Qwen Web API Proxy

Qwen API Proxy exposes the upstream chat.qwen.ai service through an OpenAI-compatible API. It provides a local Flask server and a Python API for issuing chat completions programmatically.

Features:

• OpenAI-compatible v1/chat/completions endpoint
• Python API QwenApi for synchronous completions

Installation

Install dependencies:

pip install flask flask-cors requests

The service reads configuration from environment variables:

For authentication, you can either set the QWEN_AUTH_TOKEN environment variable or create a token.txt file in the current working directory with the token. To get the token, follow the steps:

1. login to chat.qwen.ai;
2. press F12 to open the browser's developer tools;
3. navigate to the Console tab;
4. input localStorage.getItem("token") and press Enter;
5. copy the token (without quotes) and paste it into the token.txt file or set the QWEN_AUTH_TOKEN environment variable.

Usage

Run the server

python -m qwen.server

The app binds to 0.0.0.0:<PORT> and exposes the following endpoints:

• GET /health – health check
• GET /v1/models – list models
• POST /v1/chat/completions – OpenAI-compatible chat completions
• DELETE /v1/chats/<chat_id> – delete a stored chat session upstream

Python API

from qwen import QwenApi

messages = [{"role": "user", "content": "Hello, Qwen!"}]
client = QwenApi("qwen-flash")
response: str = client(
    messages,
    enable_thinking=True,
    thinking_budget=2048,
)
print(response)

QwenApi raises ConfigurationError when the token is missing and QwenAPIError for upstream request issues.

Chat parameters

• messages (list): List of messages in the format of [{"role": "user", "content": "Hello, Qwen!"}]. Or a single string containing the user query.
• model (str): Model name. The name mapping is defined in qwen/config.py, including:

  DEFAULT_MODEL_MAP: dict[str, str] = {
      "qwen": "qwen3-max",
      "qwen-think": "qwen3-235b-a22b",  # 2507
      "qwen-coder": "qwen3-coder-plus",
      "qwen-flash": "qwen-plus-2025-09-11",  # next-80b-a3b
      "qwen-vl": "qwen3-vl-plus",  # Qwen3-VL-235B-A22B
  }
  

• enable_thinking (bool): Set True to request autodocumented reasoning. Defaults to the upstream user preference when omitted. Not available on all models.
• thinking_budget (int): Optional token budget for reasoning content. Only used when enable_thinking=True.

Notes

1. rate limit unknown - not recommended to use on main account
2. currently only supports single user message
3. streaming is not supported

License

GPLv3 License