Skip to content

How recommendations work

Recommendations are computed on your server. Your reading is sent only to the trackers you connect. Anime lookups go to AniList and MyAnimeList, and detail cards fetch reviews from AniList, MyAnimeList and Kitsu.

Maki downloads a local copy of the MangaBaka database (about 350 MB, 3.5 GB on disk) and builds a vector for each series with a local ONNX embedding model. Series about similar things end up close together, which finds matches that shared genre tags would miss. See The local MangaBaka database.

Recommendations start from titles you already have. Seed them from your whole library or from specific titles, then filter by year, rating, type, status, genre, chapter count, content rating and obscurity, from popular to hidden gems.

The Browse tab is a stack of rails. Reorder or switch off any of them per user in the layout editor, and open any rail fullscreen with Show more. See Discover rails for each one.

  • Genre and catalogue rails browse the MangaBaka catalogue.
  • Readers like you looks at what people with similar reading go on to read.

The other tabs are Recommended, Your taste and Shortlist. Your taste shows your reading behaviour, grouped interests, the tags steering your picks and how your taste has moved. Shortlist holds titles you saved for later. See Discover.

Detail cards show categorized tags, alternate titles, creators, per-source ratings and reviews from AniList, MyAnimeList and Kitsu. The Discover page itself is described in Discover.