Streamed ChatGPT API With Node

A Node.js module for streaming ChatGPT API responses using the OpenAI API. Streamed ChatGPT API allows users to fetch AI-generated responses in real-time. This module was created due to issues with other modules that were causing freezing, which provided a poor user experience. Various timeouts have been implemented to automatically retry and throw errors in case of disconnections.

Introduction

ChatGPT is an advanced AI language model developed by OpenAI. This module enables you to interact with the ChatGPT API, allowing you to send messages and receive AI-generated responses in real-time. The OpenAI API provides access to various models, including the gpt-3.5-turbo model, which is used by default in this module.

Usage Example

A simple node web app showing usage of the module with streamed chat can be found here: Streamed ChatGPT API Usage Example

Installation

Install using npm:

npm install streamed-chatgpt-api

Usage

To use the module, first import it:

const { fetchStreamedChat, fetchStreamedChatContent } = require('streamed-chatgpt-api');

Then call the fetchStreamedChat function with your options and a callback function to process the streamed response. This is the simplest example, you can pass in an OpenAI API key and a single string as a prompt:

const apiKey = 'your_api_key_here';

fetchStreamedChat({
    apiKey,
    messageInput: 'Hello, how are you?',
}, (responseChunk) => {
    // get the actual content from the JSON
    const content = JSON.parse(responseChunk).choices[0].delta.content;
    if (content) {
        process.stdout.write(content);
    }
});

You can also pass in an an array as shown in the OpenAI API documentation, where you can define the System prompt, and on going conversation. Here's a simple example:

const apiKey = 'your_api_key_here';

const messages = [{ role: 'system', content: '' }, { role: 'user', content: 'capital of canada' }];

fetchStreamedChat({
    apiKey,
    messageInput: messages,
    fetchTimeout: 10000,
    }, (responseChunk) => {
    // get the actual content from the JSON
    const content = JSON.parse(responseChunk).choices[0].delta.content;
    if (content) {
        process.stdout.write(content);
    }
});

Using fetchStreamedChatContent

The fetchStreamedChatContent function is a higher-level function that simplifies the process of fetching the generated content by not requiring you to deal with individual chunks. It takes the same options as fetchStreamedChat but also accepts three optional callback functions, onResponse, onFinish, and onError.

const apiKey = 'your_api_key_here';

fetchStreamedChatContent({
    apiKey,
    messageInput: 'Hello, how are you?',
}, (content) => {
    // onResponse
    process.stdout.write(content);
}, () => {
    // onFinish
    console.log('Chat completed');
}, (error) => {
    // onError
    console.error('Error:', error);
});

You can specify the same options as specified in OpenAI's chat completion reference.

Here is a table of the parameters below. Only the API key and messageInput (prompt) are required. By default the gpt-3.5-turbo is used.

Options

The following options are available for the fetchStreamedChat and fetchStreamedChatContent functions:

Option	Type	Default	Description
apiKey	string	-	Your OpenAI API key.
messageInput	string or array of objects	-	The input message or messages to generate a chat response for.
apiUrl	string	"https://api.openai.com/v1/chat/completions"	The OpenAI API URL to use.
model	string	"gpt-3.5-turbo"	The OpenAI model to use.
temperature	number	-	The softmax temperature to use.
topP	number	-	The top_p value to use.
n	number	-	The number of responses to generate.
stop	string or array of strings	-	The sequence or sequences to stop generation at.
maxTokens	number	-	The maximum number of tokens to generate.
presencePenalty	number	-	The presence penalty value to use.
frequencyPenalty	number	-	The frequency penalty value to use.
logitBias	object	-	The logit bias object to use.
user	string	-	The user ID to use for the chat session.
retryCount	number	3	The number of times to retry if the chat response fetch fails.
fetchTimeout	number	20000	The timeout value for the fetch request.
readTimeout	number	10000	The timeout value for reading the response stream.
retryInterval	number	2000	The interval between retries.
totalTime	number	300000	The total time to allow for the chat response fetch.

License

MIT

Author

Created by Johann Dowa

jddev273/streamed-chatgpt-api