Alexey Okhrimenko obenjiro

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.

Claude Desktop Uninstallers

Scripts to fully remove Claude Desktop and all the files it leaves behind, because Anthropic's official uninstaller doesn't.

The Problem

Anthropic shipped a Cowork feature in Claude Desktop that silently spins up a Linux VM on your machine. No warning. No disk space prompt. No way to opt out at install time.

The VM bundle grows to ~10GB and lives in your AppData/Application Support folder. When you uninstall Claude Desktop through the normal method, the VM bundle stays behind. Most users will never find it because it's buried in a hidden system folder.

TypeScript 5.x to 6.0 Migration Guide

TypeScript 6.0 is a transition release bridging 5.9 and the forthcoming 7.0 (a native Go port). Most changes are new defaults and deprecations preparing for 7.0. Here is what you need to do:

Most projects need these tsconfig changes:

{
    "compilerOptions": {
        "types": ["node"],           // @types are no longer auto-discovered (see §1.6)

Install HF Code Autocomplete VSCode plugin.

We are not going to set an API token. We are going to specify an API endpoint.
We will try to deploy that API ourselves, to use our own GPU to provide the code assistance.

We will use bigcode/starcoder, a 15.5B param model.
We will use NF4 4-bit quantization to fit this into 10787MiB VRAM.
It would require 23767MiB VRAM unquantized. (still fits on a 4090, which has 24564MiB)!

Setup API

This is just a dump of some interesting undocumented features of webOS (3.8 specifically, on early 2018 4k LG TV) and other development-related tips.

Homebrew app ideas

OpenVPN frontend (OpenVPN itself is easily buildable and runs on webOS TVs: https://discord.com/channels/407937994037919756/835489130967859251/906943542457401395)
App autostart manager - dynamically update arbitrary app config and register it as input (see Registering an app as an input)
webos-vncserver frontend with autostart option
Package hyperiond into Piccap directly
Quick Screenshot (expose HTTP port that just returns PNG of current screen contents)
Custom IR remote codes editor/updater/blaster

На случай важных переговоров

[11.01.18 18:47] [Forwarded from Алексей Охрименко]

Google, Microsoft
Typescript из коробки
Единственный вреймворк с Dependency Injection из коробки
Не нужно ничего React-ить и AngularJS-ифаить. Больше никаких оберток. jQuery плагины и D3 можно использовать на прямую
Более современный фреймворк

	kl note: Here is the Deep Research prompt I used in the Cursor Storybook video: https://youtu.be/gXmakVsIbF0
	For background, this is a real-world tech feasibility task I am working on where I am trying to build out a realistic-looking fake website for an AI browsing agent to use to complete tasks. I found this random site that was close enough to what I wanted so I used it as a shortcut instead of taking the time to write out a full PRD or anything.

	...above this was just the transcript and the initial guidance...


	Act as a technical fellow and create a detailed, step-by-step guide to recreating this software using a modern stack. Here is the cursorrules for this repository:
	# .cursorrules

	Components & Naming

	#!/usr/bin/python
	# -- coding: utf-8 --

	import transformers
	from transformers import AutoTokenizer, AutoModelForCausalLM, AutoModel, AutoConfig
	from transformers.onnx import FeaturesManager, convert, export
	from pathlib import Path
	import os

	model_id = 'gpt2-large'

	/**
	* ## Angular custom webpack config for compatibility with IE11
	* ------------------------------------------------------------
	* 1. Add this file to your project root. Add the modules list that needs transpiling (see below code).
	* 2. Install dev dependencies:
	*
	* `$ npm i -D @angular-builders/custom-webpack:browser babel-loader @babel/core @babel/preset-env browserlist`
	*
	* 3. Add this to your `angular.json`:
	*


	try {
	document.createElement('canvas').getContext('webgl', { powerPreference: "high-performance" })
	} catch(e) {}