OpenAI Node.js SDK

Instrument and observe OpenAI calls

This module provides automatic instrumentation for the OpenAI Node.js SDK. which may be used in conjunction with @opentelemetry/sdk-trace-node.

Install

npm install --save @arizeai/openinference-instrumentation-openai \
  @arizeai/phoenix-otel \
  openai

Setup

To instrument your application, import and enable OpenAIInstrumentation.

Create the instrumentation.ts (or .js) file:

ESM Project

// instrumentation.ts
import { register, registerInstrumentations } from "@arizeai/phoenix-otel";
import OpenAI from "openai";
import { OpenAIInstrumentation } from "@arizeai/openinference-instrumentation-openai";

// Register Phoenix OTEL with automatic configuration
const provider = register({
  projectName: "openai-app",
});

// Manually instrument OpenAI for ESM projects
const instrumentation = new OpenAIInstrumentation();
instrumentation.manuallyInstrument(OpenAI);

registerInstrumentations({
  instrumentations: [instrumentation],
});

console.log("✅ OpenAI instrumentation registered");

ESM Projects: You must manually instrument OpenAI by calling instrumentation.manuallyInstrument(OpenAI) before using the OpenAI client.

CommonJS Project

Configuration

The register function automatically reads from environment variables:

• PHOENIX_COLLECTOR_ENDPOINT - Your Phoenix instance URL (defaults to http://localhost:6006)

• PHOENIX_API_KEY - Your Phoenix API key for authentication

You can also configure these directly:

const provider = register({
  projectName: "openai-app",
  url: "https://app.phoenix.arize.com",
  apiKey: process.env.PHOENIX_API_KEY,
});

Run OpenAI

Import the instrumentation.ts file first, then use OpenAI as usual.

// main.ts
import "./instrumentation.ts"; 
import OpenAI from "openai";

// Set OPENAI_API_KEY in environment, or pass it in arguments
const openai = new OpenAI({
  apiKey: process.env.OPENAI_API_KEY,
});

openai.chat.completions
  .create({
    model: "gpt-4o",
    messages: [{ role: "user", content: "Write a haiku." }],
  })
  .then((response) => {
    console.log(response.choices[0].message.content);
  })
  // Keep process alive for BatchSpanProcessor to flush traces
  .then(() => new Promise((resolve) => setTimeout(resolve, 6000)));

Run your application:

# Set your API keys
export OPENAI_API_KEY='your-openai-api-key'
export PHOENIX_COLLECTOR_ENDPOINT='http://localhost:6006'  # or your Phoenix URL
export PHOENIX_API_KEY='your-phoenix-api-key' 

# Run the application
node main.ts
# Or using --require flag
node --require ./instrumentation.ts main.ts

Observe

After setting up instrumentation and running your OpenAI application, traces will appear in the Phoenix UI for visualization and analysis.

Resources

• Example project

• OpenInference package for OpenAI Node.js SDK