Define a typed schema for what your application needs to decide. Connect your state. Call the Jev API and get back a strongly-typed result — classification, route, score, or safety flag — with a probability score on every output.
{
"decision": "route_to_support",
"confidence": 0.94,
"alternatives": [
{ "value": "self_serve", "probability": 0.05 },
{ "value": "escalate", "probability": 0.01 }
],
"schema_version": "v3"
}Three steps from application state to a typed decision your code can branch on.
Describe the decision as a typed structure: the input shape, the possible output values, and what confidence means in your domain.
Pass your application state — a user action, a document, a conversation turn — to the Jev REST API or TypeScript SDK.
Receive a strongly-typed decision with probability scores. Test and iterate in the playground before shipping to production.
Any place in your codebase where a conditional depends on understanding, not just data — that's a Jev decision.
Intent detection, content categorization, spam filtering, sentiment tagging — with probability distributions across all classes.
Which agent handles this? Which workflow branch? Which support tier? Typed routing instead of fragile prompt switches.
Risk scores, quality assessments, relevance rankings. A typed numeric output your business logic can threshold against.
Content policy enforcement, compliance validation, PII detection. Boolean or categorical outputs with confidence you can audit.
Replace ad-hoc prompt logic inside agent loops with versioned, testable decision nodes. Each branch gets a probability score so your orchestrator can reason about uncertainty — not just binary true/false.
Jev is in private beta. Leave your email and we'll reach out when your spot is ready — playground access, API keys, and TypeScript SDK included.