Feature
Discover: recommendations built from what you already listen to
Most recommendation engines start from scratch on a cold server. Aurral starts from your library and your history, which means it produces picks on day one and gets better as the history grows.
The Discover page is the home screen. It carries personalised recommendations, trending artists, genre tags, recent and upcoming releases, playlists, artist news and nearby shows. Every user chooses which of those sections appear and in what order, because two people running the same server want different things on their front page.
Where the seeds come from
A recommendation needs somewhere to start. Aurral collects a set of seed artists for each user from three places, and looks up artists similar to all of them.
- Artists you marked More like this. The most explicit signal you can give, so it carries the most weight.
- Your listening history, whether recorded inside Aurral or imported from Last.fm, ListenBrainz or Koito.
- A sample of your library. Recently added artists are always included so a new album immediately influences your picks. The rest of the sample changes on every refresh, which keeps the list from settling into a fixed rotation.
How the ranking actually works
Candidates are scored on three axes: how similar the artist is to your seeds, how well their genres line up with your seeds, and how many of your seeds point at them. That third axis is what stops the result collapsing into one narrow genre. An artist that four unrelated seeds all point at outranks one that a single seed matches perfectly.
Genre matching uses MusicBrainz tags, but only the ones MusicBrainz marks as a genre, plus a small set of broad fallbacks such as indie and rap. Descriptive tags like seen live or female vocalists are ignored on purpose. If they counted, a handful of mood tags would dominate every recommendation you got.
The match percentage, honestly explained
Each recommendation shows a percentage. It is not an external score borrowed from another service. It is that artist's position in your current list, scaled from 99 percent at the top to 55 percent at the bottom. It exists so you can read the ordering at a glance, and it moves as soon as you give feedback.
Trending artists deliberately carry no percentage. They are the same for everybody, so a number that implied personal relevance would be a lie.
Four ways to shape the list
Feedback reorders your list immediately and is saved for the next scheduled rebuild, which then uses the artists you liked as fresh seeds. That is what makes the loop compound rather than reset.
| Action | Immediate effect | Effect on the next rebuild |
|---|---|---|
| More like this | Artist moves up, similar artists move with it | Becomes a seed, so its neighbours get pulled in |
| Less like this | Artist and related picks move down | Removed from the seed pool entirely |
| Block artist | Removed from the page | Excluded from previews and future downloads |
| Nothing | Order stays as ranked | Library sample rotates, so new material surfaces |
Discovery modes
An admin sets one global dial in settings, and every user is affected the next time their recommendations rebuild.
- Safer stays close to what you already know. Good when your library is narrow and you want it to stay coherent.
- Balanced mixes familiar picks with new ground. The sensible default.
- Deeper deliberately reaches further outside your library. Expect more unfamiliar names, and more that you will not keep.
Last.fm or ListenBrainz, and why it matters
With no key at all, Aurral finds similar artists through ListenBrainz. That works, but ListenBrainz similarity leans towards well-known names, so the results skew mainstream. Adding a Last.fm key in settings switches similarity, trending, tag search and flows to Last.fm, and normally brings the picks closer to what you are actually into. Adding or removing the key rebuilds everyone's recommendations.
Releases, news and shows
The recent and upcoming releases list is built from a hosted metadata service and refreshed at startup, once a day, after each library scan and whenever you add an artist. Albums already in your library are left out, and if the service is unreachable Aurral keeps the last list it loaded rather than showing you nothing.
Artist news comes from RSS feeds an admin enables. A story is highlighted when an artist's full name appears in the headline, or when the publisher tagged it with that artist. Single-word names like Ghost or Live are ordinary words in English, so for publishers that do tag their stories, Aurral only highlights them when the tag names the artist. Stories are kept for 30 days even after they leave a feed.
With Ticketmaster configured, Discover also shows nearby events for recommended, trending and library artists.
Frequently asked questions
Why does the same recommendation look different for two people?
Because seeds come from your own library and history. Two people with identical libraries still have different listening history, so the ranking differs. Trending artists are the exception: those are global.
What does the match percentage actually mean?
It is the artist's rank inside your own list, mapped onto a scale from 99 percent for your top pick down to 55 percent for the last one. It is not a ListenBrainz or Last.fm score, and it changes when you give feedback.
Why do some tags never show up as genres?
Only tags that MusicBrainz marks as a genre count, plus a short list of broad ones like indie and rap. Tags like seen live or female vocalists are deliberately excluded so they do not distort genre matching.
Do I need an API key for recommendations?
No. Discover works with no account at all, using ListenBrainz for similarity and MusicBrainz for genres. An admin can add a Last.fm key, which usually gives closer matches and requires a rebuild.
What happens when I block an artist?
They disappear from your recommendations, playlist previews and future downloads. Nobody else is affected, because the blocklist belongs to your account rather than the installation.
How long until my recommendations are ready?
A new account sees a building state until the first set is generated. After that, an admin schedules how often recommendations rebuild, and your feedback from the previous cycle seeds the next one.
Related pages
Turn recommendations into scheduled playlists
Flows mix Discover with your library on a schedule and fetch the tracks.
When your library stopped being fun
The three settings that fix a stagnant library fastest.
Going deeper without the slop
Deep Dive, Focus sources and discovery modes, used together.
Last.fm integration
Why an API key gives noticeably better similar-artist matches.
See what your own library would recommend
Aurral indexes your files and builds a first set of recommendations from them. It takes one scan and no external account.