Skip to content

How it works

Open data goes in, a schedule keeps it current, and every fact is checked before you see it.

557.9K
Songs
67.9K
Artists
5.3M
Facts
178.8K
Pictures

Where the data comes from

All of it is open or openly licensed. Each row keeps its source and licence.

MusicBrainz
Songs, artists, releases and the credits between them: who wrote, produced and recorded what. We read its official database dump and ask its web service for details.
Wikidata
Facts about artists, genres and places, and artist portraits from Wikimedia Commons.
Wikipedia
Article text that claims are read from. Every claim taken from it must quote the article word for word.
Cover Art Archive
Album covers. Deezer and iTunes fill the gaps it leaves.
AcousticBrainz
Tempo, key and danceability of songs.
ListenBrainz and Deezer
Charts, fresh releases and similar artists for For you. Deezer also supplies the 30-second previews.
Last.fm and ListenBrainz accounts
Only if you connect one: your recent plays and top artists, to tune your picks.

How a fact gets onto a page

  1. Collect

    We store each claim with its source name, link, exact quote and licence. Credits come straight from MusicBrainz and Wikidata records, with no model in between.

  2. Check

    A claim taken from text is kept only if its quote appears word for word in the stored source. A second AI model, a different one, then checks that the quote supports the claim.

  3. Count

    Insights are counted from approved claims: how often two people worked together, which studio an artist used most. Each one shows how many sources stand behind it.

  4. Connect

    The graph of songs, artists, people, studios, labels and places is rebuilt from approved claims only. Nothing is guessed: places and genres need a source too.

How it stays current

Background jobs run all day. This is the current schedule, and it can change.

Every few minutes
The pages everyone sees the same, like the home page, Discover and the Explore starters, are rebuilt ahead of time. Artists that people searched for and we lacked are added. Missing artwork is filled in.
Every 15 to 30 minutes
New MusicBrainz relations and Wikidata facts become claims. Cover art and portraits are fetched and verified. New plays from connected accounts are read.
Every hour
The feed is compiled: anniversaries, connections, facts and threads. Tempo, key, genres and place coordinates are resolved. Checks make sure that nothing is shown without a claim behind it.
Every 2 hours
Wikipedia text is read for new claims.
Every day
The graph is rebuilt, and each page gets its coverage level. Charts, fresh releases and similar artists are refreshed for For you. Stored artwork is re-checked, and broken links are marked.
From time to time
The MusicBrainz database copy is reloaded from its published dump, and new songs are promoted from it.

How For you picks music

It learns from what you do, and every pick says why it is there.

  • Your signals: the artists you pick, the songs you swipe, and, if you connect them, plays from Last.fm or ListenBrainz. Your picks are rebuilt within minutes of new plays or swipes.
  • Where candidates come from: artists that listeners with a similar taste pair with yours, songs that share a producer, label or studio with what you like, new releases, and the charts.
  • How they are ordered: by your Tune settings, with a shuffle that stays stable for a day, so a page does not reshuffle while you scroll. No artist appears twice in a row.
  • What does not count: money. Nobody can pay for a place in your picks.

Open For you

How Explore works

Explore draws the graph around what you search. Lines are connections that have a source, and each result shows why it matches. You can arrange the map as a network, an orbit, clusters, a flow or a timeline, or switch to Ask and put a question in plain words.

Open Explore