Start work on general book related AI chat widget

src/calibre/gui2/dialogs/llm_book.py

@@ -0,0 +1,168 @@
+#!/usr/bin/env python
+# License: GPLv3 Copyright: 2025, Kovid Goyal <kovid at kovidgoyal.net>
+
+from collections.abc import Iterator
+from functools import lru_cache
+from typing import Any, NamedTuple
+
+from qt.core import QDialog, QUrl, QVBoxLayout, QWidget
+
+from calibre.ai import ChatMessage, ChatMessageType
+from calibre.ebooks.metadata.book.base import Metadata
+from calibre.gui2 import Application, gprefs
+from calibre.gui2.llm import ConverseWidget
+from calibre.gui2.viewer.config import vprefs
+from polyglot.binary import from_hex_unicode
+
+
+def format_book_for_query(book: Metadata, is_first: bool, num_books: int) -> str:
+    which = '' if num_books < 2 else ('first' if is_first else 'next')
+    ans = f'The {which} book is: {book.title} by {book.format_authors()}.'
+    left = get_allowed_fields() - {'title', 'authors'}
+    if 'series' in left:
+        ans += f' It is in the series: {book.series}.'
+        left.discard('series'), left.discard('series_index')
+    if 'tags' in left:
+        ans += ' It is tagged with the following tags:' + book.format_tags() + '.'
+        left.discard('tags')
+    comments = []
+    fields = []
+    for field in left:
+        m = book.metadata_for_field(field)
+        if field == 'comments' or m['datatype'] == 'comments':
+            comments.append(m.get('field'))
+        else:
+            fields.append(book.format_field(field))
+    if fields:
+        ans += ' It has the following additional metadata.'
+        for name, val in fields:
+            ans += f' {name}: {val}'
+    if comments:
+        ans += ' Some notes about this book: ' + comments
+    return ans
+
+
+def format_books_for_query(books: list[Metadata]) -> str:
+    ans = 'I wish to discuss the following book. '
+    if len(books) > 1:
+        ans = 'I wish to discuss the following books. '
+    for i, book in enumerate(books):
+        ans += format_book_for_query(book, i == 0, len(books)) + '\n---------------\n\n'
+    return ans
+
+
+def get_allowed_fields() -> set[str]:
+    return set()
+
+
+class Action(NamedTuple):
+    name: str
+    human_name: str
+    prompt_template: str
+    is_builtin: bool = True
+    is_disabled: bool = False
+
+    @property
+    def as_custom_action_dict(self) -> dict[str, Any]:
+        return {'disabled': self.is_disabled, 'title': self.human_name, 'prompt_template': self.prompt_template}
+
+    def prompt_text(self, books: list[Metadata]) -> str:
+        pt = self.prompt_template
+        return pt.format(
+            books=format_books_for_query(books),
+            books_word='book' if len(books) < 2 else 'books',
+            plural_word='is' if len(books) < 2 else 'are',
+        )
+
+
+@lru_cache(2)
+def default_actions() -> tuple[Action, ...]:
+    return (
+        Action('summarize', _('Summarize'), '{books} Provide a concise summary of the previously described {books_word}.'),
+        Action('chapters', _('Chapters'), '{books} Provide a chapter by chapter summary of the previously described {books_word}.'),
+        Action('read_next', _('Read next'), 'Suggest some good books to read after the previously described {books_word}.'),
+        Action('universe', _('Universe'), 'Describe the fictional universe the previously described {books_word} {plural_word} set in.'
+               ' Outline major plots, themes and characters in the universe.'),
+        Action('series', _('Series'), 'Give the series the previously described {books_word} {plural_word} in.'
+               ' List all the books in the series, in both published and internal chronological order.'
+               ' Also describe any prominent spin-off series.')
+    )
+
+
+def current_actions(include_disabled=False):
+    p = gprefs.get('llm_converse_quick_actions') or {}
+    dd = p.get('disabled_default_actions', ())
+    for x in default_actions():
+        x = x._replace(is_disabled=x.name in dd)
+        if include_disabled or not x.is_disabled:
+            yield x
+    for title, c in p.get('custom_actions', {}).items():
+        x = Action(f'custom-{title}', title, c['prompt_template'], is_builtin=False, is_disabled=c['disabled'])
+        if include_disabled or not x.is_disabled:
+            yield x
+
+
+class LLMPanel(ConverseWidget):
+    NOTE_TITLE = _('AI Assistant Discussion')
+
+    def __init__(self, books: list[Metadata], parent: QWidget | None = None):
+        self.books = books
+        super().__init__(parent)
+
+    def handle_chat_link(self, qurl: QUrl) -> bool:
+        match qurl.host():
+            case self.quick_action_hostname:
+                name = from_hex_unicode(qurl.path().strip('/'))
+                for ac in current_actions():
+                    if ac.name == name:
+                        self.activate_action(ac)
+                        break
+                return True
+        return False
+
+    def activate_action(self, action: Action) -> None:
+        self.start_api_call(self.prompt_text_for_action(action))
+
+    def choose_action_message(self) -> str:
+        msg = '<p>'
+        if len(self.books) > 1:
+            msg += _('{0} books selected, starting with: <i>{1}</i>').format(len(self.books), self.books[0].title)
+        else:
+            msg += _('Selected book: <i>{}</i>').format(self.books[0].title)
+        msg += self.quick_actions_as_html(current_actions())
+        msg += '<p>' + _('Or, type a question to the AI below, for example:') + '<br>'
+        msg += '<i>Discuss the literary influences in this book</i>'
+        return msg
+    ready_message = choose_action_message
+
+    def get_language_instruction(self) -> str:
+        if vprefs['llm_localized_results'] != 'always':
+            return ''
+        return self.language_instruction()
+
+    def create_initial_messages(self, action_prompt: str, **kwargs: Any) -> Iterator[ChatMessage]:
+        context_header = format_books_for_query(self.books)
+        context_header += ' When you answer the questions use markdown formatting for the answers wherever possible.'
+        if language_instruction := self.get_language_instruction():
+            context_header += ' ' + language_instruction
+        yield ChatMessage(context_header, type=ChatMessageType.system)
+        yield ChatMessage(action_prompt)
+
+    def prompt_text_for_action(self, action: Action) -> str:
+        return action.prompt_text(self.books)
+
+
+def develop():
+    app = Application([])
+    d = QDialog()
+    l = QVBoxLayout(d)
+    l.setContentsMargins(0, 0, 0, 0)
+
+    llm = LLMPanel([Metadata('The Trials of Empire', ['Richard Swan'])], parent=d)
+    l.addWidget(llm)
+    d.exec()
+    del app
+
+
+if __name__ == '__main__':
+    develop()