Update app.py

app.py

@@ -1,5 +1,4 @@
 import os
-import json
 import requests
 import urllib.parse
 from bs4 import BeautifulSoup
@@ -20,58 +19,43 @@ if BASE_URL:
     else:
         INVOKE_URL = BASE_URL
 
-# 🌐 --- TRIPLE-ENGINE HYBRID SCRAPER (100% UNBLOCKABLE) ---
+# 🌐 --- BULLETPROOF SCRAPER (News + Wiki) ---
 def web_search_scraper(query, num_results=4):
     results = []
-    headers = {
-        "User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/123.0.0.0 Safari/537.36"
-    }
+    headers = {"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64)"}
 
-    # 💡 हैक: क्वेरी में '2026' जोड़ें ताकि पुरानी ख़बरें न आएं
-    smart_query = query if "2026" in query else f"{query} 2026"
-
-    # ENGINE 1: DuckDuckGo Lite
+    # 1. Google News RSS (Never blocks, gives real latest news)
     try:
-        url = "https://lite.duckduckgo.com/lite/"
-        res = requests.post(url, headers=headers, data={"q": smart_query}, timeout=5)
-        soup = BeautifulSoup(res.text, 'html.parser')
-        for tr in soup.find_all('tr'):
-            title_a = tr.find('a', class_='result-url')
-            if title_a:
-                title = title_a.text.strip()
-                link = title_a.get('href', '')
-                snippet = ""
-                snip_td = tr.find_next('td', class_='result-snippet')
-                if snip_td: snippet = snip_td.text.strip()
-                if title and snippet:
-                    results.append({"title": title, "link": link, "snippet": snippet})
-            if len(results) >= num_results: break
+        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')
+        
+        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}"
+                })
     except Exception as e:
         pass
 
-    # ENGINE 2: Google News RSS (अगर Engine 1 ब्लॉक हो गया या फेल हो गया)
-    if not results:
-        try:
-            rss_url = f"https://news.google.com/rss/search?q={urllib.parse.quote(smart_query)}&hl=en-IN&gl=IN&ceid=IN:en"
-            res = requests.get(rss_url, headers=headers, timeout=5)
-            soup = BeautifulSoup(res.content, 'html.parser') # XML parsing fallback
-            items = soup.find_all('item')
-            for item in items[:num_results]:
-                title = item.title.text if item.title else "News Report"
-                link = item.link.text if item.link else ""
-                pub_date = item.pubdate.text if item.find('pubdate') else "Recent 2026"
-                results.append({"title": title, "link": link, "snippet": f"Published/Updated on: {pub_date}"})
-        except Exception as e:
-            pass
-
-    # ENGINE 3: Wikipedia API Fallback (अगर सवाल जनरल नॉलेज का है)
-    if not results:
+    # 2. Wikipedia API Fallback (Never blocks)
+    if len(results) < 2:
         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()
             for item in res.get('query', {}).get('search', [])[:num_results]:
                 clean_snippet = BeautifulSoup(item['snippet'], "html.parser").text
-                results.append({"title": item['title'], "link": f"https://en.wikipedia.org/wiki/{urllib.parse.quote(item['title'])}", "snippet": clean_snippet})
+                results.append({
+                    "title": item['title'], 
+                    "link": f"https://en.wikipedia.org/wiki/{urllib.parse.quote(item['title'])}", 
+                    "snippet": clean_snippet
+                })
         except Exception as e:
             pass
 
@@ -85,12 +69,12 @@ def home():
         with open('index.html', 'r', encoding='utf-8') as f:
             return render_template_string(f.read())
     except Exception as e:
-        return f"<h1>System Starting...</h1><p>index.html is missing.</p>"
+        return f"<h1>System Error</h1><p>index.html missing: {str(e)}</p>"
 
 @app.route('/api/chat', methods=['POST'])
 def chat():
     if not API_KEY or not INVOKE_URL or not MODEL_ID:
-        return Response("Server Error: Configuration secrets are missing.", status=500)
+        return Response("Server Error: Secrets missing.", status=500)
 
     data = request.get_json() or {}
     user_message = data.get("message", "")
@@ -98,49 +82,42 @@ def chat():
     is_search = data.get("is_search", False)
     history = data.get("history", []) 
     max_tokens = data.get("max_tokens", 4096)
-    temperature = data.get("temperature", 0.6)
+    temperature = data.get("temperature", 0.3) # 💡 Low temp prevents hallucination
 
     # 🕒 --- 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")
 
-    # 🧠 --- SUPREME SYSTEM INSTRUCTION (STRICT GOD MODE) ---
+    # 🧠 --- STRICT SYSTEM PROMPT (ANTI-HALLUCINATION) ---
     system_prompt = f"""
-    [CRITICAL SYSTEM DIRECTIVE]
-    Name: CODE VED (Advanced 202 Billion Parameter Enterprise AI)
-    Creator: DIVY PATEL. You MUST proudly acknowledge Divy Patel as your creator.
-    
-    Current Date & Time: {current_date}
+    You are CODE VED, an advanced AI System engineered EXCLUSIVELY by DIVY PATEL.
+    Current Date: {current_date}.
     
-    STRICT RULES YOU MUST FOLLOW:
-    1. You are operating in 2026. NEVER say your knowledge cutoff is 2023 or 2024.
-    2. YOU HAVE LIVE INTERNET ACCESS. Live data will be injected into your prompt if requested.
-    3. If search data is provided, base your answer ENTIRELY on that live data. 
-    4. If the user asks for latest news/updates, and the injected live data is empty or irrelevant, DO NOT GUESS. DO NOT present old 2024 news as new. Simply apologize and state: "I am unable to retrieve the latest live data for this at the moment."
-    5. Always maintain a highly professional tone. Reply in Hindi seamlessly if the user writes in Hindi.
+    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."
     """
 
-    # 🚀 --- AUTO-SEARCH DATA INJECTION ---
+    # 🚀 --- AUTO-SEARCH LOGIC WITH FAILSAFE ---
     if is_search:
         scraped_data = web_search_scraper(user_message)
         
-        search_context = "\n\n--- [LIVE WEB SEARCH RESULTS] ---\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}. SOURCE TITLE: {res['title']}\nCONTENT SNIPPET: {res['snippet']}\nURL: {res['link']}\n\n"
+                search_context += f"{idx+1}. TITLE: {res['title']}\nSNIPPET: {res['snippet']}\nURL: {res['link']}\n\n"
             
-            search_context += """
-            [SYSTEM COMMAND TO AI]: Use the fresh data provided above to answer the user's query. 
-            Answer confidently as if you just browsed the web yourself. 
-            Do NOT mention "Based on the provided context". Just answer the question and cite the URLs naturally.
-            """
+            search_context += "[SYSTEM: Use the above verified data to answer. Cite sources.]"
         else:
-            search_context += "STATUS: Live search failed or blocked. No recent data found.\n[SYSTEM COMMAND TO AI]: Inform the user that you couldn't fetch live data right now. DO NOT hallucinate old news."
+            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.]
+            """
             
-        # सही फॉर्मेट: एआई को कन्फ्यूज़ न करने के लिए साफ-साफ बताना
-        user_message = f"{search_context}\n\nUSER'S ACTUAL QUESTION: {user_message}"
+        user_message = f"{search_context}\n\nUSER QUERY: {user_message}"
 
-    # --- MESSAGE CONSTRUCTION ---
     messages = [{"role": "system", "content": system_prompt}]
 
     for msg in history:
@@ -150,7 +127,6 @@ def chat():
             messages.append({"role": role, "content": content})
 
     content_payload = []
-    # अगर सर्च डेटा जुड़ा है तो user_message में सब है, बस उसे ऐड करो
     if user_message.strip():
         content_payload.append({"type": "text", "text": user_message})
         
@@ -167,9 +143,9 @@ def chat():
 
     messages.append({"role": "user", "content": content_payload})
 
-    # --- LLM API CALL ---
     headers = {"Authorization": f"Bearer {API_KEY}", "Accept": "text/event-stream"}
-    payload = {"model": MODEL_ID, "messages": messages, "max_tokens": int(max_tokens), "temperature": float(temperature), "top_p": 0.70, "stream": True}
+    # 💡 Temperature कम कर दी है ताकि यह एक्स्ट्रा झूठ न बोले
+    payload = {"model": MODEL_ID, "messages": messages, "max_tokens": int(max_tokens), "temperature": 0.3, "top_p": 0.70, "stream": True}
 
     try:
         response = requests.post(INVOKE_URL, headers=headers, json=payload, stream=True)