✨ Replace the Google Books API with Open Library

[Logan2234]

Open Library becomes the sole book catalogue source, replacing the Google Books API. It is keyless and unmetered (Google Books required an API key and capped the free tier at 1,000 requests/day) and imposes no branding requirement, so the "Powered by Google" marks give way to plain courtesy attribution.

Ships as 1.6.0 (version bumped in lockstep, CHANGELOG.md entry added, Quackback changelog left in draft — publishing it sends the release newsletter, so that's Logan's call).

The provider

OpenLibraryProvider replaces GoogleBooksProvider, covering the same four capabilities:

Capability	How
search(query)	/search.json (Solr), explicit fields so the response stays small
searchByIsbn(isbn)	q=isbn:…
searchByIsbns(isbns[])	q=isbn:(A OR B …), batches of 20, mapped back through each doc's edition ISBN list
getDetails(id)	/works/{id}.json (description) + search.json?q=key:/works/{id} (page count, publishers, subjects, rating)

Notable behaviours:

• Merged works redirect. Open Library serves a type: /type/redirect stub for a work merged into another; details follow it (bounded hops) and key everything — summary, permalink, external id — off the canonical id, so the cache never stores a stale alias.
• Same-author suggestions resolve through the stable author_key and sort=rating, instead of Google's display-name matching.
• Subjects are filtered before becoming genre chips: Open Library's crowd-sourced list mixes real genres with machine tags (nyt:mass-market-monthly=2021-11-07), sentence-long entries and case variants — all dropped, capped at 10.
• Covers use the -L variant (~500px) rather than Google's ~128px thumbnails.
• Verified against the live API (search, ISBN, bulk ISBN, details, redirect) before the smoke test was removed; 15 unit tests cover the mapping.

Data model

BookSource.GOOGLE_BOOKS → OPEN_LIBRARY, addressed by work id (OL893414W).

BookItem is an on-demand cache keyed by the source's own id, so the migration drops the cached books rather than converting them — along with everything referencing one through a polymorphic, FK-less target that nothing cascades from: book reviews, comments, list items, activity events, book-linked notifications, and the reports/moderation decisions resolved through those reviews and comments. ReadingGoal references no book and is untouched.

⚠️ Breaking for self-hosters: book library entries don't survive the upgrade (BookEntry hangs off BookItem). No production data is affected. Other domains are untouched.

Trade-off accepted

Books lose their adult signal. Open Library exposes no equivalent of Google's MATURE maturity rating, so BookItem.isAdult is always false. The column, the filterAdultContent calls and the per-account gate all stay in place for the day a source provides one — documented at each site rather than removed.

Config

• GOOGLE_BOOKS_API_KEY removed from .env.example and docker/docker-compose.yml.
• MUSICBRAINZ_CONTACT → API_CONTACT, now shared by both keyless providers for the identifying User-Agent their usage policies ask for (Open Library and MusicBrainz).
• /admin/services: the Livres entry becomes keyless with no quota ceiling (no published limit), probed via search.json.

Also in here

• Fixes the seed writing its book call counter under google_books while QuotaTrackerService wrote googleBooks — the row was rendering under its raw key instead of a label on /admin/stats.
• Legal pages (mentions notice, privacy policy ×3, terms), settings → data sources, landing copy, Goodreads/StoryGraph import copy, README, apps/api/README, docs/data-model.md and the regenerated docs/erd.md.
• messages/{fr,en}.json: book_google_notice → book_openlibrary_notice, settings_datasources_googlebooks_notice → …_openlibrary_notice; the Google brand mark is dropped from provider-brands.ts (simple-icons has no Open Library glyph, so it's listed logo-less like OMDb).

Checks

pnpm test (665 passed), pnpm lint, pnpm --filter @loomkeep/web check (0 errors), pnpm knip (only pre-existing findings).

🤖 Generated with Claude Code

https://claude.ai/code/session_01WqHpRYcQuZ5safw87XgVPd

---

Generated by Claude Code

[Logan2234]

The knip check is red here, but it is not caused by this PR — it fails identically on main (latest main run, same knip job, conclusion: failure), and the job is continue-on-error: true so it doesn't gate the merge.

Cause: the knip job runs pnpm install + generate:paraglide but not pnpm build:package, unlike lint-build-test. Without packages/shared/dist, knip can't resolve how api/web consume the shared package, so it reports every exported DTO in packages/shared/src/dto/* as unused — ~200 findings spanning files this PR never touches (list.ts, music.ts, stats.ts, social.ts…).

Locally with a built dist, pnpm knip reports only 5 pre-existing findings, none introduced here.

The fix is one line in .github/workflows/ci.yml (a - run: pnpm build:package step before pnpm knip), but that's CI hygiene unrelated to the Open Library swap — happy to do it in a separate PR if you want it.

---

Generated by Claude Code