San sandin

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.

Update!

This gist is out of date and I can no longer help much, as I got rid of my Mac.

Please visit T2 Linux website for more and better information:

https://t2linux.org/

Acknowledgements

This gist is just a compilation of the hard work that others have put in. I'm not a software developer, so if there are any mistakes or better ways of doing things, I'd appreciate any suggestions. Here's a list of the real heroes who made this possible:

IDAPython cheatsheet

Important links

Basic commands

Get informations on the binary

IDA Microcode notes

References

Ilfak's presentation at Recon 2018
Microcode in pictures
Hex-Rays Microcode API vs. Obfuscating Compiler
Scripts vds10, vds11, vds12 and vds13 from Hex-Rays SDK

	// =============================================================================
	// XNU kperf/kpc demo
	// Available for 64-bit Intel/Apple Silicon, macOS/iOS, with root privileges
	//
	//
	// Demo 1 (profile a function in current thread):
	// 1. Open directory '/usr/share/kpep/', find your CPU PMC database.
	// M1 (Pro/Max/Ultra): /usr/share/kpep/a14.plist
	// M2 (Pro/Max): /usr/share/kpep/a15.plist
	// M3: /usr/share/kpep/as1.plist

	// analyzer.cpp : This file contains the 'main' function. Program execution begins and ends there.
	// compile:
	// https://stackoverflow.com/questions/35116327/when-g-static-link-pthread-cause-segmentation-fault-why
	// g++ analyzer.cpp --std=c++17 -lstdc++fs -O3 -o analyzer -g -static -lrt -pthread -Wl,--whole-archive -lpthread -Wl,--no-whole-archive
	//

	#include "httplib.h"
	#include <assert.h>
	#include <iostream>
	#include <fstream>

	This bug was also called moonshine in the beginning
	Basically the following bug is present in all bootroms I have looked at:
	1. When usb is started to get an image over dfu, dfu registers an interface to handle all the commands and allocates a buffer for input and output
	2. if you send data to dfu the setup packet is handled by the main code which then calls out to the interface code
	3. the interface code verifies that wLength is shorter than the input output buffer length and if that's the case it updates a pointer passed as an argument with a pointer to the input output buffer
	4. it then returns wLength which is the length it wants to recieve into the buffer
	5. the usb main code then updates a global var with the length and gets ready to recieve the data packages
	6. if a data package is recieved it gets written to the input output buffer via the pointer which was passed as an argument and another global variable is used to keep track of how many bytes were recieved already
	7. if all the data was recieved th

	# coding: utf-8
	# py2 origin author lrdcq
	# usage python3 unwxapkg.py filename

	__author__ = 'Integ: https://github.com./integ'

	import sys, os
	import struct

	class WxapkgFile(object):

	# Linux
	# add the following to "~/.gitconfig" file

	[merge]
	tool = intellij
	[mergetool "intellij"]
	cmd = /usr/local/bin/idea merge $(cd $(dirname "$LOCAL") && pwd)/$(basename "$LOCAL") $(cd $(dirname "$REMOTE") && pwd)/$(basename "$REMOTE") $(cd $(dirname "$BASE") && pwd)/$(basename "$BASE") $(cd $(dirname "$MERGED") && pwd)/$(basename "$MERGED")
	trustExitCode = true

	[diff]