Jev AI is a decision-making tool for software teams, with the core goal of turning real-world states into structured, probabilistic decision results that code can use directly. Developers can first enter real scenarios in the online playground, define typed questions, and view standardized decision results with probabilities. Jev runs in the playground are free, while Laya runs consume account credits.
In terms of main capabilities, Jev handles classification, routing, scoring, and safety checks. It supports three types of questions: choice, score, and noul (yes/no judgment). Users can ask typed questions about shared state and ask multiple questions in a single request to get answers with probabilities. Jev emphasizes a "bounded answer space": define possible answers in advance, then pass the results into functions, workflows, and databases. It supports classification, scoring, and yes/no judgment at the same time, without needing to chain multiple independent calls. The probabilities or confidence levels can be used to determine whether the system should continue automatically or request human intervention.
In terms of usage, Jev provides a consistent request structure, explicit schema, and manageable API keys, making it easy to move from testing to production. It provides SDK or REST examples to embed decisions into existing services. For agent integration, Jev provides an Agent Skill that can be added to Codex, Claude Code, Cursor, or other compatible agents; configure a Jev AI API key, and choose English or Simplified Chinese for guidance and examples. Then ask the agent bounded questions, and Jev returns code-usable choice, score, or yes probabilities. It should be noted that Jev is only responsible for judgment; the agent, application permissions, deterministic rules, and human approval still control the final action.
Applicable scenarios include: assessing task difficulty before calling a model, sending simple requests to a faster model, and escalating when necessary; checking intent before tool execution, and requesting confirmation for sensitive operations such as deletion and payment; classifying intent, urgency, and the need for human review together so that tickets follow the correct path; deciding in long sessions which tool results are still important, and preserving source details during context compression.
Several decision models are listed in the information, including Jev's structured decision model, Laya, Cloudflare Clef, Cloudflare Clef Flash, PPLX Decider V1 27B, Tev1 4B Experimental, and others. Some of them are open-weight, supSupports multiple languages, some of which are decision models from other vendors. Jev AI explicitly mentions itself and 13 other Decision APIs, 14 in total.
Regarding limitations, Jev currently accepts text, JSON objects, and text arrays; images, audio, and video are not yet supported. For non-English input, the official recommendation is to separately verify accuracy. Confidence is not equivalent to accuracy. Jev does not require rewriting the system; instead, it separates state, questions, and next actions, with the model handling only the middle decision-making part.
Regarding pricing and subscriptions, the materials indicate that monthly, annual, or one-time payment options are available. The one-time plan does not auto-renew and does not require a long-term subscription; the $10, $100, and $1,000 tiers include credits that remain usable after purchase. For complete details on specific prices, credit amounts, and concurrency limits, refer to the official page.
