Jason Davis jasondavis

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.

Modern CLI Tools for *nix and Windows

A collection of CLI tools I use (or have used). Most can be installed using your *nix distro’s package manager, or with Scoop on Windows. If you’re not already using Scoop, it's definitely worth trying.

✨ Other lists you might find interesting

modern-unix – A collection of modern/faster/saner alternatives to common Unix commands. Most work with both *nix and Windows.
A list of new(ish) command line tools – Julia Evans

	/**
	* File: GitHubFolderToGist-v1.1.2.js
	* Version: 1.1.2
	* Description:
	* Select a folder or ZIP archive from iOS Files and publish its contents as
	* a GitHub Gist. Nested paths are flattened into Gist-safe filenames.
	*
	* Features:
	* - Select an iOS Files folder or a .zip archive.
	* - Recursively includes files from nested folders.

	import json
	import os
	import sys
	import getpass
	from datetime import datetime, timezone

	import praw
	from praw.models import MoreComments

	"""
	The most atomic way to train and inference a GPT in pure, dependency-free Python.
	This file is the complete algorithm.
	Everything else is just efficiency.

	@karpathy
	"""

	import os # os.path.exists
	import math # math.log, math.exp

	"""
	The most atomic way to train and run inference for a GPT in pure, dependency-free Python.
	This file is the complete algorithm.
	Everything else is just efficiency.

	@karpathy
	"""

	import os # os.path.exists
	import math # math.log, math.exp

	{
	"schemaVersion": "0.1.0",
	"name": "Ultimate Hybrid Clipper v2",
	"behavior": "create",
	"noteContentFormat": "text",
	"context": "{{selectorHtml:article,main,.post-content,.entry-content,#content,.content\|strip_tags:script,style,nav,footer,aside,header,.sidebar,.comments,.advertisement\|strip_attr:class,id,style,data-*}}",
	"properties": [
	{
	"name": "url",
	"value": "{{url}}"

	{
	"schemaVersion": "0.1.0",
	"name": "Universal Smart Clipper",
	"behavior": "create",
	"noteNameFormat": "{{title\|safe_name}}",
	"path": "Clippings/Smart",
	"context": "{{selectorHtml:article,main,[role=main],.post,.entry-content,.article-content,#content,.content\|remove_html:(\"img,aside,script,style,nav,header,footer,form,button,svg,iframe,noscript,advertisement,.ad,.ads,.sidebar,.comments,.related\")\|markdown\|replace:\"/```[\\\\s\\\\S]*?```/g\":\"\"\|replace:\"/\\n{3,}/g\":\"\\n\\n\"\|slice:0,15000\|trim}}",
	"noteContentFormat": "# {{title}}\n\n> [!info]- Page Metadata\n> - Source: {{url}}\n> - Author: {{author}}\n> - Published: {{published\|date:\"YYYY-MM-DD\"}}\n> - Clipped: {{date\|date:\"YYYY-MM-DD HH:mm\"}}\n> - Word Count: {{words}}\n\n---\n\n## Executive Summary\n\n{{\"Based solely on the provided context, write a comprehensive executive summary in 2-3 paragraphs (150-250 words total). Cover the main thesis, key arguments, and conclu

	/* ! Twitter 剪藏块 */

	/* * 结合了 Callout 和 Cssclasses 所以放到一起 */

	/* Callout */

	/* 处理链接标签 */
	.cm-embed-block.cm-callout:has(.callout[data-callout="twitter"]) {
	pointer-events: fill;

	<%*
	let raindrop = tp.user.raindrop;
	let collections = await raindrop.collections();
	let collection = await tp.system.suggester((item) => item.title, collections);
	let raindrops = await raindrop.raindrops(collection._id);
	let drop = await tp.system.suggester((item) => item.title, raindrops);
	drop = await raindrop.raindrop(drop._id);

	let title = drop.title
	let url = drop.link