This solution is not lightweight, but it does work a lot more consistently than the previous REST API-based approach. For example, I'm currently using this approach to automate 10 concurrent OpenAI accounts for my [Twitter bot](https://github.com/transitive-bullshit/chatgpt-twitter-bot). 😂
To use the updated version, **make sure you're using the latest version of this package and Node.js >= 18**. Then update your code following the examples below, paying special attention to the sections on [Authentication](#authentication), [Restrictions](#restrictions), and [CAPTCHAs](#captchas).
We recently added support for CAPTCHA automation using either [nopecha](https://nopecha.com/) or [2captcha](https://2captcha.com). Keep in mind that this package will be updated to use the official API as soon as it's released, so things should get much easier over time. 💪
Lastly, please consider starring this repo and <ahref="https://twitter.com/transitive_bs">following me on twitter <imgsrc="https://storage.googleapis.com/saasify-assets/twitter-logo.svg"alt="twitter"height="24px"align="center"></a> to help support the project.
`puppeteer` is an optional peer dependency used to automate bypassing the Cloudflare protections via `getOpenAIAuth`. The main API wrapper uses `fetch` directly.
If you want to track the conversation, use the `conversationId` and `messageId` in the result object, and pass them to `sendMessage` as `conversationId` and `parentMessageId` respectively.
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.
See the [auto-generated docs](./docs/classes/ChatGPTAPI.md) for more info on methods and parameters. Here are the [docs](./docs/classes/ChatGPTAPIBrowser.md) for the browser-based version.
The authentication section relates to the REST-based version (using `getOpenAIAuth` + `ChatGPTAPI`). The browser-based solution, `ChatGPTAPIBrowser`, takes care of all the authentication for you.
We've provided an automated, Puppeteer-based solution `getOpenAIAuth` to fetch these for you, but you may still run into cases where you have to manually pass the CAPTCHA. We're working on a solution to automate this further.
You can also get these tokens manually, but keep in mind that the `clearanceToken` only lasts for max 2 hours.
6. Copy the value of the `user-agent` header from any request in your `Network` tab, or copy the result of `navigator.userAgent` command on `Console` tab. This will be your `userAgent`.
Basic Cloudflare CAPTCHAs are handled by default, but if you want to automate the email + password Recaptchas, you'll need to sign up for one of these paid providers:
- [nopecha](https://nopecha.com/) - Uses AI to solve CAPTCHAS
- Faster and cheaper
- Set the `NOPECHA_KEY` env var to your nopecha API key
- [Demo video](https://user-images.githubusercontent.com/552829/208235991-de4890f2-e7ba-4b42-bf55-4fcd792d4b19.mp4) of nopecha solving the login Recaptcha (41 seconds)
- [2captcha](https://2captcha.com) - Uses real people to solve CAPTCHAS
- More well-known solution that's been around longer
- Set the `CAPTCHA_TOKEN` env var to your 2captcha API token
Alternatively, if your OpenAI account uses Google Auth, you shouldn't encounter any of the more complicated Recaptchas —and can avoid using these third-party providers. To use Google auth, make sure your OpenAI account is using Google and then set `isGoogleLogin` to `true` whenever you're passing your `email` and `password`. For example:
The browser implementation supports setting a proxy server. This is useful if you're running into rate limiting issues or if you want to use a proxy to hide your IP address.
To use a proxy, pass the `proxyServer` option to the `ChatGPTAPIBrowser` constructor, or simply set the `PROXY_SERVER` env var. For more information on the format, see [here](https://www.chromium.org/developers/design-documents/network-settings).
```ts
const api = new ChatGPTAPIBrowser({
email: process.env.OPENAI_EMAIL,
password: process.env.OPENAI_PASSWORD,
proxyServer: '<ip>:<port>'
})
```
You can also set the `PROXY_VALIDATE_IP` env var to your proxy's IP address. This will be used to validate that the proxy is working correctly, and will throw an error if it's not.
These restrictions are for the `getOpenAIAuth` + `ChatGPTAPI` solution, which uses the unofficial API. The browser-based solution, `ChatGPTAPIBrowser`, generally doesn't have any of these restrictions.
- 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.
> Prior to v1.0.0, this package used a headless browser via [Playwright](https://playwright.dev/) to automate the web UI. Here are the [docs for the initial browser version](https://github.com/transitive-bullshit/chatgpt-api/tree/v0.4.2).
- Node.js 17, 16, and 14 were supported in earlier versions, but OpenAI's Cloudflare update caused a bug with `undici` on v17 and v16 that needs investigation. So for now, use `node >= 18`
- 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
- Huge thanks to [@waylaidwanderer](https://github.com/waylaidwanderer), [@abacaj](https://github.com/abacaj), [@wong2](https://github.com/wong2), [@simon300000](https://github.com/simon300000), [@RomanHotsiy](https://github.com/RomanHotsiy), [@ElijahPepe](https://github.com/ElijahPepe), and all the other contributors 💪
- The original browser version was inspired by this [Go module](https://github.com/danielgross/whatsapp-gpt) by [Daniel Gross](https://github.com/danielgross)
If you found this project interesting, please consider [sponsoring me](https://github.com/sponsors/transitive-bullshit) or <ahref="https://twitter.com/transitive_bs">following me on twitter <imgsrc="https://storage.googleapis.com/saasify-assets/twitter-logo.svg"alt="twitter"height="24px"align="center"></a>
