karpathy

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.

Chrisbryan17

#!/usr/bin/env python3
"""
microgpt2.py

A dependency-free, single-file GPT-style language model in pure Python.

Key properties:
- stdlib only
- explicit forward/backward kernels (no generic scalar autograd)
- flat float32 parameter buffers using array('f')

busfahrer

"""
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

logan-robbins

"""
microgpt_dnb.py — Fork of Karpathy's microgpt with Dynamic Notes Bus

Original: https://gist.github.com/karpathy/8627fe009c40f57531cb18360106ce95
Paper:    https://arxiv.org/abs/2512.10054
Code:     https://github.com/logan-robbins/parallel-decoder-transformer

Karpathy's microgpt is the complete GPT algorithm in ~200 lines of pure Python.
This fork adds ~100 lines to show where the Dynamic Notes Bus (DNB), Shared
Notes Cross-Attention (SNC), and Planner Head fit inside the transformer to

olveirap

"""
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

mplekh

import math
import random

random.seed(42)

# -----------------------------------------------------------------------------
# Tape-based Autograd Engine
# -----------------------------------------------------------------------------

class Tape:

karpathy

"""
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

nort3x

Introduction

I personally experienced slowdowns and issues while using the ja-netfilter agent.

I decided to investigate how JetBrains checks license validity, because despite explicitly configuring JetBrains to work offline, it still attempts to validate licenses.

Here are my findings:
Two domains are responsible for revoking invalid licenses:

• www.jetbrains.com
• account.jetbrains.com

jet10000

brew install caddy mkcert nss dnsmasq

mkcert -install

mkcert '*.app.test' '*.cdn.test'

# rename the certs and move them under /usr/local/etc/caddy/certs

cat <<EOF > /usr/local/etc/caddy/Caddyfile
*.app.test:443, *.cdn.test:443 {

0187773933

#!/usr/bin/env python3
import os
import time
import subprocess
from PIL import ImageGrab , Image
from pathlib import Path
import io
import codecs
import tempfile
import pyperclip