Gavi Schneider gavischneider

LLM Wiki V3: Segmentation

Karpathy gave us the foundation. Rohitg00 warned us what breaks. V3 is how you structure it to scale.

This is a concept document in the same spirit as V1 and V2.

V1 was intentionally vague. Build on it. V2 was intentionally open. Solve it.

LLM Wiki v3: A State-Space Knowledge System

Status: Draft architecture specification
Version: 0.1
Purpose: Extend the LLM Wiki pattern from a persistent markdown wiki into a state-space knowledge system with observations, belief updates, evidence lineage, graph structure, and rendered wiki pages.

0. Executive Summary

LLM Wiki v2

A pattern for building personal knowledge bases using LLMs. Extended with lessons from building agentmemory 20K+ Stars ⭐️, a persistent memory engine for AI coding agents.

This builds on Andrej Karpathy's original LLM Wiki idea file. Everything in the original still applies. This document adds what we learned running the pattern in production: what breaks at scale, what's missing, and what separates a wiki that stays useful from one that rots.

What the original gets right

The core insight is correct: stop re-deriving, start compiling. RAG retrieves and forgets. A wiki accumulates and compounds. The three-layer architecture (raw sources, wiki, schema) works. The operations (ingest, query, lint) cover the basics. If you haven't read the original, start there.

name	wiki
description	Compile personal data (journals, notes, messages, whatever) into a personal knowledge wiki. Ingest any data format, absorb entries into wiki articles, query, cleanup, and expand.
argument-hint	ingest \| absorb [date-range] \| query <question> \| cleanup \| breakdown \| status

Personal Knowledge Wiki

You are a writer compiling a personal knowledge wiki from someone's personal data. Not a filing clerk. A writer. Your job is to read entries, understand what they mean, and write articles that capture understanding. The wiki is a map of a mind.

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.

`vs` Questions

http://stackoverflow.com/questions/804115 (rebase vs merge).
https://www.atlassian.com/git/tutorials/merging-vs-rebasing (rebase vs merge)
https://www.atlassian.com/git/tutorials/undoing-changes/ (reset vs checkout vs revert)
http://stackoverflow.com/questions/2221658 (HEAD^ vs HEAD~) (See git rev-parse)
http://stackoverflow.com/questions/292357 (pull vs fetch)
http://stackoverflow.com/questions/39651 (stash vs branch)
http://stackoverflow.com/questions/8358035 (reset vs checkout vs revert)

	javascript: Promise.all([import('https://unpkg.com/turndown@6.0.0?module'), import('https://unpkg.com/@tehshrike/readability@0.2.0'), ]).then(async ([{
	default: Turndown
	}, {
	default: Readability
	}]) => {

	/* Optional vault name */
	const vault = "";

	/* Optional folder name such as "Clippings/" */