Talk to your Calibre Library

[FacetiousKnave]

Hey everyone,

Like a many of you, I have a massive Calibre library (7,000+ DeDRM EPUBs).

Although I read a lot, sometimes I struggle to know what to read next, or what books address specific topics, or how those topics are addressed, how I can connect a current read to another one to follow a topic-related path.

So, in other words, I wanted a way to actually interact with that hoard of knowledge using AI—asking questions across hundreds of books at once, finding specific philosophical arguments, or cross-referencing authors—but I absolutely refused to upload my personal library to OpenAI or pay for a cloud subscription.

So, I built Dive. It’s an open-source, local-first desktop app that acts as a secure wrapper for Ollama, but with a heavily engineered offline RAG (Retrieval-Augmented Generation) engine built specifically for massive EPUB libraries.

I'm going to try to explain how it works:

• Point it at your Calibre Folder: You don’t need to move or convert your books. Just point the app’s indexer at your Calibre Library directory, and it will safely parse the raw text from your EPUBs. The app does not edit or touch your ebooks in any way, it only reads them.
  
• Ultra-Compact Database (No Bloat): Most vector databases explode in size if you feed them thousands of books. I engineered this to use Matryoshka semantic compression (256-dimension vectors) and SQLite external-content tables. The result? You can index the mathematical meaning and text of 7,000+ books into a single database that is less than 10 GB.
  
• 100% Local & Private: It runs entirely on your hardware using Ollama and local embedding models. No internet required, no API keys, and your data never leaves your machine.
  
• Bulletproof Indexing for “Omnibus” Books: If you have massive “Complete Works” EPUBs that usually crash standard parsers, this handles them smoothly using in-memory ZIP streams. Plus, the indexing job is 100% resumable. If you close your laptop halfway through indexing your library, it picks up exactly where it left off.

If you want to give it a try, the repository is here: https://github.com/DuckTapeKiller/dive

Please note that this app only works on both Intel / Arm64 MacBooks.

[JSWolf]

Why not use your AI to convert this to a cross-platform app?

[rantanplan]

but I absolutely refused to upload my personal library to OpenAI or pay for a cloud subscription.

The current way that the price for tokens go it wouldn't make sense to use a cloud subscription for this anyway .

You can index the mathematical meaning and text of 7,000+ books into a single database that is less than 10 GB.

You know whats funny about that number? That means that the database is likely larger than your ebook library.

[dandaman]

Great! Will try on my MacBook Pro in a bit.

[JSWolf]

Hello? calibre library? Are you there?

[FacetiousKnave]

Originally Posted by rantanplan

The current way that the price for tokens go it wouldn't make sense to use a cloud subscription for this anyway .

You know whats funny about that number? That means that the database is likely larger than your ebook library.

My library is 23gb mainly because of my comic epubs, which cannot be indexed for obvious reasons.

I added int8 to the code, and the total size of the index is only 4gb.

[FacetiousKnave]

Originally Posted by JSWolf

Why not use your AI to convert this to a cross-platform app?

I don't think that would be possible. The app is not specifically designed for Calibre; the RAG is the only feature that directly works with your Calibre library, so that you can ask questions about your books.

Apart from the RAG, I built this as a personal project to automate mundane tasks. The app lets you work with Pi-Coding-Agent and local models via Ollama. Using Pi, you can perform various tasks like downloading videos playlists or podcasts from YouTube or fact-checking information with Wikipedia or Britannica.

So making this an organic part of Calibre would be a different project.

[JSWolf]

Originally Posted by FacetiousKnave

I don't think that would be possible. The app is not specifically designed for Calibre; the RAG is the only feature that directly works with your Calibre library, so that you can ask questions about your books.

Apart from the RAG, I built this as a personal project to automate mundane tasks. The app lets you work with Pi-Coding-Agent and local models via Ollama. Using Pi, you can perform various tasks like downloading videos playlists or podcasts from YouTube or fact-checking information with Wikipedia or Britannica.

So making this an organic part of Calibre would be a different project.

You won't know until you try. And yes, any app can be converted to run on another platform. You may not be able to do it as a cross platform app, but you can write an app to do what you want on other platforms.