Altay altaywtf

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.

Migrations & Seeding

What are migrations??

Migrations are a way to make database changes or updates, like creating or dropping tables, as well as updating a table with new columns with constraints via generated scripts. We can build these scripts via the command line using knex command line tool.

To learn more about migrations, check out this article on the different types of database migrations!

Creating/Dropping Tables

Ikea Bekant standing desk with Gerton table top

GPG and git on macOS

Setup

No need for homebrew or anything like that. Works with https://www.git-tower.com and the command line.

Install https://gpgtools.org -- I'd suggest to do a customized install and deselect GPGMail.
Create or import a key -- see below for https://keybase.io
Run gpg --list-secret-keys and look for sec, use the key ID for the next step
Configure git to use GPG -- replace the key with the one from gpg --list-secret-keys

	name: "Summarize Turborepo Output"
	description: "Summarize Turborepo output for GitHub Actions"
	runs:
	using: "composite"
	steps:
	- shell: bash
	name: "Summarize Turborepo Output"
	run: \|
	latest_json=$(ls -t .turbo/runs/*.json \| head -1)

	const DOUBLE_PRESS_DELAY = 300;

	// ...

	/**
	* Double Press recognition
	* @param {Event} e
	*/
	handleImagePress(e) {
	const now = new Date().getTime();

	- General
	[ ] The code works
	[ ] The code is easy to understand
	[ ] Follows coding conventions
	[ ] Names are simple and if possible short
	[ ] Names are spelt correctly
	[ ] Names contain units where applicable
	[ ] Enums are used instead of int constants where applicable
	[ ] There are no usages of 'magic numbers'
	[ ] All variables are in the smallest scope possible

	import { Component } from "React";

	export var Enhance = ComposedComponent => class extends Component {
	constructor() {
	this.state = { data: null };
	}
	componentDidMount() {
	this.setState({ data: 'Hello' });
	}
	render() {