DEV Community

Cover image for Unleashing the Power of GPT-3 with a Serverless Node.js REST API: A Step-by-Step Guide
Jones -
Jones -

Posted on • Updated on

Unleashing the Power of GPT-3 with a Serverless Node.js REST API: A Step-by-Step Guide

This is an example of a serverless Node.js application that uses create a REST API for the OpenAI GPT-3 language model (which similar in many ways to ChatGPT). The API allows REST API clients to send text prompts to the GPT-3 model and receive a response in return.

The code uses the codehooks-js library to handle routing and middleware for the API, and the node-fetch library to make HTTP requests to the OpenAI API. The code also uses the built-in Key-Value store to cache the response from the OpenAI API for 60 seconds, in order to reduce the number of requests made to the OpenAI API (If you want to create your own Node.js Express backend, you just need to replace this cache with Redis for example).

Additionally, the code also implements a middleware to rate limit the number of requests coming from the same IP address to 10/min.

This article is also available in the official docs.

Documentation and suggested reading

Create a new OpenAPI API key

Navigate to your Open AI account settings and create a new secret key.


Create a new project to host the GPT-3 REST API

coho create chatgptrestapi
Enter fullscreen mode Exit fullscreen mode

This will create a new directory for your REST API backend application.

cd chatgptrestapi
Enter fullscreen mode Exit fullscreen mode

Open the index.js file and replace the code with the source code at the end of this example.

Create a new secret environment variable for the serverless backend app

Once you have created your Open AI API secret key, you need to add this to the serverless runtime. You can use the CLI command for this.

coho set-env OPENAI_API_KEY 'XXXXXXXX' --encrypted
Enter fullscreen mode Exit fullscreen mode

Deploy the Node.js ChatGPT-like REST API to the backend

From the project directory run the CLI command coho deploy.
The output from my test project is shown below:

coho deploy 
Project: chatgptrestapi-pwnj  Space: dev

Deployed Codehook successfully! 🙌 
Enter fullscreen mode Exit fullscreen mode

Test your ChatGPT REST API endpoint

The following curl example demonstrates that we have a live REST API that takes a POST body with a text string ask and returns the response from the Open AI API.

curl -X POST \
  '' \
  --header 'x-apikey: cb17c77f-2821-489f-a6b4-fb0dae34251b' \
  --header 'Content-Type: application/json' \
  --data-raw '{
  "ask": "Tell a joke"                  
Enter fullscreen mode Exit fullscreen mode

Output from the GPT-3 REST API is shown below 😂:

Q: What did the fish say when it hit the wall?
A: Dam!                                 
Enter fullscreen mode Exit fullscreen mode

Full source code example

* GPT-3 REST API example using serverless node.js and

import { app, Datastore } from 'codehooks-js';
import fetch from 'node-fetch';

// REST API routes'/chat', async (req, res) => {
    if (!process.env.OPENAI_API_KEY) return res.status(500).end('Please add your OPENAI_API_KEY'); // CLI command: coho set-env OPENAI_API_KEY 'XXX' 
    const { ask } = req.body;
    const db = await;
    const cacheKey = 'chatgpt_cache_' + ask;

    // check cache first    
    const cachedAnswer = await db.get(cacheKey);

    // get from cache or OpenAi
    if (cachedAnswer) {
    } else { // get from Open AI API

        // pick text element from the OpenAI response by JS nested destructuring
        const { choices: { 0: { text } } } = await callOpenAiApi(ask);
        console.log(ask, text);

        // add to cache for 1 minute
        await db.set(cacheKey, text, { ttl: 60 * 1000 });
        // send text back to client

// Call OpenAI API
async function callOpenAiApi(ask) {
    const raw = JSON.stringify({
        "model": "text-davinci-003",
        "prompt": ask,
        "temperature": 0.6,
        "max_tokens": 1024,
        "stream": false

    var requestOptions = {
        method: 'POST',
        headers: {
            "Content-Type": "application/json",
            "Authorization": `Bearer ${process.env.OPENAI_API_KEY}`
        body: raw,
        redirect: 'follow'

    const response = await fetch("", requestOptions);
    return response.json();


// global middleware to IP rate limit traffic
app.use(async (req, res, next) => {
    const db = await;
    // get client IP address
    const ipAddress = req.headers['x-real-ip'];
    // increase count for IP
    const count = await db.incr('IP_count_' + ipAddress, 1, { ttl: 60 * 1000 })
    console.log(ipAddress, count);
    if (count > 10) {
        // too many calls
        res.status(429).end("Sorry too many requests for this IP");
    } else {
        // ok to proceed

// Export app to the serverless runtime
export default app.init();
Enter fullscreen mode Exit fullscreen mode

Full source is available on our GitHub example repo. We will update this example when ChatGPT becomes available as an API.

Top comments (1)

adii9 profile image
Aditya Mathur

Nice Post!! I recently wrote a blog on OpenAI's API and its usage with python. I have also mentioned the basic architecture on which chatGPT is working do give it a read..