このガイドは現在、英語と中国語で提供しています。
Start here / JevModel
Your first Jev decision
Try a support-routing example, read its Choice probabilities, and find your next steps for the JevModel API.
Begin with a real state
Open the workbench on the homepage and enter a short, non-sensitive example. A useful state describes what happened without telling the model which answer to choose. Keep only facts that affect the decision.
Run a support-routing example
Paste this state into the workbench: “The customer was charged twice and needs help correcting the invoice.” Add a Choice question with ID `queue` and ask which team should handle it. Define `billing` for duplicate charges, refunds, or invoices, and `technical` for app, API, or account failures. Explain each option so the categories do not overlap.
Read the result and test the boundary
After a successful run, inspect the selected answer and probabilities for `queue`. For example, `billing: 0.92` and `technical: 0.08` means the model prefers billing under these option definitions; it is not a 92% guarantee of correctness. Try a clear technical case and a mixed case, revise unclear boundaries, set your own action threshold, and route uncertain or high-impact cases to a person.
Fix common request errors
If validation fails, check that the state is not empty, the question ID uses lowercase letters, digits, or underscores, instructions are at least 8 characters, and Choice has 2–12 distinct options. Keep the serialized request under 8,000 characters. To call Jev from code, follow the API guide; verify your email before running, and check History and quota before retrying a request that may already have succeeded.
JevModel is independent and not affiliated with TypeSafe AI.