Open API overview
Public queries, RSS, torrents, and group automation.
AniBT provides public queries and feeds, plus API-key-authenticated publishing, deletion, and sync-status queries.
Prop
Type
The production base URL is https://anibt.net. Group API keys, personal RSS tokens, and login sessions serve different endpoints and are not interchangeable. Parse each endpoint’s envelope, nullable fields, and error format as documented.
The machine-readable contract is /openapi.yaml. Endpoint pages show request fields and response examples. Title matching is anonymous unless groupId is supplied, which requires a session and group membership.
Conventions and authentication
Rate limits and parameter bounds
HTTP caching and conditional requests
Troubleshooting
Torznab
Anime queries and ratings
scope.kind on /api/seasons/anime distinguishes season browsing, title search, and exact ID lookup. A global title search returns requestedSeason: null; each anime carries its own season. Premiere dates remain separate from season classification. premiere.precision is day, month, season, or year; a coarse date uses the start of that period.
Season browsing returns at most 200 items; title search returns at most 50. total counts the returned items and truncated=true means more results exist. Use /api/animes/search for pagination; see Anime search. rating uses a 0–10 scale and defaults to 5.0 when no valid rating is available; averageScore uses 0–100 and defaults to 50. Existing ani-rss can sort these scores. Older versions ignore truncation indicators and still display at most 50 search results.
Default ratings update automatically when a later sync obtains a valid source rating. Manually locked ratings remain unchanged. Metadata follows the existing sync frequency; page views and RSS subscriptions do not refresh upstream records individually.
Anime search
/api/animes/search, /search, /api/bgm/search, and the query of /api/seasons/anime share one title search. It compares every title and alias of an anime.
Before comparing, case and full-width forms are normalized, katakana folds to hiragana, and Traditional and Japanese shinjitai characters fold to Simplified. Spaces and punctuation are then ignored. ふりーれん finds katakana titles, and 進擊 finds Simplified titles.
Matches are ranked by tier, then by popularity (descending), then by anime ID (ascending):
- Exact: a title or alias equals the query.
- Prefix: a title or alias starts with the query.
- Substring: a title or alias contains the query.
- Multi-word: every term hits some title or alias of the same anime, in any order and possibly across different titles. Used only for 2–16 terms; with more terms this tier is skipped, without an error.
- Ordered characters: 2–4 Chinese or Japanese characters appear in order in one title or alias, not necessarily next to each other.
进巨finds 进击的巨人. - Similarity: titles or aliases spelled similarly.
/api/animes/search, /search, and /api/bgm/search return results in this order. /api/seasons/anime uses it only to choose up to 50 anime; its response is still grouped by weekday and ordered by air time and title within each group.
Limits:
- Spelling tolerance comes only from similarity. Long titles tolerate small differences, such as
kimetu no yaiba; a transposition in a short word, such asfreiren, may find nothing. - All punctuation separates terms, so
Re:ZeroandRe Zeroare the same query. A year has no special rule; it matches like any other term. - A query of 1–2 characters uses no substring or similarity matching. Whether it matches a given title or alias depends on how the two relate:
- A query equal to the whole title or alias, ignoring spaces and punctuation, always matches.
- Latin letters and digits must be a whole word or the start of a word in it.
rkandordon't match letters inside a word. - Chinese and Japanese must be a single character in it or two adjacent characters. Punctuation in the query is removed first, so
ス・ダmatches a title or alias with those two kana next to each other, not one that writes them apart with punctuation (unless the whole title or alias is just those two kana). Two characters with no space or punctuation between them can also match by ordered characters, as above. - A mix of Chinese or Japanese with letters or digits, such as
人0, matches only an identical whole fragment: a piece set off by spaces or punctuation.
- A query that folds to nothing, such as
☆★, returns an empty last page.
Anime search pagination
/api/animes/search accepts q, format, status, seasonYear, limit (default 24, range 1–50), and cursor. It returns page, isDone, and continueCursor directly, without an ok or data wrapper. Like the search page's input, q is at most 300 characters; a longer one returns 422 VALIDATION_ERROR.
continueCursor is an opaque token and an empty string on the last page. Pass it back unchanged; don't parse it or build your own.
These cursor values return 422 VALIDATION_ERROR:
- A self-built
ts_page:Nor any other malformed value. - A value not issued for the same query, filters, and search policy.
- A value sent with an empty query.
Cursors issued before the October 31, 2026 update no longer work. On 422, drop cursor and start again from the first page.
Release search
/magnets?q, an anime page's release search, other-release categories, /rss/magnets.xml?q=, /rss/other/<category>.xml?q=, and /torznab/api?q= follow these rules. Every term must hit, with no typo tolerance, no prefix completion, and no dropped terms (for the /magnets?q exceptions, see Relaxed search and Episode notation):
- An anime release is compared on its own title and spec (episode label, resolution, format, subtitle type, language), on every title and alias of its anime, and on the name and slug of its subtitle group. Different terms can hit different places, so any alias of the anime finds releases titled only in another language.
- An other-category release is compared only on its own title and category and on the name and slug of its subtitle group.
- A subtitle group's name and slug match as whole words like any other text, with no typo tolerance or prefix completion. A quoted phrase must sit entirely in the name or entirely in the slug, not half in each. A group whose status is Suspended produces no hit; an Inactive group still does. Since November 10, 2026, a keyword RSS feed or Torznab search whose keyword is a group's name or slug also returns that group's releases whose titles don't carry the name.
- Latin letters and digits must be whole words:
HEVdoesn't findHEVC,1080doesn't find1080p, andLoliHousdoesn't findLoliHouse. - Chinese and Japanese match as contiguous characters and need not be a whole word:
鬼灭finds releases of 鬼灭之刃. - Letters and digits ignore case and width on every surface, and kana and Traditional/Simplified characters fold as in anime search.
Results are listed only by the chosen sort, newest first by default, with no relevance ranking. Torznab always sorts by publish time, newest first.
Only when the strict first page is empty does one split/join fallback run: Loli Hose finds LoliHose, rezero finds Re:ZERO, and Re:Zero finds ReZero. The fallback relaxes only the boundary between adjacent words, never exclusions, and later pages keep the first page's matching mode.
Relaxed search on /magnets
For anime releases, when the first page of /magnets?q is still empty after strict matching and the split/join fallback above, the search runs once more with relaxed matching, and the page says the results are relaxed. The sort stays the same and no term is dropped. The steps run in order, and the first with results wins:
- Word prefix: the last part of a term may be just the start of a word if it has no digit and at least 3 characters:
HEVfindsHEVC, andLoliHousfindsLoliHouse. A part with a digit isn't completed, so1080still doesn't find1080p. - Anime title similarity: a single spelling difference in an anime title may match:
kimetu no yaibafinds releases of Kimetsu no Yaiba. The terms with no digit must total at least 6 characters (4 if all Chinese or Japanese); a term with a digit doesn't count, and a shorter query skips this step. Terms with a digit, terms of 1–3 characters, and multi-part terms such as quoted phrases must still hit strictly:kimetu no yaiba 1080plists only 1080p releases,kimetsu no yaiba 2160pstays empty when no 2160p release exists, and"kimetu no yaiba"in quotes is empty too. A release's own text, such as a group name, gets no such tolerance.
Neither step applies prefix or spelling tolerance to a subtitle group's name and slug; they still match as whole words.
If the relaxed read fails on the first page, the answer is the strict search's empty result, not an error; if it fails while paging through relaxed results, the answer is a 503 like any other search. An anime page's release search, other-release categories, RSS, and Torznab always match strictly, so a term that only finds results on /magnets through relaxed matching may be empty in RSS.
Episode notation on /magnets
/magnets?q reads only the last word of the query. When it is 第N集, 第N话, 第N話, EP+N, E+N, or #+N (N is 1–4 digits, leading zeros allowed, case-insensitive), that word is no longer matched as text and becomes the episode filter: Mushoku Tensei 第5集, Mushoku Tensei EP05, and Mushoku Tensei #5 give the same results as searching Mushoku Tensei filtered to episode 5. The search box and the address keep what you typed. When the strict result is empty, the relaxed search still runs and keeps the episode filter.
- Only the last word is read, and at least one other searchable word must remain:
第5集on its own is a text search. A quoted or excluded notation ("第5集",-第5集) and a bare number (Mushoku Tensei 05) are not converted. - The filter lists releases whose episode number equals N. A batch counts under its first episode:
第1集also lists a1-12batch, and第5集does not. A release a group numbered by series number (53 for a third season's fifth episode) is not found by第5集. - Only the
/magnetspage reads a query this way. Keyword RSS matches every word ofqas text and filters by episode through a separateepisode=parameter, where a batch also counts under its first episode (see Site-wide magnet RSS). Torznab doesn't recognize the notations either; it reads only trailing 2–4 digits as an episode. - The RSS link the
/magnetspage offers for such a query carries the episode: forMushoku Tensei EP05it is/rss/magnets.xml?q=Mushoku+Tensei&episode=5, so the feed filters by the same terms and episode as the page. Without a notation, the link carries the query as typed. When a subtitle-type filter is selected on the page, the link also carriessubtitle=, afterformatand beforeepisode:/rss/magnets.xml?q=Mushoku+Tensei&subtitle=INTERNAL&episode=5, so the feed filters by the same subtitle type as the page. The link leaves outlimit(the default is already 100), so when its parameters are canonical the feed answers it directly, without a308. The feed still differs from the page in two cases:- When the strict match finds nothing, the page shows relaxed results; the feed never relaxes, so it is empty.
- When the text left after removing the notation has no searchable word (
. EP05), the page searches the original text, while the link keeps the converted form and the feed is empty.
Punctuation, quotes, and exclusions
- A term with no digit splits at its punctuation, and each part matches on its own:
Nix-Raws,Re:Zero, and鬼灭/第二季are two terms each, so one can hit an anime alias and the other a release title. - A term with a digit and several parts is one phrase: the parts of
HEVC-10bit,x264-AAC, and第2期must be adjacent and in order within one text (the release's own title and spec, one title or alias of its anime, or its subtitle group's name or slug).HEVC-10bitdoesn't find10bit x264 HEVC, and第2doesn't match第20期. - A complete quoted phrase standing as its own term is one phrase and may contain spaces:
"25 1080p". A quote attached to other characters, or left unclosed, is ignored and its words are ordinary terms. -wordexcludes, and reads only the release's own title and spec, never anime aliases or a subtitle group's name and slug.-web-dlneedswebanddladjacent and in order, so it doesn't excludeWebDLwritten as one word.-"25 1080p"excludes that phrase.- A
-with whitespace on both sides is a separator:Title - 25searches for25, andTitle -25excludes25. - Use
-and"for exclusions and quoted phrases. The full-width-,﹣, and"act as these two on every surface.−and curly quotes are not operators.
Term limits and empty queries
A query carries at most 16 positive terms and 8 exclusions. A quoted phrase or a word with a digit counts as 1, a/b/c counts as 3, separators and symbol-only tokens don't count, and each exclusion counts as 1. Past the limit, each surface answers:
Prop
Type
A query with no positive term (-, ☆, or only exclusions such as -25) is not an error. Pages return an empty page, RSS returns a valid empty channel, and Torznab returns an empty list with total 0. Such a query issues no cursor, and a cursor sent with it is treated as invalid.
When an empty q= is sent, /magnets lists all releases, Torznab lists the latest releases, and RSS returns 400.
Release search pagination
- A stale
/magnetspage link returns to the first page of the current view, including search page links saved before the October 31, 2026 update. - RSS has no pagination. Torznab uses a bounded
offset.
When search is unavailable
When a search times out or can't complete, a surface never returns partial results or substitutes an empty result. The one exception is a failed relaxed read on the first page of /magnets?q: the strict search has already found nothing, and that empty result stands.
Prop
Type
The table covers search itself being unavailable. When the rate limiter of these JSON endpoints temporarily cannot count, they return a separate 503 with error.code SERVICE_UNAVAILABLE and Retry-After: 15; see Rate limits and parameter bounds.
Airing status
Automatic completion requires exact premiere and end dates, plus another source verification confirming the end after that entire day has elapsed. Dates are compared in Japan time. A year, month, season, or anime format alone cannot establish completion.
Automatic status is recalculated when synchronization receives a corrected airing plan. Manually locked statuses and explicit hiatus or cancellation take priority. Titles without sufficient evidence have an unknown status; unsupported historical statuses and planned end dates are not proof of completion.
The homepage shows only the latest episode of each anime on a given day. Genuine releases on different days remain separate.
