Andrea Gallegati andreagalle

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.

Workbench

We extract our test repos from this small Zip file.

main is a "container" repository with its working copy,
plugin is a "shared" repository with its working copy,
remotes emulates remote bare repos for both, to better resemble regular usage.

The idea is to use plugin as a subtree of main in a vendor/plugins/demo path, and allow maintenance both ways:

	#!/bin/bash
	IP=$1
	first_port=$2
	last_port=$3
	function scanner

	{
	for ((port=$first_port; port<=$last_port; port++))
	do
	(echo >/dev/tcp/$IP/$port)> /dev/null 2>&1 && echo "$port open" #\|\| echo "$port closed"

	package PasswordEncryption;

	/**
	*
	* @author daniel
	*/
	import java.io.Console;
	import java.io.IOException;
	import java.security.*;
	import java.io.ByteArrayOutputStream;

	#!/bin/bash

	db_n=9

	#make intel_debug_fast

	exe_name=pijet-4ea9e50

	dbt_name="db${db_n}-${exe_name}"

	#find . -name "*.mp4" -print0 \| du -sb --files0-from=- \| awk '{ total += $1} END { print total/1024/1024/1024 }'
	find data/vtk/ \( -name "b_mean.vtk" -o -name "b_pdf.vtk" \) -print0 \| du -sb --files0-from=- \| awk '{ total += $1} END { print total/1024/1024 }'