backend/models/gateway.py

import os
import httpx
from typing import Dict, Any

DEFAULT_MODEL = os.getenv("DEFAULT_MODEL", "zai-glm-4.7")

CEREBRAS_API_KEY = os.getenv("CEREBRAS_API_KEY", "").strip()
CEREBRAS_URL = "https://api.cerebras.ai/v1/chat/completions"


class ModelGateway:
    def __init__(self):
        self.client = httpx.AsyncClient(
            timeout=httpx.Timeout(180.0),
            follow_redirects=True,
        )

    async def generate(
        self,
        model: str,
        prompt: str,
        system: str = "You are DOLOR3V Autonomous Builder.",
    ) -> Dict[str, Any]:

        if not CEREBRAS_API_KEY:
            raise RuntimeError("CEREBRAS_API_KEY is not configured")

        payload = {
            "model": model,
            "messages": [
                {
                    "role": "system",
                    "content": system,
                },
                {
                    "role": "user",
                    "content": prompt,
                },
            ],
            "temperature": 0.2,
            "max_tokens": 4096,
        }

        response = await self.client.post(
            CEREBRAS_URL,
            headers={
                "Authorization": f"Bearer {CEREBRAS_API_KEY}",
                "Content-Type": "application/json",
            },
            json=payload,
        )

        response.raise_for_status()

        data = response.json()

        if "choices" not in data:
            raise RuntimeError(f"Unexpected Cerebras response: {data}")

        text = (
            data["choices"][0]
            .get("message", {})
            .get("content", "")
            .strip()
        )

        return {
            "text": text,
            "provider": "cerebras",
            "model": model,
            "raw": data,
        }

    async def close(self):
        await self.client.aclose()


model_gateway = ModelGateway()