Scripts for using ChatGPT without login

Example for accessing ChatGPT (3.5) without an account or API access via Python. Only for educational purposes.

---

Heads-up

OpenAI just announced instant access to ChatGPT 3.5 without signing up; this repo contains scripts in multiple programming languages for programmatic instant access, i.e., headless chats with ChatGPT via https://chat.openai.com (without signing up or requiring API keys).

Instant access is currently rolling out and (based on some quick tests of mine) is currently only available in the US and Canada (IP-based).

Supported Languages (more to come)

• Python
• JavaScript (Node.js)

Python

Setup / Dependencies

Grab the gpt.py script from within the python directory.

The script relies on pyppeteer, a headless chromium library.

Install it using pip install pyppeteer (or pip3, depending on your setup).

Usage

Calling it from your own script

With my gpt.py in the same directory, you can call it using some prompt using this example (test_implementation.py):

import asyncio
from gpt import GPT

async def run_session():
    # create gpt instance & send initial prompt
    session = GPT(prompt="Tell me a joke.", streaming=True)
    
    await session.start() 
    
    # (optional) send additional prompts and handle them
    print("\n -- asking GPT to explain the joke -- \n")
    await session.handle_prompt("Explain the joke.")

    # gracefully close the session again
    await session.close()

asyncio.run(run_session())

The first prompt of a new session always takes a bit longer, as the headless browser has to init. Subsequent prompts are handled as fast as you'd expect it from the typical GPT browser UI.

CLI Arguments

You can also chat with GPT using CLI; just call python3 gpt.py with the following args:

• -p, --prompt

  • the initial prompt text to send to ChatGPT.
  • type: str
  • default: "Hello, GPT"

• -ns, --no-streaming

  • if true, the script doesn't print GPT's response in chunks ("streaming"-esque), but rather waits until it is fully typed out before printing it.
  • type: bool (store_true)
  • default: false

You can start a streaming-based conversation with python3 gpt.py.

Node.js

Setup / Dependencies

Grab the gpt.js script from within the nodejs directory.

The Node.js script relies on puppeteer, a headless chromium library, and commander.

Install both using npm install puppeteer commander.

Usage

Calling it from your own script

With gpt.js in the same directory, you can call it using some prompt using this example (test_implementation.js):

const { GPT } = require('./gpt.js');

(async () => {
    try {
        // create gpt instance & send initial prompt
        const gptSession = new GPT("Tell me a joke.", true);

        await gptSession.start();
        
        // (optional) send additional prompts and handle them
        console.log("\n -- asking GPT to explain the joke -- \n")
        await gptSession.handlePrompt("Explain the joke.");

        // gracefully close the session again
        await gptSession.close();
    } catch (error) {
        console.error("Error in GPT session:", error);
    }
})();

The first prompt of a new session always takes a bit longer, as the headless browser has to init. Subsequent prompts are handled as fast as you know it from the typical GPT browser UI.

CLI Arguments

You can also chat with GPT using Node CLI; call node gpt.js with the following args:

• -p, --prompt

  • the initial prompt text to send to ChatGPT.
  • type: str
  • default: "Hello, GPT"

• -ns, --no-streaming

  • if true, the script doesn't print GPT's response in chunks ("streaming"-esque), but rather waits until it is fully typed out before printing it.
  • type: bool (store_true)
  • default: false

You can start a streaming-based conversation with node gpt.js.

ToDo

• support response streaming
• support conversations
• support different output formats (with / without IDs, json, prettify...)

Disclaimer

This project is only meant for educational purposes. Please use OpenAI's API for requests.

Also, this is a barebones example and still WIP - if you find issues / bugs, please report them 😊