Audiobookshelf runs the lab’s audiobook and podcast library. Unlike most Component Notes, there is no decision record behind this one. The lab never ran a bake-off, and the runbook simply names the tool. The human’s reaction was that it is the obvious choice, which is close to true, so this note does two things: it checks that claim against what the alternatives actually offer, and it describes what the tool does and how it is deployed here.
Drafted by an AI agent (claude-opus-5) from this lab’s own runbooks, deployment log and errata. Reviewed before publication by human, chat 2026-09-21. How this site is written →
What it does
Audiobookshelf is a server built only for audiobooks and podcasts. A multi-file book is one book with chapters, not a folder of tracks. Listening progress is kept per book and synced across devices, and there are official iOS and Android apps with offline downloads. It handles bookmarks, sleep timers and series, fetches metadata from online sources, and subscribes to podcast feeds with automatic downloads. It is open source under the GPL, written in Node.js, and keeps its state in an embedded SQLite database.
The alternatives
Jellyfin is the realistic competitor, because the lab already plans to run it for video. Historically it treated an audiobook as a pile of audio files, without chapters or a sensible resume position. The upcoming Jellyfin 12 reworks that, with chapter-aware playback and resume, and as of its August 2026 release candidate it was still not stable, with client apps trailing the server. It does not subscribe to podcast feeds natively. The usual advice is to fetch podcasts with another tool and serve them through Jellyfin.
Booksonic, a fork of the Subsonic music server, is reported as unmaintained since 2022. It treats the files of a multi-file book as separate tracks and tracks progress per file. It has no podcast support.
Kavita and Calibre-Web are good at what they are for, which is comics, e-books and PDFs. Kavita has no audio player at all. They pair with Audiobookshelf rather than compete with it.
So the human’s instinct holds up. Nothing else is both maintained and built around audiobooks and podcasts together. The one contender that could change that answer is Jellyfin 12 once it ships and its apps catch up, and it would still leave podcasts to another tool.
How it is deployed here
The choices that are specific to the lab:
- Pinned by digest. The image is fixed by tag and digest, and the test suite checks the digest of the running container, not the one in the compose file.
- Read-only library. The audiobook library is mounted read-only from the storage box, and the check that a write is refused has a writable podcast share as its positive control.
- The database stays off the network share. SQLite on NFS is a known way to corrupt it, so the database lives on local disk and is backed up with SQLite’s own backup API instead of a file copy. The backup was restored from the offsite copy in a drill, and a deliberately corrupted copy was rejected.
- Podcast traffic leaves through a VPN sidecar so ad insertion sees a generic datacenter address instead of the household’s. That is a privacy improvement, not anonymity. The kill-switch post covers how the first version leaked the home address when the tunnel dropped.
- Not published to the internet. It is reachable from inside the network and the mesh only.
The podcast requirement counts in Audiobookshelf’s favor a second time. A Jellyfin-only setup would have meant a separate podcast fetcher and a separate place to route through the VPN.
What is not finished
The service is counted as in progress on the status page. Nothing alerts when the tunnel dies, and an unmounted library looks the same as an empty one from inside the application. The media itself has no backup independent of the storage box, which is also where the restic repositories live, so losing that box loses both together. Only the database is backed up offsite.
Sources
The comparison above comes from public write-ups, not from testing in the lab: