From Karpathy: "A lightweight, self-hosted EPUB reader that lets you read through EPUB books one chapter at a time. This makes it very easy to copy paste the contents of a chapter to an LLM, to read along. Basically - get epub books (e.g. Project Gutenberg has many), open them up in this reader, copy paste text around to your favorite LLM, and read together and along. This project was 90% vibe coded just to illustrate how one can very easily read books together with LLMs."

The project uses uv. So for example, download Moby Dick EPUB3 to this directory as moby_dick.epub, then:

uv run reader3.py moby_dick.epub

This creates the directory moby_dick_data, which registers the book to your local library. We can then run the server:

uv run server.py

And visit localhost:8123 to see your current Library. You can easily add more books, or delete them from your library by deleting the folder.

You can now highlight passages and talk to a lightweight local LLM without leaving the reader.

1. Install and start Ollama (or another compatible server) locally, e.g.:

   ollama serve &
   ollama run llama3.2

2. (Optional) Point the reader to a different host/model:

   export OLLAMA_HOST="http://localhost:11434"
   export OLLAMA_MODEL="llama3.2"

3. Open a chapter, highlight any text, click Add to context, type a question, and press Ask question.
   Use Summarize section to get a quick recap; if no snippets are selected, the whole chapter is used.

Context is truncated to ~8k characters before sending to the model. If you run a different LLM gateway, adapt the /api/llm/query handler in server.py.

• After pulling new changes, always restart the FastAPI server so the /api/llm/query route is available:

  uv sync   # installs httpx the first time
  uv run server.py

• If the assistant UI shows “Not Found”, it means the server you’re talking to doesn’t have that endpoint. Restart the server and reload the browser tab.
• To confirm the API is alive, run:

  curl -X POST http://127.0.0.1:8123/api/llm/query \
       -H 'Content-Type: application/json' \
       -d '{"book_id":"Book_data","chapter_index":0,"question":"test","context":"hello"}'

  You should get a JSON answer instead of HTML. Once this works, the in-app assistant will work too.

MIT