Kevin Ha kcha

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.

Install the necessary packages:

sudo apt-get install apt-transport-https ca-certificates
curl -s https://syncthing.net/release-key.txt | sudo apt-key add -
echo "deb https://apt.syncthing.net/ syncthing stable" | sudo tee /etc/apt/sources.list.d/syncthing.list
sudo apt-get update
sudo apt-get install syncthing
sudo apt-get install git

Start syncthing once:

tmux cheatsheet

As configured in my dotfiles.

start new:

tmux

start new with session name:

Step 0:

Get Homebrew installed on your mac if you don't already have it

Step 1:

Install highlight. "brew install highlight". (This brings down Lua and Boost as well)

	# The UID and GID of the user used to run paperless in the container. Set this
	# to your UID and GID on the host so that you have write access to the
	# consumption directory.
	USERMAP_UID=101
	USERMAP_GID=1024

	# Additional languages to install for text recognition, separated by a
	# whitespace. Note that this is
	# different from PAPERLESS_OCR_LANGUAGE (default=eng), which defines the
	# default language used when guessing the language from the OCR output.

	#' Finds the local maxima (peaks) in the given vector after smoothing the data
	#' with a kernel density estimator.
	#'
	#' First, we smooth the data using kernel density estimation (KDE) with the
	#' \code{\link{density}} function. Then, we find all the local maxima such that
	#' the density is concave (downward).
	#'
	#' Effectively, we find the local maxima with a discrete analogue to a second
	#' derivative applied to the KDE. For details, see this StackOverflow post:
	#' \url{http://bit.ly/Zbl7LV}.