LLM Web API

中文文档

ChatGPT Web Page to API interface.

Features

Bypass Cloudflare challenge
Solve Funcaptcha by Capsolver
Use without login (supports gpt3.5 model)
Auto-login with email account (supports gpt3.5, gpt-4o, gpt-4 models)
Auto re-login after session expiration
High-speed streaming output
Model switching
Multi-turn conversations
Dynamically display supported models
Support for both event-stream and websocket responses

Compatible with the ChatGPT API.

Usage

Only Docker images are supported, The current code is not up to date.

Docker

Run with Docker

docker run --name llm-web-api --rm -it -p 5000:5000 adryfish/llm-web-api

Docker compose

See detailed configuration below for environment variables.

version: '3.8'
services:
  llm-web-api:
    image: adryfish/llm-web-api
    container_name: llm-web-api
    ports:
      - "5000:5000"
    volumes:
      # Browser data. Configure if you want to retain browser login information.
      - ./browser_data:/app/browser_data
    environment:
      # PROXY_SERVER: ""          # Proxy server address
      # USER_AGENT: ""            # Browser User-Agent
      # OPENAI_LOGIN_TYPE: ""     # Login Type,nologin or email
      # OPENAI_LOGIN_EMAIL: ""    # Login email
      # OPENAI_LOGIN_PASSWORD: "" # Login password
    restart: unless-stopped

Environment

All environment variables are optional. Regarding the CAPSOLVER_API_KEY, you do not need to fill it out unless you actually encounter a FunCaptcha.

variable	description	default
PROXY_SERVER	Proxy server address	None
USER_AGENT	User-Agent	Browser default
BROWSER_DATA	Browser data storage directory	./browser_data
OPENAI_LOGIN_TYPE	ChatGPT login type, nologin or email	nologin
OPENAI_LOGIN_EMAIL	Email account for email login type	None
OPENAI_LOGIN_PASSWORD	Password for email login type	None
FUNCAPTCHA_PROVIDER	Provider name for funcaptcha	capsolver
CAPSOLVER_API_KEY	API Key for Capsolver	None

API

Currently supports the OpenAI-compatible /v1/chat/completions API, which can be accessed using OpenAI or other compatible clients.

Chat completion

Chat completion API，compatible with Openai chat-completions-api。

POST /v1/chat/completions

Request：

{
    // If you are no-login user, use gpt-3.5-turbo or gpt-4o-mini 
    // If you are a free user, use gpt-3.5-turbo, gpt-4o-mini or gpt-4o 
    // If you are a subscribed user, use gpt-3.5-turbo, gpt-4o-mini, gpt-4o, or gpt-4 for the model name.
    "model": "gpt-4o",
    "messages": [
        {
            "role": "user",
            "content": "Hello"
        }
    ],
    // If using SSE stream, set to true, default is false
    "stream": false
}

Response：

{
    "id": "chatcmpl-ZklDQbSRpTI5gzb8zzctb6fB3YDW",
    "model": "gpt-4o",
    "object": "chat.completion",
    "choices": [
        {
            "message": {
                "role": "assistant",
                "content": "Hi there! How can I assist you today?"
            },
            "index": 0,
            "finish_reason": "stop"
        }
    ],
    "usage": {
        "prompt_tokens": 1,
        "completion_tokens": 1,
        "total_tokens": 2
    },
    "created": 1716305953
}

Examples

Using OpenAI Official Library

Python

import openai

openai.api_key = 'anything'
openai.base_url = "http://localhost:5000/v1/"

completion = openai.chat.completions.create(
    model="gpt-4o-mini",
    messages=[
        {"role": "user", "content": "Hello"},
    ],
)

print(completion.choices[0].message.content)

Node.js

import OpenAI from 'openai';

const openai = new OpenAI({
	apiKey: "anything",
	baseURL: "http://localhost:5000/v1/",
});

const chatCompletion = await openai.chat.completions.create({
  messages: [{ role: 'user', content: 'Echo Hello' }],
  model: 'gpt-4o-mini',
});

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

Notes

Nginx config

If you are using Nginx as a reverse proxy for llm-web-api, add the following configuration to optimize the streaming output and improve the user experience.

# Disable proxy buffering. When set to off, Nginx sends client requests to the backend server immediately and sends responses back to the client immediately.
proxy_buffering off;

# Enable chunked transfer encoding. This allows the server to send data in chunks for dynamically generated content without knowing the size of the content in advance.
chunked_transfer_encoding on;

# Enable TCP_NOPUSH, which tells Nginx to send data as efficiently as possible before sending data packets to the client. This is often used with sendfile to improve network efficiency.
tcp_nopush on;

# Enable TCP_NODELAY, which tells Nginx not to delay sending data and to send small data packets immediately. In some cases, this can reduce network latency.
tcp_nodelay on;

# Set the keepalive timeout, here set to 120 seconds. If there is no further communication between the client and the server within this period, the connection will be closed.
keepalive_timeout 120;

Token

Since inference is not performed on the llm-web-api side, token statistics will be returned as fixed number!!!!!

Disclaimer

This project is for learning and research purposes only and is not intended for commercial use. You should be aware that using this project may violate related user agreements and understand the associated risks. We are not responsible for any losses resulting from the use of this project.

Reference

MediaCrawler: https://github.com/NanmiCoder/MediaCrawler
Bypass Cloudflare: https://github.com/sarperavci/CloudflareBypassForScraping
ChatGPT Reverse Engine: https://github.com/PawanOsman/ChatGPT

adryfish/llm-web-api