Bernhard Wagner bwagner

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.

Given a subscribed calendar with a url like

https://example.com/example.ics

To force Google Calendar to refresh and reload the contents right now, unsubscribe from the calendar and subscribe to a new calendar with a URL like

https://example.com/example.ics#1

Adding the anchor tag will force Google Calendar to think of it as a new calendar

	"""
	The most atomic way to train and run inference for a GPT in pure, dependency-free Python.
	This file is the complete algorithm.
	Everything else is just efficiency.

	@karpathy
	"""

	import os # os.path.exists
	import math # math.log, math.exp

	"""
	Created on Sep 24, 2013

	@author: RandomHardcoreJerks

	Requires pywin32


	original: http://code.activestate.com/recipes/474121/
	# HtmlClipboard

	#!/bin/bash

	if [ "$1" = "-h" -o "$1" = "--help" -o -z "$1" ]; then cat <<EOF
	appify v3.0.1 for Mac OS X - http://mths.be/appify
	Creates the simplest possible Mac app from a shell script.

	Appify takes a shell script as its first argument:

	`basename "$0"` my-script.sh