Jev AI provides a playground and API for making structured decisions from text. Users submit plain text, JSON, or arrays and define yes/no questions, multiple-choice options, or custom score levels. Results include probabilities, selected labels, numeric scores, and calibrated confidence, allowing applications to branch on outputs. Multiple questions can run in parallel through one request. The playground includes templates for support triage, moderation, intent routing, review scoring, sales qualification, guardrails, and citation checks. Native Jev accepts text; Jev-Omni can handle one question about an image, audio, or short video in the playground. API requests use a TypeSafe-compatible JSON shape. PDF files are not directly supported, so their content must be converted to text.