Vlad Dumitrescu vladdu

LLM Wiki

A pattern for building personal knowledge bases using LLMs.

This is an idea file, it is designed to be copy pasted to your own LLM Agent (e.g. OpenAI Codex, Claude Code, OpenCode / Pi, or etc.). Its goal is to communicate the high level idea, but your agent will build out the specifics in collaboration with you.

The core idea

Most people's experience with LLMs and documents looks like RAG: you upload a collection of files, the LLM retrieves relevant chunks at query time, and generates an answer. This works, but the LLM is rediscovering knowledge from scratch on every question. There's no accumulation. Ask a subtle question that requires synthesizing five documents, and the LLM has to find and piece together the relevant fragments every time. Nothing is built up. NotebookLM, ChatGPT file uploads, and most RAG systems work this way.

	/*
	This script attempts to identify all CSS classes mentioned in HTML but not defined in the stylesheets.

	In order to use it, just run it in the DevTools console (or add it to DevTools Snippets and run it from there).

	Note that this script requires browser to support `fetch` and some ES6 features (fat arrow, Promises, Array.from, Set). You can transpile it to ES5 here: https://babeljs.io/repl/ .

	Known limitations:
	- it won't be able to take into account some external stylesheets (if CORS isn't set up)
	- it will produce false negatives for classes that are mentioned in the comments.

	#!/bin/sh
	#
	# A (client-side) git hook script which warns if there are too long lines in
	# the commit message, not fitting with 50/72 formatting.
	#
	# To use this script, copy it as .git/hooks/commit-msg and give it an executable
	# file permission.
	#
	FILE=$1

	// Use Gists to store code you would like to remember later on
	console.log(window); // log the "window" object to the console