A lightweight, OpenAI-compatible proxy server for USTC Chat. It exposes the USTC Chat backend as a standard /v1/chat/completions endpoint so you can use it with any OpenAI client.

Alternative: If you have access to the USTC LLM Public Service Platform, you can call models directly via the official API without this proxy. See the Platform API Guide below.

• OpenAI-compatible API — Drop-in replacement for OpenAI Chat Completions (/v1/chat/completions)
• Streaming & non-streaming — Full SSE streaming support with automatic recovery
• Tool calls — Supports function-calling with retry guidance for large payloads
• Reasoning content — Exposes DeepSeek reasoning chains when available
• Auto-retry — Recovers from idle timeouts, chunked encoding errors, and connection issues
• Queue management — Handles USTC queue system transparently (tick → ai-mother → wait queue → chat)
• Request capture — Optional JSONL logging of requests/responses for debugging

1. Get your USTC Chat token

   • Log in to https://chat.ustc.edu.cn/ustchat/
   • Open browser console (F12) and run:

     copy(JSON.parse(localStorage.getItem('ustchat-user-store')).state.token)

   • Paste the token into ustc.chat.txt in the same directory as the proxy script.

2. Run the proxy

   python ustc_openai_proxy.py

   The server listens on http://127.0.0.1:1958 by default.

3. Configure your client

   • Base URL: http://127.0.0.1:1958/v1
   • API Key: any non-empty string (real auth uses ustc.chat.txt)
   • Model: deepseek-v4-pro (default) or deepseek-v4-flash

oai4ustc/
├── ustc_openai_proxy.py   # Main proxy server
├── ustc.chat.txt          # USTC Chat token (create manually)
└── README.md              # This file

• The proxy reads ustc.chat.txt on every request, so you can update the token without restarting.
• deepseek-v4-pro defaults to enable_thinking=true with reasoning_effort=high.
• Responses API is not supported; only Chat Completions.

The USTC LLM Public Service Platform provides direct API access to multiple large language models hosted on campus. The API is OpenAI-compatible, so any OpenAI SDK or tool works out of the box.

1. Log in — Visit https://llm.ustc.edu.cn and authenticate via the university's unified identity system (统一身份认证).
2. Create a project — Go to Project Management → New Project. Provide a project name, usage scenario (research / teaching / development), and contact info.
3. Request an API Key — Enter the project detail page → API Key Management → Create API Key. You will receive a key in the format sk-xxxxxxxxxxxxxxxx. Save it immediately — it is only shown once.

curl https://api.llm.ustc.edu.cn/v1/chat/completions \
  -H "Authorization: Bearer sk-YourKey" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4-flash",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

from openai import OpenAI

client = OpenAI(
    api_key="sk-YourKey",
    base_url="https://api.llm.ustc.edu.cn/v1"
)

response = client.chat.completions.create(
    model="deepseek-v4-flash",
    messages=[{"role": "user", "content": "Hello"}],
    temperature=0.3
)

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

stream = client.chat.completions.create(
    model="deepseek-v4-flash",
    messages=[{"role": "user", "content": "Explain LLMs in detail."}],
    stream=True
)

for chunk in stream:
    content = chunk.choices[0].delta.content
    if content:
        print(content, end="")

• Do not expose API keys in frontend code or public repositories.
• Use separate API keys per project.
• Enable stream: true for long-form generation.
• Lower temperature for more deterministic outputs.
• Use retry and rate-limiting mechanisms for high-concurrency scenarios.
• Human-review any critical research conclusions generated by models.

For full documentation, see the Platform User Guide.