I have a problem. My phone has four note apps. My laptop has three more. My work machine has two. I once spent a weekend setting up a self-hosted wiki on a Raspberry Pi that has been offline for 14 months. I have a Notion board I open once per quarter to feel bad about.
This is not a guide from someone who has it figured out. This is a guide from someone who has tried everything, broken most of it, and kept a few pieces that actually survived contact with real life. If you are drowning in browser tabs and half-written notes, you are my people.
What I actually learned after cycling through more knowledge management systems than I can count: the tool matters way less than whether you have a method, and the method matters way less than whether you actually use it.
Why I Kept Breaking Every System
I used to think knowledge management was about finding the right app. I tried Roam (too expensive), Notion (too slow), Evernote (too 2014), OneNote (too Microsoft), TiddlyWiki (too much of a rabbit hole), and plain text files in a git repo (too lonely). Each lasted about three weeks before I found something wrong with it and started hunting for the next fix.
It was never about the app. I was treating knowledge management as a storage problem when it is actually a digestion problem. Collecting is easy. I can dump 50 bookmarks into a folder in 30 seconds. Getting anything back out that is useful, connected, and not just a wall of text I wrote while half asleep. That is the hard part.
No tool will do the thinking for me. I had to accept that. The best a PKM system can do is not get in the way when I am actually processing an idea.
The Methods That Did Not Die on Me
After burning through a dozen apps, what survived was a set of patterns rather than any single tool. Here are the ones that stuck.
Atomic Notes (The One Rule That Actually Matters)
The single most practical thing I learned: write one idea per note. Not one article per note. One idea. If a note has two ideas, split it. This sounds stupidly simple but it changes everything, because suddenly you can link notes together without feeling like you are stapling shipping containers to each other.
I fail at this constantly. About 40% of my notes are still firehoses of unrelated thoughts. But the ones that follow the one-idea rule are the ones I actually find and use months later.
The Actionability Filter
Here is a question that saved me: "What am I going to do with this?" I sort everything into four buckets, borrowed and butchered from someone smarter:
- Active work. Things with a finish line. A blog post, a PR, a decision I need to make.
- Ongoing responsibilities. Things without a finish line. My team, my health, my homelab.
- Reference. Things I might need someday. API docs, architecture diagrams, that one Slack message about database migrations.
- Archive. Things I finished or abandoned.
The thing I learned the hard way: if you are not clear on which bucket something goes into, it is probably not worth capturing at all. I spent years bookmarking articles I was never going to read. The filter is about being honest about what you will actually act on, not about organizing better.
Linking That Does Not Turn Into Noise
Bidirectional links between notes are great until you link everything to everything and your graph looks like a plate of spaghetti. I learned to ask: "Will I actually follow this link six months from now?" If the answer is no or maybe, I skip the link.
The notes I find most valuable are the ones where I wrote a short connection statement, not just a link. Instead of linking Note A to Note B with no context, I write: "Note A contradicts the approach in Note B because of X." That sentence is worth 50 raw links.
The Note Lifecycle I Accidentally Fell Into
I did not plan this, but over time my notes started following a natural progression:
- Capture. A messy thought, a quote, a link, dumped somewhere fast.
- Clarify. Rewrite it so I understand what I actually meant.
- Connect. Link it to existing notes or tag it with a context cue.
- Use. Turn it into something: a post, a decision, a fix.
Most notes die at step one. I accepted this. The ones that survive to step four are the only ones that matter.
The Tools I Keep Coming Back To
I have tried to quit both of these. I keep reinstalling them.
Obsidian (When I Want to Actually Think)
Obsidian won for me because it stores plain Markdown files that I own. There is no export anxiety. If Obsidian shuts down tomorrow, I still have a folder full of .md files that every text editor on earth can read. That peace of mind is worth more than any feature.
The catch: Obsidian is a construction site, not a finished house. You have to set it up, configure plugins, decide on a folder structure. If you like tweaking more than writing, Obsidian can become a black hole of configuration. I know. I have spent entire weekends building the perfect vault from scratch, then abandoned it on Monday.
Logseq (When I Just Want to Write)
Logseq is an outliner. It handles daily notes better than Obsidian out of the box. If your workflow is journal-first, you log what you did today and link outward from there, Logseq feels natural. It is also fully open source, which I respect.
The tradeoff: Logseq stores plain Markdown .md files, but in an opinionated outliner flavor (block bullets, property:: value lines, block refs) that is less portable than Obsidian's prose Markdown, so you are still somewhat tied to the tool. And the fork in the road is now real: the SQLite-backed 2.0 beta shipped in July 2026, and in that version the database is the source of truth rather than a folder of Markdown files. If plain files on disk are the thing you care about, that is a version to watch carefully before upgrading. Also, the outliner paradigm is love-it-or-hate-it. If you think in outlines, great. If you think in documents, you will fight it.
When I Need a Shared Wiki
For team knowledge bases, I keep coming back to the lightweight, no-database approach. DokuWiki uses plain text files and runs on almost any server. It is old school. It is not pretty. But it has never broken on me, and I cannot say that about any of the fancier options.
On AI and Knowledge Systems
I have strong opinions on this topic, all of them probably wrong.
Why RAG Is Not the Answer to Everything
Retrieval-augmented generation is everywhere right now. You dump your documents into a vector database and ask questions. It works surprisingly well for factual lookups.
But RAG has a limit that I keep hitting: it retrieves chunks, not understanding. If your question requires synthesis across multiple documents or reasoning about implicit structure, RAG gives you a plausible-sounding answer that might be completely wrong. I have had it confidently cite a note that contradicted the actual answer.
The alternative approach that interests me: instead of retrieving at query time, compile and synthesize at ingest time. Build structured knowledge pages ahead of time. This is more work upfront, but the output is more reliable because a human (or a careful AI) has done the linking and reasoning before the question is asked.
Retrieval vs. Representation
Most knowledge systems optimize for retrieval: how fast can you find something? But they under-invest in representation: how well does the system capture the structure of what you know?
I think this is backwards. If you represent knowledge well (clear notes, meaningful links, consistent structure), retrieval becomes easy. If you represent knowledge poorly, no amount of search magic will fix it because you are searching through noise.
What I Wish Someone Had Told Me
If we are being honest, here is what I wish I had heard three years and fourteen apps ago:
- Start with one method, not one tool. Pick atomic notes or the actionability filter or whatever makes sense. Use a text editor. See if the method sticks.
- The best system is the one you actually use. I know this is cliche. It is also true. I have a beautiful, elaborate Obsidian vault that I never open. I have a single text file in my home directory that I edit daily. The text file wins.
- Capture less, process more. Most of what we save, we never use. The bottleneck is attention, not storage.
- Abandon things without guilt. I have 47 half-finished systems in a digital graveyard somewhere. I stopped feeling bad about this. Starting something and dropping it taught me what did not work. That is not failure. That is research.
I am going to go close a few of those 14 apps now. Probably keep two. See you next time.