What languages does Jev support?

There are two senses of 'languages' here, and the answer differs. As a programming interface, Jev works from any language — it's one HTTPS call. As a natural-language model, it reads your state in many languages, but accuracy is best in English and varies elsewhere, so you treat non-English as something to test and watch the confidence on.

Programming languages: any

Jev isn't tied to an SDK or runtime. A decision is a single POST to https://jevtypesafeai.com/api/v1/decide with a JSON body, so you can call it from JavaScript/TypeScript, Python, Go, Rust, Java, Ruby, PHP, shell/curl — anything that can make an HTTPS request. The request and the typed answers map are the same everywhere; see the examples and quickstarts for your stack.

Natural languages: English best, others vary

The state you send and the instructions you write are natural language, and Jev reads more than English — including CJK (Chinese, Japanese, Korean) scripts. But English is its primary training language and where accuracy is strongest; other languages are handled with meaningfully more variance. TypeSafe's own guidance is to test on representative samples of your own content before relying on Jev for a non-English workload.

• English — primary training language, best accuracy and calibration
• Other languages incl. CJK — understood, but accuracy varies; validate on your data
• Watch answers.<q>.confidence more closely on non-English, and set a more conservative threshold
• Your criteria keys and instructions can stay in English even when the state is another language

A reliable pattern for non-English

If accuracy on a language matters, two approaches help. First, keep your questions precise and in English (the criteria keys and instructions), while the state stays in the user's language — the decision is bounded either way. Second, for high-stakes decisions, translate the relevant evidence to English first (with a generative model) and send that as state; Jev then decides on English text where it's strongest. Either way, calibrate your confidence threshold against labelled examples in that language before you trust it in production.

Localized site, not a localized model

Note the difference between the model and this site: jevtypesafeai.com is available in several interface languages, but that's the UI — it doesn't change how well the underlying model decides in a given language. The model's language behaviour is as above regardless of which site locale you browse.

FAQ

See also: Jev examples · How to use the API · Jev from TypeScript · What is Jev