Jev Agent

Jev Agent is an independently hosted implementation of the TypeSafe “Jev” decision model. It delivers typed yes/no decisions, multiple‑choice selections, rubric scores and probability values through a REST API and an MCP endpoint. The service is closed‑source and is not a general‑purpose text generator; it focuses on returning structured judgments about supplied text.

The tool is built for developers who need deterministic, typed evaluations inside larger workflows. Typical use cases shown on the site include routing support tickets, selecting the smallest language model that can answer a prompt, grading essays or resumes against a rubric, and classifying invoices or phishing emails. In practice you send a JSON payload that contains a “state” string, any piece of text you want evaluated, and a “questions” object that defines one or more typed queries. Each query specifies a type (choice, score, or noul) and any required parameters such as the list of options for a choice or the rubric definition for a score. The API returns a single response object where each question is answered with the chosen value, a confidence number, and a probability distribution over all possible answers. Because all questions are answered in the same call, ten questions take roughly the same time as one, which the site reports as about 155 ms median server‑side latency.

Getting started is straightforward. You can experiment in the public playground without an account, then create a developer account to receive an API key. Every account is granted five free credits each calendar month; additional credits are sold as one‑time packs, and there is no subscription model mentioned. Calls are made to `https://jev-agent.com/api/v1/systemone` with an `Authorization: Bearer <key>` header and a `content-type: application/json` body. The documentation provides ready‑to‑copy snippets for cURL, Python and TypeScript, and the API reference lists the exact request shapes and limits. The site notes that the service has handled 98 314 API calls across 689 developer accounts, indicating a modest but active user base.

Jev Agent is suitable if you need fast, typed judgments that can be programmatically consumed, such as automated triage, content moderation tiers, or rubric‑based scoring. It is less appropriate for tasks that require free‑form generation, conversational interaction, or deep reasoning beyond the predefined decision types.