Blog / Questions / FIG. 55
Does Jev Have an API? Yes: What the Ecosystem Documents
Jev's API as the ecosystem documents it: typesafe-ai/jev via the Vercel AI Gateway, an evaluate surface, choice probabilities, and SDKs in four languages.
Yes, and it's the only way anyone uses Jev: there's no chat interface to fall back on, because there's no chat. Everything below is how the ecosystem's clients and directories consistently describe the surface, offered as orientation; the syntax, auth, and limits that actually bind you live at docs.typesafe.ai, and this site deliberately never reproduces API details that rot within a week of any update.
The consistent picture: the model is addressed as typesafe-ai/jev, called through the Vercel AI Gateway (an AI_GATEWAY_API_KEY in the environment), with an evaluation surface community clients name experimental_evaluate. The native shape is the Choice API: you supply a question and its candidate answers, and Jev returns a probability for each choice, defaulting to YES / NO / UNCLEAR when you don't customize the set. Those probabilities are the confidence signal every cascade on this site gates on. One documented sizing constraint: choice sets cap around 255 options, with community tooling (jev-tree) existing to recurse past it.
SDK-wise, the ecosystem reports an official Python SDK (around 0.6.0 at the time of writing) with community clients in TypeScript (jev-harness), Swift (swift-typesafe), and Elixir (req_llm), plus an async Python alternative (jevclient). Which to use: whichever matches your stack, with the official Python client as the least-surprises default; the getting-started guide covers the mental model that matters more than any binding.
Treat every specific in this page as the ecosystem's report, dated September 2026, and the official docs as the ruling.
Frequently asked questions
How do I get a Jev API key?
Through TypeSafe's platform per the official docs; ecosystem reports describe gateway-based keys. Access mechanics change at launch speed, so follow docs.typesafe.ai, not screenshots.
Does the API return plain answers or probabilities?
Per ecosystem clients: a probability per choice, which is strictly more useful; threshold it for a verdict, route the uncertain slice per the routing pattern.
Is there an official REST spec I can read here?
Deliberately not: third-party API snapshots go stale and then mislead. We document patterns and economics; TypeSafe documents the wire format.
Numbers throughout are as reported by the build authors, not verified by shipwithjev. Code-shaped examples are pseudocode; the official docs live at docs.typesafe.ai.