Jev Model is a structured AI decision model for developers. Its core goal is not to generate paragraph text, but to make small decisions frequently and controllably within a system. By combining state with typed questions, it returns structured answers with probabilities for code to use directly.
Purpose
Jev Model is used to convert unstructured state into typed decisions, and is suitable for scenarios such as classification, routing, scoring, and safety checks. The official emphasis is that the hardest work in a product is often not generating text, but making small decisions multiple times per second without losing control. Jev Model handles only the intermediate decision step, while business logic remains under the developer's control.
Main Capabilities
- Typed questions: Supports three question types: Choice (option classification), Score (continuous scoring), and Noul (yes/no judgment), with multiple questions possible in a single request.
- Probabilities and confidence: Returns results with probabilities or confidence, which can be used to determine whether the system should continue automatically or hand off to a human.
- Multi-answer output: Can perform classification, scoring, and yes/no judgment from the same state at the same time, without chaining multiple calls.
- In-code decisions: Results can be embedded into if statements, routing, guardrails, and queue logic, without the need for an additional chat interface.
- Consistent request structure: Provides a clear schema and manageable API keys, making it easy to transition from testing to production.
How to Use
- Online Playground: Free to use, with unlimited runs. You can test Jev Model with real scenarios, define typed questions, and view structured decisions with probabilities.
- API access: Create an API key, and follow the SDK or REST examples to connect a validated decision workflow into your product. The request flow is: pass in state (such as ticket.text) → ask questions (such as intent, urgent) → code executes queueing, routing, blocking, or requests human review based on the results and probabilities.
- Agent Skill: Install the public Jev Model Agent Skill into Codex, Claude Code, Cursor, or another compatible coding agent, configure an API key, and choose English or Simplified Chinese for guidance and examples. After that, you can have the agent ask bounded questions, and Jev Model returns Choice, Score, or yes probabilities for code to use.
Applicable Scenarios
- Classification, routing, and safety decisions connected to production products.
- Agent integration: provides typed decisions for routing, guardrails, validation, and completion checks.
- For scenarios where it is necessary to determine whether to hand off to a human, use probability or confidence to decide whether to continue automatically or request a human.
Limitations
Jev Model 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 verify accuracy separately. Confidence is not equal to accuracy.
Plans and Usage
The official offering includes monthly, annual, or one-time access plans. One-time plans do not renew automatically, and credits remain available after purchase. Specific plans include:
- Entry level: 100,000 credits, no expiration, 1 concurrent request, support for Choice, Score, and Noul questions, typed decision output, probability and confidence results, online Playground, API key management.
- Pro: 1,000,000 credits, no expiration, unlimited workspaces, 3 concurrent requests, parallel questions per request, usage and request history, priority support.
- Higher tier: 11,000,000 credits (including 10% extra), includes all Pro features, 10 concurrent requests, dedicated fast track, team workspaces and collaboration, custom integration support, security and permissions guidance, dedicated support, priority production processing, product roadmap feedback.
Documentation and Resources
The official documentation is organized according to the actual developer path: understand the model, define questions, send states, and integrate probabilities into code. The blog provides product explanations, technical details, and practical methods, such as articles like "What Is the Jev AI Model? A Practical Guide to Typed AI Decisions."
