Now that GPT-4o has been released, let’s look into how we can implement our custom GPT-4o assistant.

To start, go to an empty directory and create a npm project and install the dependencies:

• npm init -y
• npm install openai

Because I’m running the code directly, I’m adding the type: "module" to my package.json:

{
  "name": "tutorial-gpt-4o",
  "type": "module",
  "version": "1.0.0",
  ...
}

Now that we have all of our preparation, it is time to create our script.

// We import the OpenAI library
import { OpenAI } from "openai";

// We set our key in a variable
// Pssss.. you should use an environment variable
const key = "YOUR-API-KEY";

async function main() {
	// The OpenAI constructor accepts an `apiKey` string parameter
    const openai = new OpenAI({ apiKey: key });

    console.log("Writing message");

	// We need to call a chained function
    const chat = await openai.chat.completions.create({
		// This is where we assign GPT-4o
        model: "gpt-4o",
        messages: [{
			// Every message has a `role` and `content`
            role: "user", content: "What color is the sky?"
        }]
    });

	// Finally we log our response
    console.log("Response:", chat.choices[0].message.content);
}

// Don't forget to call your function!
main();

If you run node index.js, you’ll have an output similar to:

Writing message Response: Hello! I'm just a computer program, so I don't have feelings, but thanks for asking. How can I assist you today?

Deep diving into the types

Now let’s analyze the code in detail, as some parts may be a little confusing:

messages: [{
	role: "user", content: "What color is the sky?"
}]

Every message object in the array has two values, the first one is role, which represents who is sending the message. We are using user, which indicates the model that it has to reply. We can also use system, which will be the core instructions for the model, but it will still be waiting for a second message coming from user.

A working example would be the following:

messages: [
	{ role: "system", content: "You are a meteorologist"}
	{ role: "user", content: "Why does it rain?"}
]

By setting the system instructions, the model now will know that, to every query it receives, it needs to remember that it is a meterologist.

The other confusing element could be the value that we receive:

chat.choices[0].message.content

Every time we call the completion endpoint it will return an array of choices, by default, we will always receive a single choice, so, unless that you manually change this (because you want to compare results), you should always obtain the element 0 in the array.

message is a wrapper with 3 values, content, which is the answer, role, which we just saw before. In this case, it would be assistant, and finally tool_calls, but we won’t see that here (you can see it in the documentation).

And that’s it! You can now start experimenting with your model.