Press "Enter" to skip to content

Calibre 8.16’s AI Wants to Talk Books. It Can’t Even Find Mine.

After looking at the new Calibre 8.16 which puts AI on the playing field, our writer wonders if that’s a good idea.

The FOSS Force Linux App of the Week — Calibre

Calibre main page.

A couple of weeks ago, Calibre 8.16 was released, with a new AI feature that might excite or anger some users.

Once upon a time — prior to getting hooked up with my current publisher — I relied on the Calibre ebook management tool to help get my novels published on the likes of Amazon. Calibre was an absolute essential, and I would have struggled mightily without it.

Calibre’s New AI Feature

With the latest iteration of Calibre, you can set up an AI provider and then discuss a book via queries. It’s quite a fascinating feature that could help you really understand a book you’ve just read or get back up to speed on a book you read some time ago. You can even ask the AI what book to read next, based on what you just read, or ask it to find similar books. You can even run local LLMs (such as those from Ollama) with Calibre.

This is the first time Calibre has been given a jolt of AI.

How do I feel about that and how well does it work?

As someone who makes a living from writing and has over 70 published books, I’m torn. On one hand, I do not want AI anywhere near creative work because I find it to be counter to the craft. I see AI as nothing more than a creative crutch and would never use it as such.

On the other hand, I can see the value in gaining further insights that you might have missed when reading a novel. I did run a few tests with this and, quite honestly, found the AI to pretty much miss the point on all fronts.

For example, I added my book “Dracula Theory” into Calibre 8.16.2 and asked it some questions. Every question pretty much came back with a similar answer… which was wrong. Given that I’m the author of the book, I should know.

Converting Calibre’s Quick Start Guide to another format, and editing the layout along the way.

Part of the problem could be the local LLMs that are available to use, which are:

  • gpt-oss:latest
  • llama3.2-vision:latest
  • llama3.2:latest
  • qwen3-coder:latest

I tried all four, and the AI was still unable to answer my questions about the book. In fact, every question I asked the AI started with “Unfortunately, I couldn’t…”

Nextcloud resilient communication and collaboration.

This does not bode well.

I then simplified my query to just “Summarize.”

Nope. Pretty much every query wound up with an answer that was way off the mark.

Editing the metadata for Calibre’s Quick Start Guide.

Would this be the case had I used Stoker’s “Dracula”? Maybe, maybe not. But shouldn’t the AI be able to dive into the text of a book, regardless of which book it is, and be able to discuss it?

Sadly, that’s not the case here.

I decided to convert the book to MOBI to see if that would help. Unfortunately, the AI was still off the mark. Finally, I asked it to recommend another book written by me to which it replied, “I couldn’t find any information about a novel called ‘Dracula Theory.'” “Dracula Theory” was the very book I was using with the AI.

Come on!

Looking to purchase Carla Schroder’s book on Audacity.

My conclusion with the Calibre AI is that it’s not exactly ready for prime time. It’s a great idea, but it needs a bit more work — especially with the local LLMs — to be functional. It’s possible that using a cloud-based LLM might have better results, but I value my privacy too much to even touch them. This is especially true when working with my own books.

The public reception of AI in Calibre has not been all that warm. In a Reddit thread that was created when it was just rumored that the app would be getting get AI, some of the responses were telling:

“What bloody questions could you possibly have for an AI that couldn’t be answered by just having a text search function and reading the damn books yourself? Literary analysis?? An AI wouldn’t even be good at that, and there’s zero reason to ask an AI to do it because you can already do that yourself.”

What Else Is New

Other than the big-ticket AI addition, version 8.16.2 is mostly about bug fixes. Some of the more important bug fixes include

  • Use a named local timezone for better display of historical dates in the local timezone
  • Improved PDF conversion, which is important because PDF conversion has been iffy at best. In the past, I’ve attempted to convert PDFs to ebooks with not much luck or consistency. It seems the developer has finally fixed some of those issues, and PDF conversion is more reliable.
  • The Amazon.it store plugin has been updated.
  • Fixed the incorrect series index when downloading metadata from amazon.co.jp.

Getting It Running

The latest version also comes with an explicit warning against distro-package builds being outdated. Because of this, it’s crucial that you install Calibre via the command line with:

sudo -v && wget -nv -O- https://download.calibre-ebook.com/linux-installer.sh | sudo sh /dev/stdin

Do not install Calibre from your default package manager, or you will not see the new features or bug fixes in place.

Although the AI does need a lot of work, Calibre still remains the dominant FOSS ebook management tool. As I see it, the injection of AI goes against the very heart of what Calibre has stood for, which is manually reading and enjoying books. Sure, AI could be used for other purposes, but I could also imagine users simply telling Calibre’s AI to summarize a book for them (once it starts working), and the user would consider the book read.

That’s cheating, and I think we all know how the Linux community feels about cheating.

The good news is that the AI feature will not function without it being configured. So, if you don’t want AI involved, don’t set it up.

Things I like about Calibre 8.16

Things I don’t like about Calibre 8.16

  • A good round of bug fixes goes a long way.
  • The more consistent PDF conversion is a big plus.
  • The UI hasn’t changed much, and that’s a good thing.
  • AI doesn’t work well, especially with lesser-known books.

Be First to Comment

Leave a Reply

Your email address will not be published. Required fields are marked *