totalizator

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.

Overview

Here's some docs on how I connected to OCTOI using a Raspberry Pi, an icE1usb, and my Cisco 2921 voice router.

Getting a config file

I contacted LaF0rge from #retronetworking on libera.chat. He asked me to email him at (redacted).

DEPRECATION NOTICE

Consider this gist to be deprecated. It's full of information that may or may not be correct, given recent findings. Consider the in-development PIDDatabase, and UMSKT, instead: https://umskt.github.io/PIDDatabase-viewer https://github.com/UMSKT/UMSKT

If you still want to see the old gist, see the last revision.

DigitalOcean does not provide a way to download a snapshot of your droplet locally. You can use rsync to accomplish this instead.

On your local machine, assuming you have added your-server in your SSH config:

rsync -aAXHv --append-verify --exclude={"/dev/*","/proc/*","/sys/*","/tmp/*","/run/*","/mnt/*","/media/*","/lost+found"} your-server:/

-a : archive mode (all files, with permissions, etc.)
-A : preserve ACLs/permissions (not included with -a)

Before starting, make sure you have a backup, and make sure to have a linux live boot ready to rescue your system. It's easy to mess this up!

Use gdisk to convert the partition table to GPT.

gdisk /dev/sda
Create the "BIOS boot" partition that GRUB needs.

n to create a new partition. Needs to be about 1MB. You can probably squeeze this in from sectors 34-2047. Use L or l to look up the code for "BIOS boot" (ef02).
Write the new partition table.

w
Reload the partition table. > partprobe /dev/sda

Making a UEFI bootable iPXE USB drive

Build the UEFI executable for iPXE

# First we'll clone iPXE
$ git clone git://git.ipxe.org/ipxe.git
# Go into the src directory of the cloned git repo
$ cd ipxe/src
# Compile the UEFI iPXE executable

	#!/bin/bash

	##############################################################################
	# An rclone backup script by Chris Tippett (c.tippett@gmail.com)
	#
	# Originally adapted from the great work by Jared Males (jaredmales@gmail.com)
	# https://gist.github.com/jaredmales/2f732254bb10002fc0549fa9aa0abdd7
	#
	# Copyright (C) 2020 Chris Tippett (c.tippett@gmail.com)
	#

	#!/bin/bash

	##############################################################################
	# An rclone backup script by Chris Tippett (c.tippett@gmail.com)
	#
	# Originally adapted from the great work by Jared Males (jaredmales@gmail.com)
	# https://gist.github.com/jaredmales/2f732254bb10002fc0549fa9aa0abdd7
	#
	# Copyright (C) 2020 Chris Tippett (c.tippett@gmail.com)
	#

	#!/bin/bash

	# Copyright (C) 2020 Kia
	# Licensed under Zero Clause BSD License


	if [[ $# -ne 3 ]]; then
	echo "usage: $0 (BMC I2C BUS NUMBER) (FAN PAGE) (DUTY CYCLE FLOOR)"
	echo ""
	echo "BMC I2C BUS NUMBER can be found as follows"

	<!doctype html>
	<title>Site Maintenance</title>
	<style>
	body { text-align: center; padding: 150px; }
	h1 { font-size: 50px; }
	body { font: 20px Helvetica, sans-serif; color: #333; }
	article { display: block; text-align: left; width: 650px; margin: 0 auto; }
	a { color: #dc8100; text-decoration: none; }
	a:hover { color: #333; text-decoration: none; }
	</style>