OpenAI Zod Validator

A powerful TypeScript library for validating OpenAI SDK 6 structured outputs and function calls using Zod 4 schemas. Ensures type-safe AI responses with full TypeScript inference.

Features

✅ Full TypeScript support with type inference
✅ Compatible with OpenAI SDK 6.x
✅ Built for Zod 4.x schemas
✅ Validates structured outputs, function calls, and tool calls
✅ Convert Zod schemas to OpenAI function parameters
✅ Comprehensive error handling
✅ Zero configuration required

Installation

npm install openai-zod-validator openai zod

Quick Start

import OpenAI from 'openai';
import { z } from 'zod';
import { createOpenAIValidator } from 'openai-zod-validator';

// Define your Zod schema
const ResponseSchema = z.object({
  name: z.string(),
  age: z.number(),
  email: z.string().email(),
});

// Create a validator
const validator = createOpenAIValidator(ResponseSchema);

// Use with OpenAI
const openai = new OpenAI({ apiKey: process.env.OPENAI_API_KEY });

const completion = await openai.chat.completions.create({
  model: 'gpt-4',
  messages: [{ role: 'user', content: 'Generate user data' }],
  response_format: { type: 'json_object' },
});

// Validate the response
const result = validator.validateCompletion(completion);
console.log(result.data); // Fully typed as { name: string; age: number; email: string }

Usage with OpenAI SDK 6

1. Structured Outputs (JSON Mode)

import OpenAI from 'openai';
import { z } from 'zod';
import { createOpenAIValidator } from 'openai-zod-validator';

const openai = new OpenAI();

// Define schema
const RecipeSchema = z.object({
  name: z.string(),
  ingredients: z.array(z.string()),
  steps: z.array(z.string()),
  cookingTime: z.number(),
});

const validator = createOpenAIValidator(RecipeSchema);

async function generateRecipe() {
  const completion = await openai.chat.completions.create({
    model: 'gpt-4-turbo-preview',
    messages: [
      {
        role: 'user',
        content: 'Generate a recipe for chocolate chip cookies',
      },
    ],
    response_format: { type: 'json_object' },
  });

  const result = validator.validateCompletion(completion);

  if (result.success) {
    console.log('Recipe:', result.data); // Fully typed!
  } else {
    console.error('Validation errors:', result.errors);
  }
}

2. Function Calling

import OpenAI from 'openai';
import { z } from 'zod';
import { createOpenAIValidator, zodToOpenAIParameters } from 'openai-zod-validator';

const openai = new OpenAI();

// Define function parameter schema
const GetWeatherSchema = z.object({
  location: z.string(),
  unit: z.enum(['celsius', 'fahrenheit']),
});

const validator = createOpenAIValidator(GetWeatherSchema);

async function callWithFunctions() {
  const completion = await openai.chat.completions.create({
    model: 'gpt-4',
    messages: [
      {
        role: 'user',
        content: 'What is the weather in San Francisco?',
      },
    ],
    functions: [
      {
        name: 'get_weather',
        description: 'Get the current weather in a location',
        parameters: zodToOpenAIParameters(GetWeatherSchema),
      },
    ],
    function_call: 'auto',
  });

  const message = completion.choices[0]?.message;

  if (message?.function_call) {
    const result = validator.validateFunctionCall(message.function_call);

    if (result.success) {
      console.log('Function arguments:', result.data);
      // result.data is typed as { location: string; unit: 'celsius' | 'fahrenheit' }
    }
  }
}

3. Tool Calls (OpenAI SDK 6 Tools API)

import OpenAI from 'openai';
import { z } from 'zod';
import { createOpenAIValidator, zodToOpenAIParameters } from 'openai-zod-validator';

const openai = new OpenAI();

const SearchSchema = z.object({
  query: z.string(),
  limit: z.number().optional(),
});

const validator = createOpenAIValidator(SearchSchema);

async function callWithTools() {
  const completion = await openai.chat.completions.create({
    model: 'gpt-4-turbo-preview',
    messages: [
      {
        role: 'user',
        content: 'Search for TypeScript tutorials',
      },
    ],
    tools: [
      {
        type: 'function',
        function: {
          name: 'search',
          description: 'Search for content',
          parameters: zodToOpenAIParameters(SearchSchema),
        },
      },
    ],
  });

  const message = completion.choices[0]?.message;

  if (message?.tool_calls) {
    for (const toolCall of message.tool_calls) {
      const result = validator.validateToolCall(toolCall);

      if (result.success) {
        console.log('Tool arguments:', result.data);
      }
    }
  }
}

4. Direct Data Validation

import { z } from 'zod';
import { createOpenAIValidator } from 'openai-zod-validator';

const UserSchema = z.object({
  id: z.string(),
  name: z.string(),
  email: z.string().email(),
  age: z.number().min(0),
});

const validator = createOpenAIValidator(UserSchema);

// Validate any data
const data = {
  id: '123',
  name: 'John Doe',
  email: 'john@example.com',
  age: 30,
};

const result = validator.validate(data);

if (result.success) {
  console.log('Valid user:', result.data);
} else {
  console.error('Validation errors:', result.errors);
}

API Reference

`createOpenAIValidator(schema, options?)`

Creates a validator instance for a Zod schema.

Parameters:

schema: A Zod schema
options?: Optional configuration
- throwOnError?: boolean - Whether to throw on validation errors (default: true)
- stripUnknown?: boolean - Whether to strip unknown keys (default: false)

Returns: Validator instance with methods:

validate(data) - Validate any data
validateCompletion(completion) - Validate OpenAI chat completion
validateFunctionCall(functionCall) - Validate function call arguments
validateToolCall(toolCall) - Validate tool call arguments
getSchema() - Get the underlying Zod schema

`zodToOpenAIParameters(schema)`

Converts a Zod schema to OpenAI function parameters format.

Parameters:

schema: A Zod schema

Returns: OpenAI-compatible parameter object

Advanced Examples

Custom Error Handling

const validator = createOpenAIValidator(MySchema, {
  throwOnError: false
});

const result = validator.validate(data);

if (!result.success) {
  result.errors?.errors.forEach(error => {
    console.log(`Error at ${error.path.join('.')}: ${error.message}`);
  });
}

Complex Nested Schemas

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

const PersonSchema = z.object({
  name: z.string(),
  age: z.number(),
  addresses: z.array(AddressSchema),
  metadata: z.record(z.string()),
});

const validator = createOpenAIValidator(PersonSchema);

// Full type inference works with nested objects
const result = validator.validate(data);
// result.data is fully typed with all nested properties

Using with Streaming

const validator = createOpenAIValidator(ResponseSchema);

const stream = await openai.chat.completions.create({
  model: 'gpt-4',
  messages: [{ role: 'user', content: 'Generate data' }],
  stream: true,
  response_format: { type: 'json_object' },
});

let accumulatedContent = '';

for await (const chunk of stream) {
  const content = chunk.choices[0]?.delta?.content || '';
  accumulatedContent += content;
}

// Validate accumulated content
const result = validator.validate(JSON.parse(accumulatedContent));

TypeScript Support

This library is built with TypeScript and provides full type inference:

const Schema = z.object({
  name: z.string(),
  count: z.number(),
});

const validator = createOpenAIValidator(Schema);
const result = validator.validate(data);

if (result.success) {
  // result.data is automatically typed as:
  // { name: string; count: number }
  console.log(result.data.name.toUpperCase());
  console.log(result.data.count * 2);
}

Error Handling

The library provides detailed error information:

const validator = createOpenAIValidator(Schema, { throwOnError: false });
const result = validator.validate(invalidData);

if (!result.success) {
  result.errors?.errors.forEach(error => {
    console.log('Path:', error.path);
    console.log('Message:', error.message);
    console.log('Code:', error.code);
  });
}

Requirements

Node.js >= 18.0.0
OpenAI SDK >= 6.0.0
Zod >= 4.0.0

License

MIT

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Support

For issues and questions, please open an issue on GitHub.

default

This tool is currently broken

Interactive Playground

Installation & Usage

1. Install the package

2. Import the tool

3. Use with AI SDK

Parameters

README