LLM can be powerful allies and they can do amazing things to help you with knowledge management.
The knowledge base problem. You know that moment. You are sure you wrote something about an idea. You search. One query returns unrelated notes. Another query returns nothing useful. You wanted to share this idea to your LLM to start a new research project. Now you have to type the idea by hand in the chat window. The note is still there, somewhere in the knowledge base. But, it is lost.
If you already work with an LLM. Don't take on yourself to maintain the knowledge base. LLM changes the problem. A knowledge base now has two readers: the person and the machine. The Markdown folders behind Obsidian and Andrej Karpathy's LLM wiki show the same thing: plain Markdown files work for both. dimnesis starts from that collision between knowledge management and AI. It sees two knowledge bases, yours and the LLM's, that work together and stay separate.
dimnesis helps knowledge workers combine knowledge management with an understanding of AI, so they collaborate better with it. The tools are simple to use, light on your computer and modular. They work with any knowledge base made of Markdown files and folders. Your knowledge stays on your own machine.
What dimnesis believes: local-first, LLM-agnostic, Markdown-native, MIT licence, and a human in the loop. The human has the last word. What dimnesis refuses: to take control of your information, or to keep you captive.
$ pkms-lint ~/notes 0 errors · 3 warnings
At work, I am a Product Owner in software development. In my free time, I am a cross-disciplinary researcher. My topics are knowledge management, AI and LLM best practices.
I have my own personal knowledge management problems, and I don't think I am the only one. So I decided to share my solutions.
Being a Product Owner shapes how I build them. My job is to find the problem worth solving before anyone writes code, to listen to the people who use the product, and to deliver small pieces that work. Each dimnesis tool starts from a real problem, does one job, and gets better with your feedback.
If a tool solves a problem you have, or misses one, write to me.
Clip any part of any web page into a library inside Chrome, then export it as Markdown for NotebookLM, Perplexity, or any LLM.
Personal knowledge management is the practice of collecting, organizing and reusing what you learn.
A good system makes the next search short. You capture a note once, you give it a place, and you find it again when you need it.
Tools change. The habit stays: capture, connect, review.
A good system makes the next search short. You capture a note once, you give it a place, and you find it again when you…
--- tag: PKM exported: 2026-11-01 clips: 1 --- # PKM --- ## Personal knowledge management *Saved: 2026-11-01 14:32* [Source](https://example.org/notes/personal-knowledge-management) *Tags: PKM* A good system makes the next search short. You capture a note once, you give it a place, and you find it again when you need it. ---
Fully compatible with Obsidian, Foam and Karpathy’s LLM Wiki.
Your notes stay on your computer. The reader opens your folder and shows your notes. It never writes to a file, so it cannot move, rename or change a note. There is no account, no cloud sync and no server. The reader makes no network request of any kind.
Choose the folder that holds your Markdown notes. You do this once.
Chrome will ask for permission. On the first grant, Chrome offers a single Allow — accept it.
Nothing leaves your machine. The reader never writes to your files.
Grant vault
2 notes
Choose a note in the tree.
An assistant keeps more than one kind of memory. Each tier trades speed for size.
The fast tier is the context window. It holds what the model reads now.
The slow tier is your notes. retrieval brings the right note back when you need it.
The context window is the text that a model reads in one session. It is measured in tokens.
When the session ends, the window is empty again. Nothing carries over unless you write it down in a note.
A prompt cache keeps a long prefix, so the model can reuse the same tokens in the next request.
It saves time and money. See context-window.
Tools change. The habit stays: capture, connect, review. This note keeps the three steps short.
Write the idea down when it comes. One idea per note. Give the note a title that you can search for later.
Link the new note to one note that you already have. Use a memory-tiers link or a plain Markdown link.
Once a week, open the notes of the week. Follow their links. Keep what is still true, and add what is new.
A review is short. Ten minutes is enough to find what you wrote and to see what is missing.
The extension can view the files in notes during this visit.
Find broken links in a Markdown vault. Wikilinks and reference-style ID links. Point it at any folder of Markdown files. It needs no index file, no plugin, and no configuration. It never writes to your notes unless you pass --fix.
uv tool install pkms-lintCopiedOr pipx install pkms-lint · pkms-lint on PyPI →
Product Owner and cross-disciplinary researcher, I explore information science, design, and AI. On Substack, I write about human-AI collaboration for knowledge work.
Subscribe on SubstackA question, a bug report, an idea for a tool. Write to me, and I will answer.