🪄📚 Spellbinder

Convert PDFs to EPUBs using AI

Pre-LLM tools in this vein usually output an unreadable mess. Spellbinder leverages the built-in PDF processing and large context window of the Gemini API and is able to output high quality EPUBs ready for ereaders (usually).

Features

• Clean, accurate outputs
• Table of contents and chapter structure
• Fixes OCR errors
• Auto-retry when encountering API errors

Setup

Install:

npm i -g spellbinder-cli

Get a Gemini API key and either set it to SPELLBINDER_API_KEY in your shell, or pass it via the -k/--api-key flag when running spellbinder.

Usage

You can convert a PDF simply by passing it as the first argument:

spellbinder a-book.pdf

Progress logs will be emitted chapter-by-chapter during the conversion process and the final EPUB will be written to a-book.epub.

You can explicitly pass a different output path with a second argument:

spellbinder a-book.pdf a-better-book.epub

You can also change the model or output raw HTML. Here's the full set of options:

spellbinder <input> [output] [options]

Positionals:
  input   Input PDF file path                                           [string]
  output  Output file path (optional)                                   [string]

Options:
      --version  Show version number                                   [boolean]
  -k, --api-key  Gemini API key                              [string] [required]
  -m, --model    Gemini model       [string] [default: "gemini-3-flash-preview"]
  -h, --help     Show help                                             [boolean]

Current limitations

• No images
• No footnote linking
• Some books could be chunked in a way that exceeds the output token limit

Pull requests are welcome if you figure out clever ways to address any of these.

Like books?

Try my other project, Emdash.