The OpenAI Responses API is designed for building interactive experiences. Metorial provides support for this API through a dedicated adapter.

What is the Responses API?

The Responses API is optimized for multi-turn conversations and real-time interactions. It's particularly useful for building conversational interfaces where the AI needs to interact with tools during the conversation.

Basic Setup

import { Metorial } from 'metorial';
import { metorialOpenAI } from '@metorial/openai';
import OpenAI from 'openai';

const metorial = new Metorial({
  apiKey: process.env.METORIAL_API_KEY
});

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

await metorial.withProviderSession(
  metorialOpenAI.responses, // Note: using .responses instead of .chatCompletions
  { serverDeployments: ['your-server-deployment-id'] },
  async session => {
    // Your code here
  }
);

Key Differences from Chat Completions

Tool Format

The Responses API uses a slightly different tool format with strict mode enabled by default:

// Tools are automatically formatted with strict: true
const tools = session.tools; // Already in the correct format

Calling Tools

The callTools method returns FunctionCallOutput objects:

const toolOutputs = await session.callTools(toolCalls);
// Returns an array of FunctionCallOutput objects

When to Use the Responses API

• Building conversational interfaces with multiple turns
• Creating interactive experiences with real-time tool usage
• When you need more control over the conversation flow
• For applications that benefit from structured tool calling

When to Use Chat Completions API

• Simple request-response patterns
• Batch processing
• When you don't need the additional features of the Responses API

Both APIs work seamlessly with Metorial's 600+ integrations. Choose based on your application's requirements.