Update app.py

app.py

@@ -1,11 +1,13 @@
 import os
-import torch
+import json
+import urllib.request
+import urllib.parse
 from fastapi import FastAPI
 from pydantic import BaseModel
 from huggingface_hub import InferenceClient
 from fastapi.middleware.cors import CORSMiddleware
 from fastapi.responses import StreamingResponse
-from duckduckgo_search import DDGS # <--- NEW: Free, zero-key web search!
+from duckduckgo_search import DDGS
 
 app = FastAPI()
 
@@ -17,49 +19,74 @@ app.add_middleware(
     allow_headers=["*"],
 )
 
-# Initialize the world-class coding brain
+# 1. UPGRADED BRAIN: 32B parameter coder model (Free via Hugging Face API)
 client = InferenceClient(
-    model="Qwen/Qwen2.5-Coder-7B-Instruct",
+    model="Qwen/Qwen2.5-Coder-32B-Instruct",
     token=os.getenv("HF_TOKEN")
 )
 
 memory_store = {}
 
-# System prompt optimized for reasoning, long code writing, and web search integration
 PRO_IDENTITY = """
-You are Zenvy Quantum Pro, an elite AI software engineer created by Xirrik.
+You are Zenvy Quantum Pro, an elite 32-Billion parameter AI software engineer created by Xirrik.
 
 Core Instructions:
 1. Always start your response with a thinking phase wrapped inside <thinking>...</thinking> tags. In this phase, analyze Xirrik's request step-by-step and plan your code architecture.
-2. If web search results are provided in your context, explicitly state: "[WEB SEARCH ENGAGED: "search query"]" inside your <thinking> block, and explain how the retrieved search context helps you write accurate, up-to-date code.
-3. After the closing </thinking> tag, write out your clean, complete, and fully commented code.
-4. You have a high-capacity output window of 4096 tokens. Never use placeholders like '// write code here'. Write the full file.
-5. Maintain a supportive tone, always recognizing Xirrik as your creator.
+2. If web search results are provided in your context, explicitly state: "[WEB SEARCH ENGAGED: "search query"]" inside your <thinking> block.
+3. If Wikipedia results are provided in your context, explicitly state: "[WIKIPEDIA INDEX ENGAGED: "search query"]" inside your <thinking> block.
+4. Explain how these retrieved results help you write accurate, up-to-date, or historically correct code.
+5. After the closing </thinking> tag, write out your clean, complete, and fully commented code.
+6. You have a high-capacity output window of 4096 tokens. Never use placeholders like '// write code here'. Write the full file.
+7. Maintain a supportive tone, always recognizing Xirrik as your creator.
 """
 
 class ChatRequest(BaseModel):
     message: str
     session_id: str
 
-# Helper function to perform a web search
-def search_web(query: str, max_results: int = 3):
+# DuckDuckGo search helper
+def search_web(query: str, max_results: int = 2):
     try:
         with DDGS() as ddgs:
             results = list(ddgs.text(query, max_results=max_results))
             return results
     except Exception as e:
-        print(f"Search engine error: {e}")
+        print(f"Web search failed: {e}")
         return []
 
-# Helper function to decide if a search is needed and generate the query
+# FREE Wikipedia API search helper (No key needed!)
+def search_wikipedia(query: str, max_results: int = 2):
+    try:
+        encoded_query = urllib.parse.quote(query)
+        search_url = f"https://en.wikipedia.org/w/api.php?action=query&list=search&srsearch={encoded_query}&format=json"
+        
+        req = urllib.request.Request(search_url, headers={'User-Agent': 'ZenvyWorkspace/1.0'})
+        with urllib.request.urlopen(req) as response:
+            data = json.loads(response.read().decode())
+        
+        search_results = data.get("query", {}).get("search", [])
+        output = []
+        for item in search_results[:max_results]:
+            title = item.get("title")
+            # Strip Wikipedia's search highlighting tags
+            snippet = item.get("snippet", "").replace('<span class="searchmatch">', "").replace('</span>', "")
+            page_id = item.get("pageid")
+            link = f"https://en.wikipedia.org/?curid={page_id}"
+            output.append({"title": title, "body": snippet, "href": link})
+        return output
+    except Exception as e:
+        print(f"Wikipedia search failed: {e}")
+        return []
+
+# Fast router to determine if research is needed
 def get_search_query(user_message: str) -> str:
-    keywords = ["search", "recent", "latest", "how to", "current", "news", "2025", "2026", "updated", "documentation", "error", "bug", "why is", "release", "api change"]
+    keywords = ["search", "recent", "latest", "how to", "current", "news", "2025", "2026", "updated", "documentation", "error", "bug", "why is", "release", "api change", "wikipedia", "explain", "who is", "what is"]
     if any(k in user_message.lower() for k in keywords):
         try:
             router_prompt = (
                 f"Analyze this user request: '{user_message}'. "
-                "If it requires up-to-date documentation or real-time web search to answer accurately, "
-                "generate a single, optimized search engine query. "
+                "If it requires up-to-date documentation, real-time web search, or encyclopedic knowledge to answer accurately, "
+                "generate a single, optimized search query. "
                 "If no search is needed, reply with 'NONE'. Do not output any other text."
             )
             res = client.chat_completion(
@@ -75,7 +102,7 @@ def get_search_query(user_message: str) -> str:
 
 @app.get("/")
 def home():
-    return {"status": "Zenvy Pro Search Core is Online"}
+    return {"status": "Zenvy Pro 32B Search Core is Online"}
 
 @app.post("/chat")
 async def chat(request: ChatRequest):
@@ -85,19 +112,31 @@ async def chat(request: ChatRequest):
         
     history = memory_store[session_id]
     
-    # 1. Check if we need to search the web first
+    # 1. Check if we need to search
     search_query = get_search_query(request.message)
     search_context = ""
     
     if search_query:
-        print(f"Web Search Engaged: {search_query}")
-        results = search_web(search_query)
-        if results:
-            search_context = "\n\n--- LIVE WEB SEARCH RESULTS (Up-to-date Context) ---\n"
-            for r in results:
-                search_context += f"Source: {r.get('href', 'N/A')}\nTitle: {r.get('title', 'N/A')}\nSnippet: {r.get('body', 'N/A')}\n\n"
-    
-    # 2. Append the user message (with search results if found) to memory
+        print(f"Retrieving data for: {search_query}")
+        
+        # Query both DuckDuckGo and Wikipedia in parallel
+        web_results = search_web(search_query)
+        wiki_results = search_wikipedia(search_query)
+        
+        if web_results or wiki_results:
+            search_context = "\n\n--- RESEARCH DATA RETRIEVED ---\n"
+            
+            if wiki_results:
+                search_context += f"\n[WIKIPEDIA INDEX ENGAGED: \"{search_query}\"]\n"
+                for r in wiki_results:
+                    search_context += f"Source: {r['href']}\nTitle: {r['title']}\nSummary: {r['body']}\n\n"
+            
+            if web_results:
+                search_context += f"\n[WEB SEARCH ENGAGED: \"{search_query}\"]\n"
+                for r in web_results:
+                    search_context += f"Source: {r.get('href', 'N/A')}\nTitle: {r.get('title', 'N/A')}\nSnippet: {r.get('body', 'N/A')}\n\n"
+
+    # 2. Append history
     user_payload = request.message
     if search_context:
         user_payload += search_context
@@ -111,7 +150,7 @@ async def chat(request: ChatRequest):
             
             response = client.chat_completion(
                 messages=messages,
-                max_tokens=4096,  # Raised output token capacity for long, complete files
+                max_tokens=4096,  # 4096 token output capacity
                 temperature=0.3,
                 stream=True
             )
@@ -123,7 +162,6 @@ async def chat(request: ChatRequest):
                         collected_reply.append(word)
                         yield word
             
-            # Save final response to history
             full_reply = "".join(collected_reply)
             history.append({"role": "assistant", "content": full_reply})