Langchain Structured Output

langchain-structured-output (JavaScript/TypeScript)

Overview

Structured output transforms unstructured model responses into validated, typed data. Instead of parsing free text, you get JSON objects conforming to your schema - perfect for extracting data, building forms, or integrating with downstream systems.

Key Concepts:

• responseFormat: Define expected output schema
• Zod Validation: Type-safe schemas with automatic validation
• withStructuredOutput(): Model method for direct structured output
• Tool Strategy: Uses tool calling under the hood for models without native support

Decision Tables

When to Use Structured Output

Schema Options

Code Examples

Basic Structured Output with Agent

import { createAgent } from "langchain";
import { z } from "zod";

const ContactInfo = z.object({
  name: z.string(),
  email: z.string().email(),
  phone: z.string(),
});

const agent = createAgent({
  model: "gpt-4.1",
  responseFormat: ContactInfo,
});

const result = await agent.invoke({
  messages: [{
    role: "user",
    content: "Extract: John Doe, john@example.com, (555) 123-4567"
  }],
});

console.log(result.structuredResponse);
// { name: 'John Doe', email: 'john@example.com', phone: '(555) 123-4567' }

Model Direct Structured Output

import { ChatOpenAI } from "@langchain/openai";
import { z } from "zod";

const MovieSchema = z.object({
  title: z.string().describe("Movie title"),
  year: z.number().describe("Release year"),
  director: z.string(),
  rating: z.number().min(0).max(10),
});

const model = new ChatOpenAI({ model: "gpt-4.1" });
const structuredModel = model.withStructuredOutput(MovieSchema);

const response = await structuredModel.invoke("Tell me about Inception");
console.log(response);
// { title: "Inception", year: 2010, director: "Christopher Nolan", rating: 8.8 }

Complex Nested Schema

import { z } from "zod";

const AddressSchema = z.object({
  street: z.string(),
  city: z.string(),
  state: z.string(),
  zip: z.string(),
});

const PersonSchema = z.object({
  name: z.string(),
  age: z.number().int().positive(),
  email: z.string().email(),
  address: AddressSchema,
  tags: z.array(z.string()),
});

const agent = createAgent({
  model: "gpt-4.1",
  responseFormat: PersonSchema,
});

Enum and Literal Types

import { z } from "zod";

const ClassificationSchema = z.object({
  category: z.enum(["urgent", "normal", "low"]),
  sentiment: z.enum(["positive", "neutral", "negative"]),
  confidence: z.number().min(0).max(1),
});

const agent = createAgent({
  model: "gpt-4.1",
  responseFormat: ClassificationSchema,
});

const result = await agent.invoke({
  messages: [{
    role: "user",
    content: "Classify: This is extremely important and I'm very happy!"
  }],
});
// { category: "urgent", sentiment: "positive", confidence: 0.95 }

Optional Fields and Defaults

import { z } from "zod";

const EventSchema = z.object({
  title: z.string(),
  date: z.string(),
  location: z.string().optional(),
  attendees: z.array(z.string()).default([]),
  confirmed: z.boolean().default(false),
});

Union Types (Multiple Schemas)

import { z } from "zod";

const EmailSchema = z.object({
  type: z.literal("email"),
  to: z.string().email(),
  subject: z.string(),
});

const PhoneSchema = z.object({
  type: z.literal("phone"),
  number: z.string(),
  message: z.string(),
});

const ContactSchema = z.union([EmailSchema, PhoneSchema]);

const agent = createAgent({
  model: "gpt-4.1",
  responseFormat: ContactSchema,
});
// Model chooses which schema based on input

Array Extraction

import { z } from "zod";

const TaskListSchema = z.object({
  tasks: z.array(z.object({
    title: z.string(),
    priority: z.enum(["high", "medium", "low"]),
    dueDate: z.string().optional(),
  })),
});

const agent = createAgent({
  model: "gpt-4.1",
  responseFormat: TaskListSchema,
});

const result = await agent.invoke({
  messages: [{
    role: "user",
    content: "Extract tasks: 1. Fix bug (high priority, due tomorrow) 2. Update docs"
  }],
});

Include Raw AIMessage

import { ChatOpenAI } from "@langchain/openai";
import { z } from "zod";

const schema = z.object({ name: z.string(), age: z.number() });

const model = new ChatOpenAI({ model: "gpt-4.1" });
const structuredModel = model.withStructuredOutput(schema, {
  includeRaw: true,
});

const response = await structuredModel.invoke("Person: Alice, 30 years old");
console.log(response);
// {
//   raw: AIMessage { ... },
//   parsed: { name: "Alice", age: 30 }
// }

Error Handling

import { createAgent } from "langchain";
import { z } from "zod";

const StrictSchema = z.object({
  email: z.string().email(),
  age: z.number().int().min(0).max(120),
});

const agent = createAgent({
  model: "gpt-4.1",
  responseFormat: StrictSchema,
});

try {
  const result = await agent.invoke({
    messages: [{ role: "user", content: "Email: invalid, Age: -5" }],
  });
} catch (error) {
  console.error("Validation failed:", error);
  // Model will retry or return error
}

Boundaries

What You CAN Configure

✅ Schema structure: Any valid Zod schema ✅ Field validation: Types, ranges, regex, etc. ✅ Optional vs required: Control field presence ✅ Nested objects: Complex hierarchies ✅ Arrays: Lists of items ✅ Enums: Restricted values

What You CANNOT Configure

❌ Model reasoning: Can't control how model generates data ❌ Guarantee 100% accuracy: Model may still make mistakes ❌ Force valid data if context lacks it: Model can't invent missing info

Gotchas

1. Accessing Response Wrong

// ❌ Problem: Accessing wrong property
const result = await agent.invoke(input);
console.log(result.response);  // undefined!

// ✅ Solution: Use structuredResponse
console.log(result.structuredResponse);

2. Missing Descriptions

// ❌ Problem: No field descriptions
const schema = z.object({
  date: z.string(),  // What format?
  amount: z.number(),  // What unit?
});

// ✅ Solution: Add descriptions
const schema = z.object({
  date: z.string().describe("Date in YYYY-MM-DD format"),
  amount: z.number().describe("Amount in USD"),
});

3. Over-constraining

// ❌ Problem: Too strict for model
const schema = z.object({
  code: z.string().regex(/^[A-Z]{2}-\d{4}-[A-Z]{3}$/),  // Very specific!
});

// ✅ Solution: Validate post-processing or use looser schema
const schema = z.object({
  code: z.string().describe("Format: XX-0000-XXX (letters and numbers)"),
});

4. Not Handling Validation Errors

// ❌ Problem: No error handling
const result = await agent.invoke(input);
const data = result.structuredResponse;  // May throw!

// ✅ Solution: Try/catch or check for errors
try {
  const result = await agent.invoke(input);
  const data = result.structuredResponse;
} catch (error) {
  console.error("Failed to get structured output:", error);
}

5. Confusing responseFormat with tools

// ❌ Problem: Using responseFormat with tools incorrectly
const agent = createAgent({
  model: "gpt-4.1",
  tools: [searchTool],
  responseFormat: MySchema,  // Will extract from FINAL response only
});
// Tools run first, then schema extracted from final response

// ✅ This is correct if you want tools + structured final output
// Just understand the flow

Links to Documentation

• Structured Output Overview
• Model Structured Output
• Agent Structured Output