Jev API: integration without confusing it with a chat API

The direct TypeSafe API

The documentation describes POST https://api.typesafe.ai/v1/systemone with an Authorization: Bearer header. The request contains the model, the state (state) and a map of questions (questions). S08 The examples in the documentation use the alias jev-latest; as of 23 September 2026, both it and jev-preview point to jev-1.13.0. For stable results, we recommend specifying a fixed version. S01

You can prepare a sample request body in the question builder. It is not a record of paid calls that were actually made.

Access via OpenRouter

OpenRouter documents a decisions endpoint, POST https://openrouter.ai/api/alpha/decisions, and the identifier typesafe/jev-1.13. So it is not enough to simply change the model name in an existing chat adapter. S17

The adapter design JEV.sk recommends

Keep the provider separate from your internal response schema. Set a timeout, validate the response, handle errors and send unclear results for review. Store the version of the task definition together with the model used. Log usage, and do not needlessly log entire private messages.

JEV.sk does not call the live API: it explains the principle and generates the sample JSON directly in the browser. Live calls would require a separate server endpoint, a budget, abuse protection and clear notice to visitors that their text is being sent to a third party.

Never put the key in the browser

For a future live demo, we recommend a server-managed key. Do not put it in JavaScript, HTML, a URL, the repository or logs. Users should know in advance whether they are working with a demo or with the real service.

During an outage, the application must not make up a successful result. It returns a clear error or offers a safe alternative procedure.