thednp

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.


	const pubSub = (() => {
	const topicSubscriberMap = new Map();
	const throwError = (topic) => {
	throw new Error(`
	No ${topic} named topic found in the registry.
	Available topics:- ${JSON.stringify(getAllTopic())}
	`);
	};
	const getAllTopic = () => Array.from(topicSubscriberMap.keys());

	/**
	* Credits
	* @Bkucera: https://github.com/cypress-io/cypress/issues/2839#issuecomment-447012818
	* @Phrogz: https://stackoverflow.com/a/10730777/1556245
	*
	* Usage
	* ```
	* // Types "foo" and then selects "fo"
	* cy.get('input')
	* .type('foo')

	/--------------Library code----------------/
	/**
	* BezierEasing - use bezier curve for transition easing function
	* by Gaëtan Renaudeau 2014 – MIT License
	*
	* Credits: is based on Firefox's nsSMILKeySpline.cpp
	* Usage:
	* var spline = BezierEasing(0.25, 0.1, 0.25, 1.0)
	* spline(x) => returns the easing value \| x must be in [0, 1] range
	*

	// Gist adapted from: https://gist.github.com/cms/369133
	export getStyle(el: Element, styleProp: string): string {
	let value;
	const defaultView = el.ownerDocument.defaultView;
	// W3C standard way:
	if (defaultView && defaultView.getComputedStyle) {
	// sanitize property name to css notation (hypen separated words eg. font-Size)
	styleProp = styleProp.replace(/([A-Z])/g, '-$1').toLowerCase();
	return defaultView.getComputedStyle(el, null).getPropertyValue(styleProp);
	} else if (el['currentStyle']) { // IE

	{
	"alignleft": {
	"selector": "p,h1,h2,h3,h4,h5,h6,td,th,div,ul,ol,li,table,img",
	"classes": "left"
	},
	"aligncenter": {
	"selector": "p,h1,h2,h3,h4,h5,h6,td,th,div,ul,ol,li,table,img",
	"classes": "center"
	},
	"alignright": {

	// Run from the dev tools console of any Youtube video
	// Accurate as of July 2, 2020.
	//
	// Copy and paste this into the dev console in a browser with the desired video loaded.
	//
	// NOTE: Some Youtube videos do not directly expose the video url in the response.
	// This script doesn't currently attempt to handle those. It will work for most other general video types though.

	(async () => {
	const html = await fetch(window.location.href).then((resp) => resp.text()).then((text) => text);

	<?php
	/**
	* @package Joomla.Site
	* @subpackage Layout
	*
	* @copyright Copyright (C) 2005 - 2015 Open Source Matters, Inc. All rights reserved.
	* @license GNU General Public License version 2 or later; see LICENSE.txt
	*/

	defined('_JEXEC') or die;

	var touchstartX = 0;
	var touchstartY = 0;
	var touchendX = 0;
	var touchendY = 0;

	var gesuredZone = document.getElementById('gesuredZone');

	gesuredZone.addEventListener('touchstart', function(event) {
	touchstartX = event.screenX;
	touchstartY = event.screenY;