app.py

import os
import gradio as gr
from gradio_client import Client

# Configuration
GRADIO_SPACE = os.environ.get("GRADIO_SPACE", "amd/gpt-oss-120b-chatbot")
API_KEY = os.environ.get("PROXY_API_KEY")
GRADIO_API = "chat"  # remove leading slash for gradio_client
client = Client(GRADIO_SPACE)

def chat_proxy(messages, api_key=""):
    # Check API key
    if not API_KEY:
        return {"error": "Server API key not set."}
    if api_key != API_KEY:
        return {"error": "Invalid API key."}

    # Extract user and system messages
    user_message = ""
    system_prompt = "You are a helpful assistant."
    for m in reversed(messages):
        if m["role"] == "user" and not user_message:
            user_message = m["content"]
        elif m["role"] == "system":
            system_prompt = m["content"]

    # Call Gradio space
    try:
        result = client.predict(
            message=user_message,
            system_prompt=system_prompt,
            temperature=0.7,
            api_name=GRADIO_API
        )
    except Exception as e:
        return {"error": f"Gradio client error: {e}"}

    # Return OpenAI-compatible response
    return {
        "id": "chatcmpl-proxy",
        "object": "chat.completion",
        "created": 0,
        "model": "gradio-proxy",
        "choices": [
            {
                "index": 0,
                "message": {"role": "assistant", "content": str(result)},
                "finish_reason": "stop"
            }
        ],
        "usage": {"prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0}
    }

# Expose as a pure API via Gradio Interface
demo = gr.Interface(
    fn=chat_proxy,
    inputs=[gr.JSON(label="Messages"), gr.Textbox(label="API Key")],
    outputs=gr.JSON(label="Response")
)

# Launch without share (HF Spaces does not support share)
demo.launch()