[GUI Plugin] Chapter Summarizer — Flash and detailed AI summaries, ready to use

[Comfy.n]

Chapter Summarizer

Generates summaries of your books directly inside Calibre — flash overviews, chapter-by-chapter breakdowns, or key events/ideas extraction. Works with any AI provider you already have configured in Calibre, plus two built-in zero-config demo servers to try it without setup.

Quick start

Chapter-by-chapter summary (EPUB, KEPUB, AZW3)

1. Select a book in your library and open Chapter Summarizer
2. Click Select chapters… and check the chapters you want
3. Click Summarize now

Flash summary (any format)

1. Select a book in your library and open Chapter Summarizer
2. Check Flash summary
3. Click Summarize now

Tips

• The zero-config demo servers are good for a first test. For regular use, set up your own provider key in Calibre's AI preferences.
• For NVIDIA Free, OpenRouter, and LM Studio setup, use the built-in Free API Guide (accessible from the About tab).
• Chapter summaries are cached — use "Select uncached" in the chapter selector to resume an interrupted job without re-summarizing what's already done.

Summary modes

• Flash overview — single summary of the whole book with a highlight
• Per-chapter — one summary per chapter or section, with optional key events or key ideas extraction
• Custom prompt — write your own instruction for the AI

Output options

• Append to the book's Comments field
• Save as a new book format in your library (EPUB, HTML, or plain text)
• Save to Calibre data files (Markdown, HTML, or EPUB)
• Open directly in your browser

AI providers
Uses whatever Calibre AI providers you have installed and configured — GitHub AI, Google OpenAI-compatible, OpenRouter, and local (tested with LM Studio).
NVIDIA is also available and works with a free API key.

Two built-in zero-config demo servers let you test the plugin immediately without an API key. They rotate through a curated model pool and may be rate-limited or temporarily unavailable — intended for first use, not regular work.

Chapter selection
For EPUB, KEPUB, and AZW3 books, a "Select chapters" dialog lets you check exactly which chapters to summarize, with a live filter and cached/uncached status shown per chapter.

Caching
Completed chapter summaries are cached locally. Re-running a job skips already-summarized chapters by default. The cache manager lets you inspect, clear, or export cached entries per book.

Output language
Summaries can be generated in the book's source language or translated to any target language you specify.

UI translations currently available: Simplified Chinese, Vietnamese, Serbian, Brazilian Portuguese, European Portuguese, Italian, French, Persian, Spanish, and German. Contributions are welcome.

Compatibility
Calibre 8.16.2 or later
Windows, macOS, Linux (tested on Windows — feedback welcome for other platforms)

License
GNU GPL v3

Release notes

Spoiler:

Version 1.4.0
- Additional fixes for language drifts

Version 1.3.3
- Fix auto-append logic when batch mode is enabled

Version 1.3.2
- When appending to comments or custom column, omit run information

Version 1.3.1
- Fix a variable-scoping issue.

Version 1.3
- Fix language drift and redundancy on some highlights / key events section
- UI fixes

Version 1.2
- Fix an infrequent node selection issue that would cause content beyond the selected chapter boundaries to be sent - would happen only with some ToC structures.
- Fix ToC nodes count not showing below Pages number. Implemented lazy loading to help performance.
- Cache information: implemented lazy loading to help performance.
- Miscellaneous fixes

Version 1.1
- New: custom column support for the Append to comments feature: choose any Long text, like comments column, or create it instantly by clicking the plus button on config setting.
- Miscellaneous fixes

Version 1.0.0

• Initial release

[Comfy.n]

reserved

[Wiggo]

Very nice!

I’m not yet sure which output format to choose. Which one do you use and why?
I also have an annotation column (long text, like comments), but I cannot choose it to append the summary.

...and a small glitch.

[Comfy.n]

Originally Posted by Wiggo

Very nice!

I’m not yet sure which output format to choose. Which one do you use and why?
I also have an annotation column (long text, like comments), but I cannot choose it to append the summary.

...and a small glitch.
Attachment 224019

Great idea, and thank you for the heads-up!

The option to append to comments was a simpler approach I chose to implement, but indeed a long text dedicated column looks more fitting.

I have .MD format set for data files, and EPUB for the summaries added to library. This choice was made mostly for testing purposes but I'll probably keep it...

[Comfy.n]

@Wiggo, I forgot to address the small glitch observation: please check if resizing the window solves it. Custom column support is implemented and soon ready for release.

[Wiggo]

Comfy.n! Of course resizing window doesn't change anything.

Edit: ChatGPT says "Schlüsselaspekt" would be a good replacement for all three Ereignis, Erkenntnis und Kernidee to make it shorter

[Comfy.n]

Version 1.1

- New: custom column support for the Append to comments feature: choose any Long text, like comments column, or create it instantly by clicking the plus button on config setting.

- Miscellaneous fixes

[Wiggo]

Thx!

I find your PI very useful.
The second part of a series has just come out this June, but I could only vaguely remember the first one. Chapter summary – voilà

[Comfy.n]

Originally Posted by Wiggo

Thx!

I find your PI very useful.
The second part of a series has just come out this June, but I could only vaguely remember the first one. Chapter summary – voilà

Glad you like it.

[Wiggo]

Comfy.n, which of the both free ones would you recommend, Ollama or LMStudio? I’m not familiar with this topic at all yet.

[Comfy.n]

Welp, never had much luck with Ollama. LM Studio works fine. I plan to list some models I used on tests and delivered good results. Some tiny Gemma models sized less than 1GB worked flawlessly with LM Studio.

[Wiggo]

Thx!

Which model would have been the next question.

[MuffyVS]

I had no problem getting Ollama up and running and use it now. Have also heard good things about LM Studio. I settled on Gemma 3:12b myself but you will want to play around with a few models to see what works best with your hardware. I found gemma 3:12b to be more conversational and fairly quick on my newer, but not optomised for local AI (no VRAM) windows computer. I tried several different sizes of gemma, qwen, and mistral to find the one that worked best for what I will use it for. Gemma 4 ran on my system but was too slow for me to want to use it. People who code will suggest a qwen model but that was one of the worst for me. It took a full 5 minutes to respond to my test prompt "Hi!" The thinking trace was bonkers. Anyway, play around with it, try the small models first and see how they go. The model files are small, you can have several models sitting on your harddrive, you only load 1 into ram at a time. Get a good logic problem to ask and compare the models. Can they correctly figure out the answer and how long does it take? That will get you where you want to go.

[Wiggo]

Thx!

[Comfy.n]

Version 1.2
- Fix an infrequent node selection issue that would cause content beyond the selected chapter boundaries to be sent - would happen only with some ToC structures.
- Fix ToC nodes count not showing below Pages number. Implemented lazy loading to help performance.
- Cache information: implemented lazy loading to help performance.
- Miscellaneous fixes

Oh, and the tiny model I mentioned above is this one:
https://lmstudio.ai/models/google/gemma-3-1b