ARKit face tracking is shrouded in a ton of myth and mystique.
I reverse engineered the internal face tracking API, and this is how it actually works.
This is based on reverse engineering the API of Apple's face detection framework and testing it with my own code. Some of it is educated guesswork based on names and strings. I did not reverse engineer the details of how ARKit/FaceKit works, but given what I've seen, I'm quite confident in the overall big picture.
한국어 번역 (Korean translation): nikke-pc-launcher-compromise_ko.md
Caution
From 2026-09-27, four official JavaScript files on Level Infinite's CDN contained attacker code. The NIKKE PC launcher loads these files. From Oct 23 at the latest, opening the launcher could install a remote-control agent: while the launcher was open, the attacker could run any command on your machine, read any file and take screenshots.
Update, 2026-11-03 22:40 GMT: three of the files were replaced on Nov 3 at 17:04, 17:13 and 17:21 GMT. The fourth, FeedList.feb647c9.js, was cleaned at the origin on Oct 28, but two CDN servers still delivered the modified copy at 22:26 GMT. The attacker's server no longer answered at 18:31 GMT. None of this undoes anything on machines that were already infected, and the old files may still be in your launcher's cache. **Do not open the NIKKE PC launcher until Level Infinite confirms the fix,
| Character code (character) | Duration (ms) | Alias to | Pronounced as |
|---|---|---|---|
| 27 (�) | 656 | None | Escape |
| 33 (!) | 1333 | None | Exclamation mark |
| 35 (#) | 409 | None | Hash |
| 36 ($) | 425 | None | Dollar |
| 37 (%) | 631 | None | Percent |
| 38 (&) | 329 | None | And |
| 43 (+) | 396 | None | Plus |
| 45 (-) | 706 | None | Dash |
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.
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.
