Update app.py

app.py

@@ -8,40 +8,64 @@ from flask import Flask, request, Response, stream_with_context, render_template
 app = Flask(__name__)
 
 # 🔐 --- SECURE ENVIRONMENT VARIABLES ---
-# यह आपके Hugging Face Secrets से अपने आप लोड हो जाएगा
-API_KEY = os.environ.get("NVIDIA_API_KEY") or os.environ.get("YOUR_VEDIKA_API_KEY") 
-MODEL_ID = os.environ.get("MODEL_ID")
+API_KEY = os.environ.get("YOUR_VEDIKA_API_KEY") 
+MODEL_ID = os.environ.get("MODEL_ID") # Default fallback
 
-# NVIDIA का डायरेक्ट URL (BASE_URL का लॉजिक पूरी तरह हटा दिया गया है)
+# NVIDIA's official invoke URL
 INVOKE_URL = "https://integrate.api.nvidia.com/v1/chat/completions"
 
-# 🌐 --- BULLETPROOF SCRAPER (News + Wiki) ---
+# 🌐 --- SUPER-HYBRID UNBLOCKABLE SCRAPER --- 🌐
 def web_search_scraper(query, num_results=4):
+    """
+    Yahoo Search + Google News + Wikipedia (Never blocks on Hugging Face)
+    """
     results = []
-    headers = {"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64)"}
+    headers = {
+        "User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/124.0.0.0 Safari/537.36",
+        "Accept-Language": "en-US,en;q=0.9"
+    }
 
-    # 1. Google News RSS (Never blocks, gives real latest news)
+    # 1. YAHOO SEARCH (Highly reliable for Datacenter IPs like Hugging Face)
     try:
-        rss_url = f"https://news.google.com/rss/search?q={urllib.parse.quote(query)}&hl=en-IN&gl=IN&ceid=IN:en"
-        res = requests.get(rss_url, headers=headers, timeout=5)
-        soup = BeautifulSoup(res.content, 'xml') # Use XML parser
-        items = soup.find_all('item')
+        url = f"https://search.yahoo.com/search?p={urllib.parse.quote(query)}"
+        res = requests.get(url, headers=headers, timeout=8)
+        soup = BeautifulSoup(res.text, 'html.parser')
         
-        for item in items[:num_results]:
-            title = item.title.text if item.title else ""
-            link = item.link.text if item.link else ""
-            pub_date = item.pubDate.text if item.pubDate else ""
-            if title:
-                results.append({
-                    "title": title,
-                    "link": link,
-                    "snippet": f"Published: {pub_date}"
-                })
+        for div in soup.find_all('div', class_='algo'):
+            title_tag = div.find('h3', class_='title')
+            desc_tag = div.find('div', class_='compText')
+            
+            if title_tag and desc_tag:
+                title = title_tag.text.strip()
+                link = title_tag.find('a').get('href', '') if title_tag.find('a') else ""
+                snippet = desc_tag.text.strip()
+                
+                results.append({"title": title, "link": link, "snippet": snippet})
+            
+            if len(results) >= num_results:
+                break
     except Exception as e:
-        pass
+        print(f"Yahoo Scraper Error: {e}")
 
-    # 2. Wikipedia API Fallback (Never blocks)
-    if len(results) < 2:
+    # 2. GOOGLE NEWS RSS FALLBACK (If Yahoo fails or for latest news)
+    if not results:
+        try:
+            rss_url = f"https://news.google.com/rss/search?q={urllib.parse.quote(query)}&hl=en-IN&gl=IN&ceid=IN:en"
+            res = requests.get(rss_url, headers=headers, timeout=5)
+            soup = BeautifulSoup(res.content, 'xml')
+            items = soup.find_all('item')
+            
+            for item in items[:num_results]:
+                title = item.title.text if item.title else ""
+                link = item.link.text if item.link else ""
+                pub_date = item.pubDate.text if item.pubDate else ""
+                if title:
+                    results.append({"title": title, "link": link, "snippet": f"News Published on: {pub_date}"})
+        except Exception as e:
+            print(f"News Scraper Error: {e}")
+
+    # 3. WIKIPEDIA FALLBACK
+    if not results:
         try:
             wiki_url = f"https://en.wikipedia.org/w/api.php?action=query&list=search&srsearch={urllib.parse.quote(query)}&utf8=&format=json"
             res = requests.get(wiki_url, headers=headers, timeout=5).json()
@@ -78,41 +102,37 @@ def chat():
     is_search = data.get("is_search", False)
     history = data.get("history", []) 
     max_tokens = data.get("max_tokens", 4096)
-    temperature = data.get("temperature", 1.00) # Updated to NVIDIA default if not passed
+    temperature = data.get("temperature", 1.0) # NVIDIA Recommended
 
     # 🕒 --- REAL-TIME IST INJECTION ---
     ist_time = datetime.now(timezone.utc) + timedelta(hours=5, minutes=30)
     current_date = ist_time.strftime("%A, %d %B %Y, %I:%M %p IST")
 
-    # 🧠 --- STRICT SYSTEM PROMPT (ANTI-HALLUCINATION) ---
+    # 🧠 --- GOD MODE SYSTEM PROMPT ---
     system_prompt = f"""
     You are CODE VED, an advanced AI System engineered EXCLUSIVELY by DIVY PATEL.
-    Current Date: {current_date}.
+    Current Live Date and Time: {current_date}.
     
     STRICT DIRECTIVES:
     1. NEVER invent, guess, or hallucinate product launches, dates, news, or facts.
-    2. If you are provided with live search data, base your answer PURELY on that data.
-    3. If you lack data to answer a question about a recent event, you MUST say: "I do not have the verified live data for this right now."
+    2. If you receive "LIVE WEB SEARCH RESULTS", you MUST base your answer ENTIRELY on that data.
+    3. Do NOT say "Based on the provided search results". Just answer naturally and confidently, citing the sources/links if needed.
     """
 
-    # 🚀 --- AUTO-SEARCH LOGIC WITH FAILSAFE ---
+    # 🚀 --- AUTO-SEARCH INJECTION ---
     if is_search:
         scraped_data = web_search_scraper(user_message)
         
+        search_context = "\n\n--- [LIVE VERIFIED WEB DATA] ---\n"
         if scraped_data:
-            search_context = "\n\n--- [LIVE VERIFIED WEB DATA] ---\n"
             for idx, res in enumerate(scraped_data):
                 search_context += f"{idx+1}. TITLE: {res['title']}\nSNIPPET: {res['snippet']}\nURL: {res['link']}\n\n"
             
-            search_context += "[SYSTEM: Use the above verified data to answer. Cite sources.]"
+            search_context += "[SYSTEM COMMAND: Use the above live data to answer the user accurately.]"
         else:
-            search_context = """
-            \n\n[CRITICAL SYSTEM ALERT: LIVE SEARCH FAILED OR BLOCKED. 
-            YOU HAVE NO NEW DATA FOR THIS QUERY. 
-            MANDATORY RULE: DO NOT INVENT ANY NEWS OR DATES. TELL THE USER HONESTLY THAT YOU COULD NOT FETCH THE LATEST LIVE DATA.]
-            """
+            search_context += "[SYSTEM ALERT: Live search blocked. Rely on your existing knowledge, but DO NOT hallucinate recent news.]"
             
-        user_message = f"{search_context}\n\nUSER QUERY: {user_message}"
+        user_message = f"USER QUERY: {user_message}\n\n{search_context}"
 
     messages = [{"role": "system", "content": system_prompt}]
 
@@ -130,7 +150,6 @@ def chat():
         att_type = att.get("type")
         b64_data = att.get("data")
         if att_type == "image":
-            # Image formatting compatible with standard APIs
             content_payload.append({"type": "image_url", "image_url": {"url": f"data:image/jpeg;base64,{b64_data}"}})
         elif att_type in ["audio", "file"]:
             content_payload.append({"type": "input_audio", "input_audio": {"data": b64_data, "format": "wav"}})
@@ -144,16 +163,16 @@ def chat():
         "Authorization": f"Bearer {API_KEY}", 
         "Accept": "text/event-stream"
     }
-    
-    # 🌟 --- NEW NVIDIA FORMAT PAYLOAD INTEGRATED HERE --- 🌟
+
+    # 🌟 --- NVIDIA FORMAT INTEGRATED ---
     payload = {
-        "model": MODEL_ID,  # Dynamically fetching from your secrets
+        "model": MODEL_ID,
         "messages": messages,
         "max_tokens": int(max_tokens),
         "temperature": float(temperature),
-        "top_p": 0.95, # As per your format
-        "stream": True, # Stream is kept True for fast UI response
-        "chat_template_kwargs": {"enable_thinking": True} # 👈 New NVIDIA feature added
+        "top_p": 0.95,
+        "stream": True,
+        "chat_template_kwargs": {"enable_thinking": True} # Thinking logic enabled
     }
 
     try: