Update api/utils.py

api/utils.py

@@ -6,7 +6,8 @@ import random
 import string
 from typing import Any, Dict, Optional
 import os
-from fastapi import HTTPException, Request
+from fastapi import HTTPException, Request, Response
+from fastapi.responses import JSONResponse
 from dotenv import load_dotenv
 
 import httpx
@@ -60,7 +61,7 @@ def check_rate_limit(ip: str):
                 logger.warning(f"Rate limit exceeded for IP {ip}.")
                 raise HTTPException(
                     status_code=429,
-                    detail=f"Rate limit exceeded. Maximum {REQUEST_LIMIT_PER_MINUTE} requests per minute allowed."
+                    detail={"error": {"message": "Rate limit exceeded. Please wait and try again.", "type": "rate_limit"}},
                 )
 
 # Helper function to create chat completion data