Librarr 1.1.0-beta

First release since 1.0.0-beta, and the first with images for ARM.

Docker

docker pull ghcr.io/rorqualx/librarr:1.1.0-beta   # or rorqualx/librarr:1.1.0-beta

One manifest covering linux/amd64, linux/arm64 and linux/arm/v7 — Docker picks the architecture for you, so a Raspberry Pi and an x86 server run the same command. 1.0.0-beta was amd64-only.

arm64 is runtime-verified on real aarch64 hardware: healthcheck, UI, live Open Library search, root-folder scan, a Library Import run and a full discography refresh, no errors. arm/v7 has only been exercised under emulation, where QEMU's own instruction translator fell over rather than anything in Librarr — treat 32-bit ARM as untested, and please report either way if you run it.

Added

Library Import wizard (Library → Library Import). Point Librarr at a root folder holding books you already have; every folder inside it that no author occupies becomes a row, pre-matched against Open Library with an editable search box where the guess is wrong. Each author is attached to the folder already on disk, so nothing is moved or re-created. Readarr never shipped this page — it was dropped when Readarr forked from Lidarr.

Rescan button per root folder (Settings → Media Management), so an existing collection can be picked up without waiting for the scheduled task.

Fixed

Author search returned OpenLibrary's ranking unchanged, and the wizard auto-selects the first result. Searching Tolkien, J.R.R. — an ordinary Calibre folder convention — put a 1-work archaeology report ahead of the real 355-work J.R.R. Tolkien, because the report's title happens to contain (Tolkien, J. Results are now ranked on name match, with work count only as a bounded tiebreak so a prolific unrelated author can never displace an exact match. Nothing is filtered out; a stub record is demoted, never hidden.

Same-named authors were impossible to tell apart. OpenLibrary carries three "Stephen King" records — 606 works, 48 works, and 7 works by the author of Principles of Macroeconomics. The work count and best-known title were already in the response and were being discarded. The wizard now shows them.

Interactive search returned HTTP 500 for any book with no monitored edition — measured at 42% of books in a freshly added library.

Cover fetches could be rejected by OpenLibrary. Only the ISBN-keyed form is metered, and only that form is now throttled.

Docker images were never version-stamped, so every containerised install was classified as a non-production build.

Changed

Librarr now identifies itself honestly to OpenLibrary, Wikidata and Audnex with a real User-Agent carrying app name, version and contact URL, per each service's stated policy.

No Windows build in this release

The Windows package needs the net6.0-windows target framework, which only publishes from a Windows runner. CI is Linux-only, and build.sh skips the package rather than ship one missing Readarr.Windows.dll. Windows users should run the Docker image or build from source until a Windows leg is added to the release matrix.

Installing

Verify downloads against SHA256SUMS first. Upgrading an existing Readarr library: mount your old config directory at /config and start it — see docs/migrating-from-readarr.md. Coming from a folder of books with no database, use the new Library Import wizard instead.

Full detail in CHANGELOG.md.