns before you type a word.

It lives in a folder called 04-Claude inside your vault and gets uploaded to a Claude Project as permanent context.

CLAUDE.md Template

Who I Am [How you evaluate information. What you trust. What you are sceptical of. Not a biography — a reasoning profile.]# How This Vault Works 00-Inbox: raw captures, unprocessed 01-Sources: processed material, one note per source, no note counts as processed without a reaction paragraph 02-Ideas: my own positions and theses, no external sources 03-Projects: active research threads 04-Claude: this file plus everything you produce# What I Am Working On Right Now [Three active projects. For each: the specific open question you have not answered.]# What I Want From You Read my vault before answering anything. Surface connections I have not made. Flag contradictions between notes. Never pad. If you are uncertain, say so at the start, not buried at the end.# Hard Rules Primary sources over secondary sources. Always. Specific over general. Always. Contradiction over false consistency. Always.

Setup

1. Create CLAUDE.md in your 04-Claude folder, fill in the five sections above 2. Create a Claude Project at claude.ai 3. Upload CLAUDE.md as the first piece of project knowledge 4. Upload five of your most important existing notes as seed context 5. Start a conversation and ask Claude what it notices before asking it anything

Keep this file under 120 lines. Every line should change an output. If it does not, it does not belong.

2. The Daily Synthesis Automation

A note-taking app shows you what is there. This automation shows you what is implied by what is there.

It runs at 6am via N8N, reads the last seven days of vault notes, and writes a synthesis to your Inbox before you sit down.

N8N Setup

Node 1: Schedule Trigger → 6:00am daily Node 2: Read Binary Files → vault path, filter modified in last 7 days Node 3: Anthropic node → model claude-sonnet-4-6 System prompt: You are reading an Obsidian vault. Synthesise, do not summarise. User prompt: Read these vault notes from the last 7 days. Produce four sections: Connections (two non-obvious links between separately captured notes, name both notes, if the connection is obvious it does not qualify), Pattern (one theme across three or more notes, one sentence), Contradiction (two notes where positions conflict, quote both), Best capture (the single note most worth developing and why). Node 4: Write Binary File → 00-Inbox/brief-{{date}}.md

This is the output a note-taking app structurally cannot produce, because it requires reading across files simultaneously and reasoning about what they imply together, not retrieving what matches a search term.

3. The Contradiction Check

No note-taking app has ever asked whether your notes contradict each other. That is not a retrieval question. It is a reasoning question, and it is the one that most clearly proves the vault is doing something a filing system cannot.

It runs at 7am, reading your 02-Ideas positions against your 01-Sources captures from the last 30 days, looking specifically for conflict.

N8N Setup

Node 1: Schedule Trigger → 7:00am daily Node 2: Read Binary Files → 02-Ideas folder Node 3: Read Binary Files → 01-Sources folder, modified in last 30 days Node 4: Anthropic node Prompt: Here are my current thesis notes from 02-Ideas. Here are source notes from the last 30 days. Your only job is contradiction. For each idea, tell me whether any recent source note contradicts, complicates, or materially updates that position. Do not look for agreement. If there is no conflict, say so in one word: Clear. Node 5: Write Binary File → 00-Inbox/contradictions-{{date}}.md

A thesis that comes back Clear every morning is either genuinely well-supported or not being tested by what you are reading. Both are worth knowing.

The Test

You do not need all three running to check whether this is working. Ten minutes proves it either way.

Take ten notes you have already written. Add one honest sentence to each: your actual reaction, not a summary. Paste all ten into a single Claude conversation and ask it to find one connection between any two that you had not noticed.

If it finds something real, your vault is already closer to a reasoning substrate than a note-taking app. Build the three things above and that becomes automatic instead of manual.

If it finds nothing, the problem is not Claude. It is that your notes contain other people’s sentences, not your own thinking. Fix that first. Nothing else in this article works without it.

It was never trying to be a better Notion. It was always capable of being something a note-taking app cannot.

Follow @damidefi on X for daily Claude AI tools, crypto analysis, and the full journey to 100K. Bookmark this. Share it with one person still comparing Obsidian to Notion on folder structure.