Feature Proposal: Gemini/LLM API Integration in E-book Viewer - Page 3

kovidgoyal · 11-20-2025, 07:46 AM

Beneficial how?

lucask · 11-20-2025, 03:54 PM

While I share many of the reservations people have here about relying on LLMs I still think this would be an excellent addition.

Here's a possibly less controversial use case: I like to embed the abstract and table of contents for my books and papers in the comments field. This is more granular than a plain tags system and allows me to search for particular results or theorems across my entire library. There are basically no public databases with that kind of data, well cleaned, so I've had to open the file, use the text selection tool of my viewer and enter them manually (with a little script to remove things like excess whitespace and page numbers) over a long time. Of course no purely syntactic program can do that perfectly, but with an LLM and the right prompt I could instantly get clean text requiring only occasional touchups.

The problem is when doing this for multiple books at once outside Calibre the prompt gets forgotten after a while, and you get nonsense unless you periodically start fresh. With this plugin embedded directly in the viewer this need never occur.

kovidgoyal · 11-21-2025, 12:39 AM

It's already implemented in the viewer.

ppiotr · 11-23-2025, 07:33 AM

This is an excellent idea (in fact, I have found your post when searching for a tool which would allow me to do what you suggest).

Do you have the prototype available somewhere for trying ?

You mentioned the usage in reading texts of a more scientific / formal nature (I completely agree that LLMs are an amazing tool. Modern agentic systems are able to provide you with precise references to source texts (and those systems will get better). This means that in the unlikely event of hallucination, links to source texts will drop from the LLM. You can verify sources before informing others.

I disagree with the statement that this is not useful when reading novels. As a non-native english-speaker, I sometimes get lost in more complex prose (e.g. trying to read Ulysses ... but not only). Also, after making a longer break I may not remember all the past details or names of all characters. What I am looking for is passing all the book (up to the place where I am currently reading) as a query context and being able to ask specific questions. For example "What does character A mean by ..." "What do we know about character B" "What were the past interactions between character B and C".

Great job

View Poll Results: Want LLM/AI (e.g., Gemini) features in Calibre Viewer?
Yes, this would significantly enhance my reading experience.	4	12.90%
No, I prefer using external tools or don't need this feature.	27	87.10%
Voters: 31. You may not vote on this poll

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
New Feature Proposal - Leverage ChatGPT AI for Automatic Book Summary and Analysis	mikhail_fil	Calibre	27	10-02-2023 11:24 AM
Calibre - Feature proposal for server	Mr. Nietzsche	Server	2	08-21-2019 10:48 AM
E-Book Viewer feature suggestion?	arthurh3535	Calibre	3	10-20-2018 12:00 AM
Feature request: E-book Viewer TOC doesn't show where you are	alessandro	Calibre	5	11-21-2013 11:16 AM
Feature proposal: attachments	chrisberkhout	Calibre	1	08-07-2013 11:40 PM

11-20-2025, 07:46 AM	#31
kovidgoyal creator of calibre Posts: 45,659 Karma: 28549046 Join Date: Oct 2006 Location: Mumbai, India Device: Various	Beneficial how?

11-20-2025, 03:54 PM	#32
lucask Member Posts: 10 Karma: 10 Join Date: Aug 2017 Location: Uganda Device: none	While I share many of the reservations people have here about relying on LLMs I still think this would be an excellent addition. Here's a possibly less controversial use case: I like to embed the abstract and table of contents for my books and papers in the comments field. This is more granular than a plain tags system and allows me to search for particular results or theorems across my entire library. There are basically no public databases with that kind of data, well cleaned, so I've had to open the file, use the text selection tool of my viewer and enter them manually (with a little script to remove things like excess whitespace and page numbers) over a long time. Of course no purely syntactic program can do that perfectly, but with an LLM and the right prompt I could instantly get clean text requiring only occasional touchups. The problem is when doing this for multiple books at once outside Calibre the prompt gets forgotten after a while, and you get nonsense unless you periodically start fresh. With this plugin embedded directly in the viewer this need never occur.

11-21-2025, 12:39 AM	#33
kovidgoyal creator of calibre Posts: 45,659 Karma: 28549046 Join Date: Oct 2006 Location: Mumbai, India Device: Various	It's already implemented in the viewer.

11-23-2025, 07:33 AM	#34
ppiotr Junior Member Posts: 1 Karma: 10 Join Date: Nov 2025 Device: kobo	This is an excellent idea (in fact, I have found your post when searching for a tool which would allow me to do what you suggest). Do you have the prototype available somewhere for trying ? You mentioned the usage in reading texts of a more scientific / formal nature (I completely agree that LLMs are an amazing tool. Modern agentic systems are able to provide you with precise references to source texts (and those systems will get better). This means that in the unlikely event of hallucination, links to source texts will drop from the LLM. You can verify sources before informing others. I disagree with the statement that this is not useful when reading novels. As a non-native english-speaker, I sometimes get lost in more complex prose (e.g. trying to read Ulysses ... but not only). Also, after making a longer break I may not remember all the past details or names of all characters. What I am looking for is passing all the book (up to the place where I am currently reading) as a query context and being able to ask specific questions. For example "What does character A mean by ..." "What do we know about character B" "What were the past interactions between character B and C". Great job

Advert