Kevin McBride krmcbride

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.

The fastest remote directory rsync over ssh archival I can muster (40MB/s over 1gb NICs)

This creates an archive that does the following:

rsync (Everyone seems to like -z, but it is much slower for me)

a: archive mode - rescursive, preserves owner, preserves permissions, preserves modification times, preserves group, copies symlinks as symlinks, preserves device files.
H: preserves hard-links
A: preserves ACLs

	float ease(float x) {
	return pow(1.0 - x, 10.0);
	}

	float sdBox(in vec2 p, in vec2 xy, in vec2 b)
	{
	vec2 d = abs(p - xy) - b;
	return length(max(d, 0.0)) + min(max(d.x, d.y), 0.0);
	}