9 August 2026

Personal Knowledge Bases

A personal knowledge base is a store of what you have learned, kept in your own words, somewhere you can find it again. The form barely matters. A folder of text files qualifies, and so does a box of index cards.

I want to argue for keeping one. But the argument is narrower than the one the productivity world makes, so it is worth being precise about what holds up and what does not. The practice is old and useful. The industry around it oversells it. And most attempts fail in the same place: the system becomes the work.

An old habit with new software

Readers have kept commonplace books since antiquity, and working scholars ran card indexes long before software. The example everyone cites is Niklas Luhmann, the German sociologist whose Zettelkasten, a cabinet of about 90,000 linked index cards, is credited with powering seventy books and several hundred papers. His method became the template for a wave of tools in the early 2020s, Roam and Obsidian most visibly, along with the digital garden movement of public, slowly tended notes.

The legend needs some deflation. The Bielefeld University project that digitized Luhmann's actual cards makes them available to anyone, and decades of access to the method have produced no second Luhmann. He was an unusually driven writer who kept notes, not an ordinary writer with a magic cabinet.

The skeptics have a point

Brennan Brown asked a blunt question in What have note-taking PKMs accomplished, really?, and his answers are uncomfortable. The most visible output of the personal knowledge management scene is more personal knowledge management: courses about note-taking, videos about Obsidian setups, books about building second brains. The prolific writers held up as proof, Robert Greene and John McPhee among them, mostly work from disposable notes scoped to one project and start fresh for the next. And nobody has controlled evidence that elaborate note systems raise anyone's output.

Behind all of this sits what the Zettelkasten community itself calls the collector's fallacy: saving something feels like learning it. It is not. A clipped article changes nothing about what you know, and a thousand of them change nothing a thousand times.

Measured by published work, then, personal knowledge bases look like a hobby wearing productivity's clothes.

What that argument misses

Published work is the wrong measuring stick. Emacs never wrote a paper either. Tools do not produce anything; they change what a person can do, and most of what a knowledge base does for a person is private. The best response to Brown's essay spends several thousand words on this point, and I think it wins it.

There is also one mechanism here with real support: writing itself. Explaining an idea in your own words forces you to notice what you have not understood. The understanding is the product, and the note is close to a byproduct. Andy Matuschak holds that position despite being the person most cited on note-taking technique: better notes only matter in service of better thinking.

So the failures the skeptics document are real, but they are failures of system-building. People burn their attention on plugins, schemas, and folder taxonomies, and have none left for the subject.

What survives

Five habits hold up, and none of them requires special software.

Write, don't clip. A saved article is a bookmark. A paragraph in your own words about why the article mattered is knowledge. If something is not worth rewriting, it was not worth keeping.

Own your files. Plain text in ordinary files has outlived every note app so far and will outlive the current ones. Export-hostile tools hold your notes hostage precisely when you have accumulated enough to care.

Organize for retrieval, not filing. You will search your notes far more often than you will browse them. Filing schemes reward the librarian fantasy; findability rewards the writing. Link notes when a link is cheap and obvious, and skip the ontology.

Let projects pull. Notes exist to serve the things you make. A base that feeds nothing degrades into a museum, and museums are where the collector's fallacy retires.

Keep the system boring. Every hour spent redesigning the vault is an hour taken from the subject. Novelty belongs in the work.

The part that is changing

Knowledge bases have always been write-mostly. Reading your old notes back costs attention that almost nobody spends, and that gap is exactly where the collector's fallacy lives: things go in, nothing comes back out.

Software that reads is new. An assistant that lives with your files can actually read what you kept, surface the relevant note while you work, and notice that two of them disagree. Retrieval used to be the expensive half of the habit, and it is becoming the cheap half; that shift is a large part of why we built ROS as a drive of ordinary files with an assistant living on it.

The habit the machines reward is the same one the skeptics already recommended. Notes in your own words become more valuable, not less, because they are the ground truth the software reasons from. The caveat survives too: an assistant will cheerfully index a hoard of unread clippings, and garbage retrieved at scale is still garbage.

Where to start

Make a folder of plain text files. When you learn something worth keeping, write down what you understood in your own words, with enough context that a stranger could follow it, because in six months that stranger is you. Link when it helps. Delete freely; a smaller base you trust beats a large one you avoid.

Then watch for one signal. In any given week, if the system got more of your attention than the subjects inside it, the system has failed, whatever the productivity industry says.