yep .. I know, another note taking post
How I Take Notes
For me, taking notes is fundamental: it’s how I build knowledge and actually get things done over time.
I like the idea of leaving breadcrumbs for my future self. The small versions of me from 2019 still show up in my vault today, and their random ideas have slowly stratified in unexpected, non‑linear ways. That alone tells me that how I take notes matters and deserves to be thought through carefully.
I take notes in Obsidian, one of my favourite softwares. But this post is not about why Obsidian is better than everything else. It’s called “How I Take Notes” and not “How I Use Obsidian” for a reason: everything I describe here could theoretically be replicated with Notepad. The tool matters less once the habits are formed. The principles are what last.
That said, Obsidian helped me think in minimal terms and introduced me to a set of solid foundational habits:
- File over app : data ownership, accessibility, and longevity win over transient apps. The files you create are designed to outlast the software itself.
- Avoid vendor lock-in : your notes should be readable everywhere, even on offline hardware
- Keep it simple stupid : don’t spend time on organize stuff, just capture in a couple of folders → linking and basic tagging are the way.
- Zero friction on capture : if it’s hard to write down, you won’t.
“Building things, not migrating data.”
The Architecture
My vault is the simplest thing you can imagine. At the top level, just two folders:
kb/(Knowledge Base): thousands of atomic notes on anything (from specific concepts, to ideas or inputs like movies, books, articles,videos).projects/: where the work happens: anything related to projects, ideas, things that occupy my life for more time (from days to years).
The glue between everything is links like [link](other note) and tags like #input (but also nested tags like #input/book and #input/movie ).
My Journal (projects/journal/) is the most important thing in the vault. It acts as the temporal anchor for everything else. Every note I create links back to the daily note of its creation, so each atomic concept carries context: not just what I learned, but when, and what else I was doing that day.
(If you’ve read about Zettelkasten or PARA: yes, I’ve stolen from both. No, this isn’t either. I don’t have the patience for Luhmann’s numbering or Forte’s quarterly reviews. Two folders. That’s the system.)
Years of over-optimizing taught me that the best system is the one you actually use.
“You must have the discipline to do less.”
My plugins are few: Periodic Notes, Calendar, Full Calendar, Note Refactor, Timestamper. I call them “UX sugar” → none of them store data in a hidden database. Full Calendar, for example, just scrapes my markdown notes for events and renders them visually. Delete the plugin, and the data remains as plain text.
One habit I’m particularly attached to: whenever I modify an existing note, I add a --- separator at the bottom and write the new content below it. This gives me a linear history of my thinking on any topic, just by scrolling — no Git log needed.
# Atomic Note
[2021-01-23](2021-01-23.md)
Example Original thought from 2021.
---
[2024-05-05](2025-05-05.md)
New angle after reading X
---
[2025-08-20](2025-08-20.md)
Now I think the opposite, here's why.The hardest part isn’t the setup. It’s the discipline to stick to a methodology without constantly being tempted to “improve” it.
Nested Vaults
Under projects/ I have dedicated spaces for work, university, and other projects. These are technically separate vaults that live inside my larger Second Brain directory.
The separation is invisible but powerful. For example some advantages:
- At work I only open the work vault. No personal journal in view but still when I’m at home I see what I did at work, like conceptually I see work as a big subset of our life.
- My old university notes are a nested vault that is also a public GitHub repository, I can deploy them as a website without exposing my private journal. This blog post you are reading lives in
projects/blog. I run a script that picks it up, processes it with a framework Hugo (or Quartz), and deploys it. The website is just a mirror of a subfolder of my vault.
It’s not a bug, it’s a feature
I have been using a Remarkable tablet for years for pre-verbal thinking. But I don’t sync everything automatically. Once a year I move the PDF of my handwritten notes into Obsidian. That’s it.
I’m fully aware this is a limitation of my system. Handwritten notes aren’t searchable. They don’t link to anything. They sit in a PDF, opaque to every plugin and agent I use. I know. And I genuinely don’t care.
Yes, I could set up OCR pipelines to automatically transcribe everything into Markdown but I don’t give a shit. I will save my ass saying that the mess on my Remarkable shouldn’t be indexed, it should be filtered.
Not a bug, it’s a feature.
The Remarkable is where I think before I know what I think. I have an annual move from the Remarkable to Obsidian which forces me to quickly re-read months of raw thinking to see if anything still gives me inspiration. Most of it doesn’t, and that’s fine.
Syncing and Backup
A collateral benefit of keeping everything on the filesystem: trivial to back up to a NAS and encrypt in the cloud for long-term storage.
On AI Agents: Read-Only
Plain text wins again here. Pointing an AI agent at a folder of Markdown files is trivially simple no connectors bullshit, no exports.
Links and backlinks give agents a traversable graph of concepts, making the vault a powerful long-term memory layer. I’ve experimented with using them as thinking partners: turning quick captures into structured notes, annual reviews from journals, critiquing ideas, and extracting lessons from project retrospectives.
But after trying Opencode with various models, OpenClaw, and similar setups, I made a deliberate decision: agents read my vault. They don’t write to it.
That’s it. The vault is a knowledge base I feed into context, not a database that gets modified by something that isn’t me. Capture stays manual. Refactoring stays manual. The agent can be used as a tool to help me reason over my notes, or find connections I missed. But it doesn’t touch the files.
For now, this is the constraint I choose to keep. The goal isn’t a museum of automated artifacts:
- Intentionality over automation: automate the writes and you lose the cognitive engagement that makes the note valuable. The friction of manually refactoring a note is the filter.
- Simplicity scales : the more complex the system, the more fragile it becomes → a broken automation is worse than no automation.
- Productivity porn is a trap: chasing the perfect setup is procrastination.
No agent, plugin, or app is going to do the thinking for you. Obsidian taught me the principles, but the software isn’t the secret. Once the habits are formed, the tool matters less.