New AI feature: What book to read next

[Leonatus]

I guess it's a nice feature, but every click on any book results to this message:

calibre, version 8.16.0
FEHLER: Unbehandelter Ausnahmefehler: <b>AttributeError</b>:'function' object has no attribute 'prompt_text'

calibre 8.16  embedded-python: True
Linux-6.8.0-88-generic-x86_64-with-glibc2.39 Linux ('64bit', 'ELF')
('Linux', '6.8.0-88-generic', '#89-Ubuntu SMP PREEMPT_DYNAMIC Sat Oct 11 01:02:46 UTC 2025')
Python 3.11.14
Interface language: de
EXE path: /opt/calibre/bin/calibre
Successfully initialized third party plugins: Annotations (1, 17, 13) && Count Pages (1, 14, 4) && Ebook Translator (2, 4, 1) && Kobo Utilities (2, 24, 5) && Modify ePub (1, 8, 8)
Traceback (most recent call last):
  File "calibre/gui2/actions/similar_books.py", line 46, in ask_ai
  File "calibre/gui2/actions/llm_book.py", line 49, in ask_ai_with_action
  File "calibre/gui2/dialogs/llm_book.py", line 220, in activate_action
  File "calibre/gui2/dialogs/llm_book.py", line 243, in prompt_text_for_action
AttributeError: 'function' object has no attribute 'prompt_text'

Just wonder what this means.

[BetterRed]

I think you have to configure an AI provider via the View->Discuss selected book(s) with AI (Ctrl+Alt+A) action before you can get anywhere:



Because I think it also applies to the Similar Books->Ask AI for what to read next action, maybe that dialogue should be in Preferences.

BR

[Leonatus]

I got the key, and "discuss selected book with AI" works (sometimes), but at closing the window, Calibre shuts entirely down. And the "What book to read next" feature results in the same message above mentioned.
Edit: I just see that there are two options of "Ask AI what to read next": One in "Open book" -> "Discuss selected book with AI" (which works), and one in "Similar books", which ends with the mentioned message.

[kovidgoyal]

It was a bug: https://github.com/kovidgoyal/calibr...91a77d01e7d208

You can access the same feature view the view button menu and then clicking "Read next"

[Leonatus]

Ah, thank you! Remains the fact of the unexpected shutdown after using the "summarize" feature.

[kovidgoyal]

Originally Posted by Leonatus

Ah, thank you! Remains the fact of the unexpected shutdown after using the "summarize" feature.

I cannot replicate that, steps I tried, select a book, click discuss with AI, click summarize. Then click close after summarization is completed.

[thiago.eec]

I have another problem: in the 'Metadata' section of the configure dialog, I cannot select any box. Only clicking 'select all' works.

[kovidgoyal]

Originally Posted by thiago.eec

I have another problem: in the 'Metadata' section of the configure dialog, I cannot select any box. Only clicking 'select all' works.

Click the actual text not the checkbox and it will work

[Leonatus]

Originally Posted by kovidgoyal

I cannot replicate that, steps I tried, select a book, click discuss with AI, click summarize. Then click close after summarization is completed.

Maybe it's due to the system: Linux Mint 22.2 Cinnamon Zara. the issue persists with me.

[kovidgoyal]

I tested on Linux as well, are you waiting for the summarization to be completed before clicking close?

[Leonatus]

Originally Posted by kovidgoyal

I tested on Linux as well, are you waiting for the summarization to be completed before clicking close?

Yes, of course.

[Sirtel]

Just from curiosity, I tried the new feature "Ask AI what to read next" (not that I need more options in addition to the thousands of books currently in my calibre library...). It's pretty nifty. Of course most of the recommended books are well known and I've either read them already or am planning to do so, but I did get a couple of recs I hadn't heard about before. I do hope those weren't just hallucinations.

[Comfy.n]

I was amazed with LM Studio, for its wide customising features, and now I found about Openrouter. It's astonishing.

Using local models requires more recent and powerful hardware so I was playing more with the LiquidAI ones, given my GPU is an ancient MX230. Also, local models don't have much factual knowledge and can't browse the web to do any cross-checking.

The Openrouter large models (and I mean the free ones) are blazing fast since they don't depend on user hardware and are quite knowledgeable.

[kovidgoyal]

Originally Posted by Leonatus

Yes, of course.

See
https://bugs.launchpad.net/calibre/+bug/2133992