[Metadata Source Plugin] BiblioSleuth AI

[terry.trent]

BiblioSleuth AI researches the exact edition of selected EPUB books with OpenAI web
search, then presents evidence-backed title, author, series/index, tag,
identifier, publication-date, publisher, and formatted-description suggestions
for field-level review—or a separately warned bulk-accept workflow—before
anything is written to Calibre.

Author: Terry Trent
License: MIT
Platforms: Windows, macOS, Linux
Minimum Calibre version: 7.0.0
Current version: 1.0.0
calibre-bibliosleuth-ai GitHub Repository

Requirements
- A selected Calibre book containing EPUB
- An OpenAI Platform API key with billing/credits
- Internet access

Installation
1. Download the single attached BiblioSleuth-AI.zip file without extracting it.
2. In Calibre, choose Preferences → Plugins → Load plugin from file.
3. Restart Calibre and add BiblioSleuth AI under Preferences → Toolbars & menus.
4. Open the arrow beside the BiblioSleuth AI icon and choose Configure BiblioSleuth AI.

Privacy and cost
Only selected OPF metadata and bounded text from confidently identified title
and copyright pages are sent. Unidentified pages and chapters are excluded; the
full book is not uploaded and EPUB files are not rewritten. API and web-search calls are
billable; the plugin displays reported usage and approximate cost. Persistent
keys use the operating-system credential vault rather than Calibre preferences.

Support
Post the BiblioSleuth AI/Calibre versions, operating system, optimization preset, and
a redacted error. Never post API keys or copyrighted book text.

Version History

Spoiler:

1.0.0 — Cross-platform CI and release reliability
- Fixed UTF-8 documentation tests on Windows
- Added concurrency cancellation and removed duplicate Dependabot branch/PR runs
- Added a pinned blocking Ruff correctness gate and retained broader Qlty reports
- Consolidated deterministic package/checksum verification after the test matrix

1.0.0 — Initial stable release
- Exact-edition web research and structured metadata review
- Background jobs, field-specific searches, editable proposals, series index, and sort-field updates
- Economy/Balanced/Thorough/Custom optimization, usage/cost reporting, and validated custom prompts
- Title/copyright-page-only disclosure, secure credentials, strict schemas, safe HTML, and hardened EPUB/XML handling
- Settings-aware session caching, cached model discovery, connection reuse, and explicit fresh research
- First-run setup, accessible settings/review, pending notifications, batch and bulk workflows, and session undo
- Privacy-safe performance statistics plus previewable, locally saved redacted diagnostic bundles
- Deterministic self-contained packaging with pinned dependencies and automated test/security/release gates