chatgpt-api/readme.md

12 KiB

Update December 11, 2022

Today, OpenAI added additional Cloudflare protections that make it more difficult to access the unofficial API. This is affecting all ChatGPT API wrappers at the moment, including the Python ones. See this issue.

As a temporary workaround, make sure you're using the latest version of this package and Node.js >= 18:

  1. Log into https://chat.openai.com/chat and copy a fresh session token (same instructions as below).

  2. Copy the value of the cf_clearance cookie and store it in a CLEARANCE_TOKEN environment variable in addition to your SESSION_TOKEN.

  3. Copy your browser's user-agent header from any request in your browser's network tab.

  4. Use both tokens when creating the API wrapper:

const api = new ChatGPTApI({
  sessionToken: process.env.SESSION_TOKEN,
  clearanceToken: process.env.CLEARANCE_TOKEN,
  userAgent: '' // replace to match your real browser's user agent
})

await api.ensureAuth()

Restrictions on this method:

  • Cloudflare cf_clearance tokens expire after 2 hours, so right now you'll have to manually log in and extract it by hand every so often
  • Your user-agent and IP address must match from the real browser window you're logged in with to the one you're using for ChatGPTAPI.
    • This means that you currently can't log in with your laptop and then run the bot on a server or proxy somewhere.
  • You must use node >= 18. I'm using v19.2.0 in my testing, but for some reason, all fetch requests using Node.js v16 and v17 fail at the moment (these use undici under the hood, whereas Node.js v18 and above use a built-in fetch based on undici).
  • You should not be using this account while the bot is using it, because that browser window may refresh one of your tokens and invalidate the bot's session.

We're working hard in this issue to make this process easier and more automated.

Cheers, Travis


Example usage

ChatGPT API

Node.js client for the unofficial ChatGPT API.

NPM Build Status MIT License Prettier Code Formatting

Intro

This package is a Node.js wrapper around ChatGPT by OpenAI. TS batteries included.

You can use it to start building projects powered by ChatGPT like chatbots, websites, etc...

Install

npm install chatgpt

Usage

import { ChatGPTAPI } from 'chatgpt'

async function example() {
  // sessionToken is required; see below for details
  const api = new ChatGPTAPI({
    sessionToken: process.env.SESSION_TOKEN
  })

  // ensure the API is properly authenticated
  await api.ensureAuth()

  // send a message and wait for the response
  const response = await api.sendMessage(
    'Write a python version of bubble sort.'
  )

  // response is a markdown-formatted string
  console.log(response)
}

ChatGPT responses are formatted as markdown by default. If you want to work with plaintext instead, you can use:

const api = new ChatGPTAPI({
  sessionToken: process.env.SESSION_TOKEN,
  markdown: false
})

If you want to automatically track the conversation, you can use ChatGPTAPI.getConversation():

const api = new ChatGPTAPI({
  sessionToken: process.env.SESSION_TOKEN
})

const conversation = api.getConversation()

// send a message and wait for the response
const response0 = await conversation.sendMessage('What is OpenAI?')

// send a follow-up prompt to the previous message and wait for the response
const response1 = await conversation.sendMessage('Can you expand on that?')

// send another follow-up to the same conversation
const response2 = await conversation.sendMessage('Oh cool; thank you')

Sometimes, ChatGPT will hang for an extended period of time before beginning to respond. This may be due to rate limiting or it may be due to OpenAI's servers being overloaded.

To mitigate these issues, you can add a timeout like this:

// timeout after 2 minutes (which will also abort the underlying HTTP request)
const response = await api.sendMessage('this is a timeout test', {
  timeoutMs: 2 * 60 * 1000
})

You can stream responses using the onProgress or onConversationResponse callbacks. See the docs for more details.

Usage in CommonJS (Dynamic import)
async function example() {
  // To use ESM in CommonJS, you can use a dynamic import
  const { ChatGPTAPI } = await import('chatgpt')

  const api = new ChatGPTAPI({
    sessionToken: process.env.SESSION_TOKEN
  })
  await api.ensureAuth()

  const response = await api.sendMessage('Hello World!')
  console.log(response)
}

Docs

See the auto-generated docs for more info on methods and parameters.

Demos

A basic demo is included for testing purposes:

# 1. clone repo
# 2. install node deps
# 3. set `SESSION_TOKEN` in .env
# 4. run:
npx tsx src/demo.ts

A conversation demo is also included:

# 1. clone repo
# 2. install node deps
# 3. set `SESSION_TOKEN` in .env
# 4. run:
npx tsx src/demo-conversation.ts

Session Tokens

This package requires a valid session token from ChatGPT to access it's unofficial REST API.

To get a session token:

  1. Go to https://chat.openai.com/chat and log in or sign up.
  2. Open dev tools.
  3. Open Application > Cookies. ChatGPT cookies
  4. Copy the value for __Secure-next-auth.session-token and save it to your environment.

If you want to run the built-in demo, store this value as SESSION_TOKEN in a local .env file.

Note This package will switch to using the official API once it's released.

Note Prior to v1.0.0, this package used a headless browser via Playwright to automate the web UI. Here are the docs for the initial browser version.

Projects

All of these awesome projects are built using the chatgpt package. 🤯

If you create a cool integration, feel free to open a PR and add it to the list.

Compatibility

This package is ESM-only. It supports:

  • Node.js >= 16.8
    • If you need Node.js 14 support, use v1.4.0
  • Edge runtimes like CF workers and Vercel edge functions
  • Modern browsers
    • Mainly meant for chrome extensions where your code is protected to a degree
    • We recommend against using chatgpt from client-side browser code because it would expose your private session token
    • If you want to build a website using chatgpt, we recommend using it only from your backend API

Credits

License

MIT © Travis Fischer

If you found this project interesting, please consider sponsoring me or following me on twitter twitter