🧠 Unofficial ChatGPT API Node.js

A developer-focused Node.js + Puppeteer-powered backend that exposes an unofficial OpenAI ChatGPT API by automating browser interaction with chat.openai.com—ideal for local testing, prompt chaining, and AI chatbot exploration without using official API keys.

🚀 Why This Project?

While OpenAI’s official APIs are powerful, they come with rate limits, cost barriers, and limited conversation thread support. This project enables developers to:

Use their personal ChatGPT account to interact with ChatGPT programmatically.
Automate login and session persistence using Puppeteer and stealth plugins.
Send prompts and get responses in a structured, customizable format.
Mimic reasoning and web search modes for enhanced answers (optional).
Simulate a local API-like development flow for chatbot prototyping and AI experimentation.

Note

“Reliance on UI behavior can cause this API to be unreliable in production; it is recommended for local use only, but I'm continually working to make it more robust."

🧰 Tech Stack

Node.js (Express) – API service
Puppeteer + Stealth Plugin – ChatGPT automation
dotenv – Credential & config management
HTML parsing (in-progress) – To extract & process response
CORS, Body-Parser – Clean JSON APIs

🛠️ Setup Guide

1. 📦 Clone & install dependencies

# Clone the repo
git clone https://github.com/roxylius/ChatGPT_unofficial_API_Node.git

# Move to the repo folder
cd ChatGPT_unofficial_API_Node

# Install all dependencies
npm install

2. ⚙️ Configure environment variables

Note: Google Auth support is not added, Signup and generate email/password for auth

Create a .env file at the project root:

OPENAI_EMAIL=your-chatgpt-login-email
OPENAI_PASSWORD=your-chatgpt-password

Replace chatgpt-login-email and your-chatgpt-password with your actual OpenAI account credentials.

3. ▶️ Run the server

Start the server by running:

node server.js

or

npm run dev

The server runs at http://localhost:3001/ and will confirm “Server is up and Running……”.

🧪 Example Test Prompt

Run a test interaction:

npm run test

Test workflow:

🌐Launch Chrome -> 📁Load Chrome-user-data -> 🔐Login -> ✉️Send Prompt -> ⏳Poll Response -> 📄Extract Text -> 💬Return JSON

🔁 API Endpoints

POST /api/prompt

As of latest ChatGPT update, reason automatically searches when required no need for both, only use Search when no Reasoning is required

Description: Sends a prompt to ChatGPT and retrieves the response.
Request Body:

{
  "prompt": "Your prompt here",
  "options": { 	// this object is optional 
    "reason": false,
    "search": true,
    "threadId": "optional_thread_id"
  }
}

or

{
  "prompt": "Your prompt here"
}

Field	Type	Description	Required
`prompt`	String	The text prompt to send to ChatGPT.	Yes
`options.reason`	Boolean	Enables Reason mode (default: false).	No
`options.search`	Boolean	Enables Search mode (default: false).	No
`options.threadId`	String	Specifies an existing conversation thread ID.	No

Response:

{
  "threadId": "the_thread_id",
  "response": "The response from ChatGPT"
}

Field	Type	Description
`threadId`	String	The ID of the conversation thread.
`response`	String	The cleaned response from ChatGPT.

Sample Response:

{
  "threadId": "681a6cba-c0fc-8004-977c-f34adf806988",
  "response": "Why don't scientists trust atoms? Because they make up everything!"
}

Note: Response times may vary based on prompt complexity and ChatGPT’s server load. Parsing may occasionally be inconsistent, particularly in Reason mode.

🔌 OpenAI-Compatible API (`/v1`)

In addition to the native /api/openai/prompt endpoint above, the server exposes an OpenAI-compatible surface so you can point any existing OpenAI SDK / client at it as a drop-in base URL.

Base URL: http://localhost:3001/v1

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "http://localhost:3001/v1",
  apiKey: "not-needed-unless-you-set-API_KEY", // any string, or your API_KEY
});

const completion = await client.chat.completions.create({
  model: "gpt-4o",
  messages: [{ role: "user", content: "Tell me a joke" }],
});

console.log(completion.choices[0].message.content);

GET /v1/models

Returns a static list of model ids for client compatibility (gpt-4o, gpt-4o-search, gpt-4o-reason, gpt-4, gpt-3.5-turbo).

POST /v1/chat/completions

Standard OpenAI request/response shape: { model, messages, stream } in, an OpenAI chat.completion (or chat.completion.chunk SSE stream, if stream: true) object out, including a usage block with estimated token counts.

Reason / Search modes — since the OpenAI schema has no field for these, pick them via the model name: gpt-4o-reason or gpt-4o-search (or any model id containing reason/search, e.g. o1, o3).
Streaming — the underlying automation only ever returns the fully-settled response text (it polls the page until the text stops changing), so stream: true re-chunks that final text into word-sized SSE deltas rather than true token-by-token generation. Clients built for SSE still work correctly; just don't expect lower latency-to-first-token from it.
Thread continuity — the API is otherwise stateless like OpenAI's, but each response includes a non-standard thread_id field with the underlying ChatGPT conversation id. The server also best-effort remembers threads across calls (matched by hashing the message history), so a normal "send full history back" chat loop will usually keep reusing the same ChatGPT thread automatically. To force it, pass thread_id explicitly in the request body.
Auth — unauthenticated by default (matches the rest of this project's local-use posture). Set API_KEY in .env to require Authorization: Bearer <API_KEY> on all /v1/* routes.
Concurrency — there's only one shared browser page under the hood, so /api/openai/prompt and /v1/chat/completions requests are queued and processed one at a time, not run in parallel.

🐳 Docker

# 1. Configure credentials
cp .env.example .env
# edit .env: set OPENAI_EMAIL / OPENAI_PASSWORD (and optionally API_KEY)

# 2. Build & run
docker compose up --build

The server is then reachable at http://localhost:3001 (native API) and http://localhost:3001/v1 (OpenAI-compatible API), exactly as when run with node server.js locally.

Why Docker needs a virtual display: this project intentionally runs Chrome headful (not headless) to avoid ChatGPT's bot detection — see the comment in src/services/puppeteerService.js. Since a container has no physical display, docker-entrypoint.sh starts Chrome under Xvfb, a virtual X server, before launching the app — no code changes needed, this happens automatically. Set PUPPETEER_HEADLESS=true in .env to skip Xvfb and run headless instead (smaller/faster, but more likely to trip login/prompt detection).

Session persistence: docker-compose.yml mounts a named volume over /app/chrome-user-data, the same directory Puppeteer already persists cookies/localStorage to (see the directory tree below). This means you generally only have to sit through the login flow once — subsequent container restarts reuse the saved session. Without that volume, every restart re-triggers performLoginWithBasicAuth.

Manual docker build / docker run (equivalent to the compose file above, if you'd rather not use Compose):

docker build -t chatgpt-unofficial-api .

docker run -d \
  --name chatgpt-api \
  -p 3001:3001 \
  --shm-size=1gb \
  --env-file .env \
  -v chatgpt-chrome-user-data:/app/chrome-user-data \
  -v chatgpt-logs:/app/logs \
  chatgpt-unofficial-api

Note: as with running this locally, headful browser automation against ChatGPT's login flow is inherently fragile — CAPTCHAs, 2FA, or new anti-bot checks can still interrupt the automated login inside the container. If it fails and you need to intervene visually, connect to the container's Chrome over its remote-debugging port with a VNC/noVNC setup, or run the equivalent flow outside Docker once to get familiar with the login UI it automates before troubleshooting the containerized version.

📂 Key Components & Dictory Tree

.
├── chrome-user-data/      # Persists browser session data (cookies, localStorage) — Docker volume
├── logs/                  # log4js output — Docker volume
├── src/
│   ├── configs/
│   │   └── log4js-config.json  # Logger config
│   ├── flows/
│   │   ├── openai_emailAuth.js   # Handles email/password login automation
│   │   └── openai_promptFlow.js  # Handles sending prompts and polling for responses
│   ├── router/
│   │   ├── index.js        # Mounts /api/openai and /v1 routers, CORS/body-parser middleware
│   │   ├── openai.js       # Native POST /api/openai/prompt route
│   │   └── v1.js            # OpenAI-compatible /v1/chat/completions + /v1/models routes
│   ├── services/
│   │   └── puppeteerService.js # Manages shared Puppeteer browser and page instance
│   └── utils/
│       ├── helpers.js       # Utility functions (e.g., login check, HTML→text, timeouts)
│       ├── logger.js        # log4js wrapper
│       └── pageLock.js      # Serializes requests against the single shared page
├── .env                   # Environment variables (OpenAI credentials, Port, API_KEY)
├── .env.example           # Example environment file
├── docker-compose.yml     # Build + run with persistent volumes
├── Dockerfile             # Headful Chrome + Xvfb image (see 🐳 Docker section)
├── docker-entrypoint.sh   # Starts Xvfb, then the app
├── example-test.js        # Standalone test script for Puppeteer automation
├── feature.md             # List of features and bug fixes
├── insights.md            # Important notes and observations
├── package.json           # Project metadata and dependencies
├── server.js              # Main application entry point, starts Express server and Puppeteer
└── README.md              # This file

🔧 Planned Features

Add markdown/HTML parser for formatted output
Add file/image support
Improve “Reason” mode polling
Signup support
Enhanced thread context management

⚠️ Known Issues

Small viewport may trigger mobile view and change behavior.
“Reason” mode writes to alternate DOM nodes. [Not working Currently!]
Some long responses split across multiple elements.

📊 Workflow Diagram

graph TD
A[Client Request] -->|/api/prompt| B[Express Server]
B --> C{Check Auth?}
C -->|Yes| D[Use existing session]
C -->|No| E[Run Login Flow]
E --> F[Persist Session]
F --> G[Load ChatGPT Page]
D --> G
G --> H[Inject Prompt]
H --> I[Poll for Response]
I --> J[Extract Response HTML/Text]
J --> K[Return JSON to Client]

👩‍💻 Author

Developed with ☕ by Roxylius

📄 License

MIT License

Navigation Menu

Repository files navigation

🧠 Unofficial ChatGPT API Node.js

🚀 Why This Project?

Note

🧰 Tech Stack

🛠️ Setup Guide

1. 📦 Clone & install dependencies

2. ⚙️ Configure environment variables

3. ▶️ Run the server

🧪 Example Test Prompt

🔁 API Endpoints

POST /api/prompt

🔌 OpenAI-Compatible API (`/v1`)

GET /v1/models

POST /v1/chat/completions

🐳 Docker

📂 Key Components & Dictory Tree

🔧 Planned Features

⚠️ Known Issues

📊 Workflow Diagram

👩‍💻 Author

📄 License

About

Releases

Used by

Contributors

Languages

Name		Name	Last commit message	Last commit date
src		src
.dockerignore		.dockerignore
.env.example		.env.example
.gitignore		.gitignore
Dockerfile		Dockerfile
LICENSE		LICENSE
Readme.md		Readme.md
bun.lockb		bun.lockb
docker-compose.yml		docker-compose.yml
docker-entrypoint.sh		docker-entrypoint.sh
example-test.js		example-test.js
feature.md		feature.md
package.json		package.json
server.js		server.js

Latest commit

History

Folders and files

Repository files navigation

🧠 Unofficial ChatGPT API Node.js

🚀 Why This Project?

Note

🧰 Tech Stack

🛠️ Setup Guide

1. 📦 Clone & install dependencies

2. ⚙️ Configure environment variables

3. ▶️ Run the server

🧪 Example Test Prompt

🔁 API Endpoints

POST /api/prompt

🔌 OpenAI-Compatible API (/v1)

GET /v1/models

POST /v1/chat/completions

🐳 Docker

📂 Key Components & Dictory Tree

🔧 Planned Features

⚠️ Known Issues

📊 Workflow Diagram

👩‍💻 Author

📄 License

About

Topics

Resources

Stars

Watchers

Forks

Releases

Used by

Contributors

Languages

🔌 OpenAI-Compatible API (`/v1`)