Jev tutorial FAQ (TypeSafe Jev / System One)
What is Jev / System One?
Jev (typesafe-ai/jev) is TypeSafe’s System One evaluation model on Vercel AI Gateway. It returns typed Choice, Score, and Boolean answers with probabilities instead of free-form prose.
Do I need to build my own API relay?
No. Call AI Gateway directly from your app or serverless function. A self-built relay is out of scope and discouraged.
Can I use the OpenAI-compatible Gateway endpoint?
Evaluation models require the AI SDK evaluate API (AI SDK 7+). They are not supported through OpenAI-compatible endpoints.
What does the template pack include?
S1 starters: T1 ticket routing (tech/sales/billing/human), T2 content moderation (allow/block/human), T3 livestream labeling (buy_intent/price_question/after_sales/spam/other), and optional T4 agent risk preview. Sold once on Gumroad for $49. Placeholder listing.
Does any template process payments or refunds?
No. Templates only suggest routes or labels. Money decisions stay human or other systems.
Is this CMD Guard or a browser DOM extension?
No. Those products are explicitly out of scope. T4 agent risk is a preview pattern only.
How should I set confidence thresholds?
High confidence → auto-suggest; low → human queue. Calibrate on labeled data from your workflow. Score values are expected rubric levels, not accuracy probabilities. Do not promise accuracy.