Ankur Tyagi 7h3rAm

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.

This gist contains a proof of concept project (hacked together during one evening) aimed at connecting an AI chat assistant to the Meshtastic network, enhancing its capabilities. The assistant is designed to provide concise and helpful responses within the constraints of the Meshtastic environment.

Project Overview:

This project brings together two favorite elements: the Meshtastic network and an AI chat assistant. The assistant is based on the "FuseChat-7B-VaRM-Q5_K_M" model, and the code is written in Python.

Q: What is the cube MAC address?
A: MAC is the Bluetooth device hardware address, each device instance has it unique. It has 48 bits and typically represented as 6 bytes written in hexadecimal and separated by colons or hyphens, for example AB:12:34:5D:34:12.

Q: Why MAC address is ever needed?
A: The GAN smart cube application-layer protocol encrypts all data using AES cypher. Encryption keys are unique for each cube instance, and generated based on cube's MAC address. So to decrypt any data sent by cube it is required for application to know cube MAC address.

Q: But I used another GAN cube with Cubeast, or different smartcubing software, and it does not ever ask me for the MAC address.
A: For some GAN smart cubes range used for MAC addresses was small enough, so Cubeast try to bruteforce address automatically. This can be done in reasonable amount of time. But this is not an option for newer cubes and MAC can't be guessed. Also some software like platform-native applications h

	#include <stdio.h>
	#include <stdlib.h>
	#include <string.h>
	#include <ctype.h>
	#include <openssl/sha.h>

	#define TARGET_PREFIX "20250327"
	#define MAX_WORDS 256
	#define MAX_TEXT 2048
	#define MAX_ATTEMPTS (1ULL << 32) // 2^32 attempts (~4.3B, enough for 8-char prefix)

	they're not Particularly Rare, But I think they're more pleasing to See.

	For instance This "2822302" Makes me Especially Happy: https://github.com/BrowserBox/BrowserBox/commit/2822302387c4cb7ff71c4239da3dc5fa4c07e165

	It Even Extends up to 10 digits! That's Not Particularly rare - Roughly 1% chance (10^10/16^10 i think) - But i just think they look nice.

	Are there any other people out there who are particularly pleased when they hit that? sorta like hitting 7777 on the odometer, or whatever.

	I'm also a fan of the purely Numeric Identifiers Twitter/X Uses (and has for ages).

	- After making changes, ALWAYS make sure to start up a new server so I can test it.
	- Always look for existing code to iterate on instead of creating new code.
	- Do not drastically change the patterns before trying to iterate on existing patterns.
	- Always kill all existing related servers that may have been created in previous testing before trying to start a new server.
	- Always prefer simple solutions
	- Avoid duplication of code whenever possible, which means checking for other areas of the codebase that might already have similar code and functionality
	- Write code that takes into account the different environments: dev, test, and prod
	- You are careful to only make changes that are requested or you are confident are well understood and related to the change being requested
	- When fixing an issue or bug, do not introduce a new pattern or technology without first exhausting all options for the existing implementation. And if you finally do this, make sure to remove the old implementation afterwards so we d

	# -----------------------------------------------------------------------------
	# AI-powered Git Commit Function
	# Copy paste this gist into your ~/.bashrc or ~/.zshrc to gain the `gcm` command. It:
	# 1) gets the current staged changed diff
	# 2) sends them to an LLM to write the git commit message
	# 3) allows you to easily accept, edit, regenerate, cancel
	# But - just read and edit the code however you like
	# the `llm` CLI util is awesome, can get it here: https://llm.datasette.io/en/stable/

	gcm() {

	# ==============================================================================
	# ShellGPT
	# ==============================================================================

	# ------------------------------------------------------------------------------
	if which sgpt >/dev/null 2>&1; then
	# ------------------------------------------------------------------------------

	alias sgpt-chat="sgpt --repl chat"
	alias sgpt-code="sgpt --repl code --code"

	#!/bin/sh
	# rename-pictures.sh
	# Author: Justine Tunney <jtunney@gmail.com>
	# License: Apache 2.0
	#
	# This shell script can be used to ensure all the images in a folder
	# have good descriptive filenames that are written in English. It's
	# based on the Mistral 7b and LLaVA v1.5 models.
	#
	# For example, the following command: