@anvia/jev

Use Jev for typed decisions in ordinary applications: classification, tagging, routing, prioritization, record matching, and verification. Create a decision model and pass it to Anvia's provider-neutral decide() operation.

Install

pnpm add @anvia/jev @anvia/core

The adapter uses the official @typesafe-ai/sdk and requires Node.js 20 or newer.

Quickstart

Set TYPESAFE_API_KEY in your server environment.

import { JevClient, JEV_LATEST } from "@anvia/jev";
import { choice, decide } from "@anvia/core/decision";

const client = new JevClient({ apiKey: process.env.TYPESAFE_API_KEY });
const model = client.decisionModel({ modelId: JEV_LATEST });

const { answers } = await decide({
  model,
  state: { message: "Please refund my duplicate payment." },
  questions: {
    department: choice({
      instructions: "Which department should handle this?",
      options: {
        billing: "Payments, invoices, and refunds",
        technical: "Product bugs and technical support",
        general: "Other requests",
      },
    }),
  },
});

console.log(answers.department.choice); // "billing" | "technical" | "general"
console.log(answers.department.confidence);

Capabilities

• Native choice, rubric score, and boolean probability questions.
• Multiple labels composed from one independent Noul question per label in the same request.
• Mixed questions, normalized token usage, and the original SDK result in rawResponse.
• Model listing, custom endpoints, headers, and an injected TypeSafeClient.
• Core-controlled retries, cancellation, and ordered batch execution.

Learn more

• Jev usage guide
• Decision API guide
• Official TypeSafe documentation