LLM assisted book reader by Karpathy (github.com)

Andrej Karpathy released a minimal, self-hosted EPUB reader designed for "reading together" with LLMs: it loads EPUBs one chapter at a time so you can easily copy/paste chapters into your favorite model and follow along. The repo is intentionally lightweight and "90% vibe coded" — no promises of support or future development — but it’s ready to run locally (example: uv run reader3.py dracula.epub creates dracula_data, uv run server.py launches a web UI at localhost:8123). Books (e.g., Project Gutenberg EPUBs) are registered simply by creating folders; the project is MIT-licensed and meant as an inspirational, hackable starting point. This matters because it lowers the friction for interactive, privacy-conscious book exploration with LLMs: self-hosting avoids cloud lock-in and makes it trivial to iterate on reading workflows, chunking strategies, and LLM prompts for summarization, question-answering, or inline annotation. Key technical notes: ephemeral/simple code, local server interface, per-book data dirs, and explicit encouragement to use an LLM to modify the code. Caveats: it’s experimental, unsupported, and copying text into third-party LLMs can expose content — good for prototyping but not a polished production reader.