The JSON API

The dictionary as files a program can read. No key, no rate limit, no server: every URL below is a static file written when the site is built, served from the same place as the pages.

The endpoints

/api/v1/index.json

Every published entry in one file — 9,109 of them — as slug, cyrillic, latin, pos and gloss. Start here.

/api/v1/words/<slug>.json

One entry: the head-word in both scripts, its parts of speech, and its published senses with their definitions, labels and example sentences. The slug is the one in the page address, so /api/v1/words/at.json describes at.

/api/v1/forms/<slug>.json

Every form the engine builds for that entry — cases, possessives, degrees, and the full conjugation — each with the key that names the cell (case.dative.sg, owner.my.sg, degree.comparative, present.a.0). It is a separate file because a verb's paradigm is far larger than its definitions and most readers of the word file do not want it.

/api/v1/meta.json

The data version this build was made from, the entry, record and sense counts, the licence (identifier, URL and attribution line), and the list of endpoints.

Both scripts, and which one is the source

Karachay is written here in Cyrillic and in Karachay Latin, and the two are not equal: the Cyrillic is the source and the Latin is derived from it by rule, one way. Every Karachay string in the API therefore ships as a pair — cyrillic and latin, or cyrillic beside kb-style Latin on an example — so that no program ever has to romanise anything itself. There is no Latin-to-Cyrillic conversion anywhere in this project, deliberately: the dative of kün is written künnge with two н, and a program working backwards from the Latin spelling would write one.

What is in it, and what is not

The API carries exactly what the site shows: published entries and published senses. Drafts are not in it, unreviewed material is not in it, and neither are the unadapted Russian citation forms that are kept as source evidence but never presented as Karachay vocabulary. A paradigm table that an entry does not print is not in the forms file either — the forms it would contain are the ones being curated away.

The source dictionaries are cited on the site and never reproduced, so the API carries no citation quotes and no per-sense source list in version 1. Editorial fields — confidence, review status, corpus frequency — are also absent: they mean something inside the review process and nothing outside it.

The promise about version 1

The version is in the address. Inside /api/v1/ changes are additive only: new fields and new endpoints may appear, and an existing field will not be removed, renamed, or made to mean something else. If that ever has to happen, it happens at /api/v2/ and this tree stays where it is. Treat an unknown field as something to ignore rather than an error.

The content does change: definitions are corrected, senses are added, entries are published. dataVersion is on every document and names the state of the dictionary it was built from; compare it to know whether anything could have moved.

The four older files, which are not an API

The site has served machine-readable files for longer than this API: /search-index.json, /lexicon.json, /tagger.json and /sitemap.xml. The first three exist to make the search box, the analyzer and the tagger work. Their shape belongs to those features and changes with them, they are requested with a version stamp and cached as permanent, and nothing here promises they will keep the fields they have. Please build on /api/v1/ instead.

Being polite

There is no rate limit and no key, because there is no server to protect — these are static files behind a CDN, and they answer cross-origin requests from any page. Fetch what you need rather than the whole tree: index.json in one request is cheaper for both of us than 9,109 word files, and a forms file is only worth fetching for a word you are actually showing. Cache what you fetch; the files change when the dictionary does, not more often.

Licence: CC BY-SA 4.0

Everything this API serves is the dictionary data, and the dictionary data is licensed Creative Commons Attribution-ShareAlike 4.0 International (CC-BY-SA-4.0). You may copy it, build on it, and use it commercially, on two conditions: credit it, and release what you build on it under the same licence. meta.json carries the identifier, the licence URL and this attribution line:

Karachay dictionary (karachay.com) by Davud Kaplan, CC BY-SA 4.0

The engine that generates the forms is a separate work under the MIT licence. The source dictionaries are cited, never reproduced, and nothing here licenses them: they remain their authors' and publishers' work.

Citing it

If you use this data in published work, please cite the site and name the data version, because the dictionary changes as entries are reviewed.

Karachay dictionary (karachay.com), data version 4b2225c8, retrieved 18 September 2026.

The abbreviations the definitions use are explained on abbreviations and conventions, and the two spellings on spelling and scripts.