Structured output

Use createParsedCompletion when your app needs data, not prose.

When to use this

Structured output is useful for extraction and classification:

• Pull fields from a support ticket.
• Classify a message by priority.
• Convert free text into an object your app can store.

Prerequisites

Install Zod in addition to the core and provider packages:

pnpm add zod

Define a schema

import { createParsedCompletion } from "@anvia/core";
import { OpenAIClient } from "@anvia/openai";
import { z } from "zod";

const ticketSchema = z.object({
  customer: z.string(),
  priority: z.enum(["low", "medium", "high"]),
  summary: z.string(),
});

const client = new OpenAIClient({
  apiKey: process.env.OPENAI_API_KEY,
});

const model = client.completionModel("gpt-5");

const ticket = await createParsedCompletion(model, {
  schema: ticketSchema,
  input: "Acme Co. reports checkout failures. Priority is high.",
});

console.log(ticket.data);

What happens

createParsedCompletion sends the schema to the provider as an output schema, reads the model text as JSON, and parses it with Zod.

Use ticket.data for validated application data. If the response is not valid JSON or does not match the schema, the call throws.

Check yourself

Run the example and confirm ticket.data is a typed object with customer, priority, and summary.

Next

Return to the main path and wrap model behavior in an agent.

Build your first agent