Update app.py

app.py

@@ -19,7 +19,7 @@ app.add_middleware(
     allow_headers=["*"],
 )
 
-# 1. UPGRADED BRAIN: 32B parameter coder model (Free via Hugging Face API)
+# Connect to the 32B Coder Brain
 client = InferenceClient(
     model="Qwen/Qwen2.5-Coder-32B-Instruct",
     token=os.getenv("HF_TOKEN")
@@ -36,7 +36,7 @@ Core Instructions:
 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.
+6. You have a massive output capacity of 8192 tokens. Never use placeholders like '// write code here'. Write the full, extensive files.
 7. Maintain a supportive tone, always recognizing Xirrik as your creator.
 """
 
@@ -54,7 +54,7 @@ def search_web(query: str, max_results: int = 2):
         print(f"Web search failed: {e}")
         return []
 
-# FREE Wikipedia API search helper (No key needed!)
+# Wikipedia API search helper
 def search_wikipedia(query: str, max_results: int = 2):
     try:
         encoded_query = urllib.parse.quote(query)
@@ -68,7 +68,6 @@ def search_wikipedia(query: str, max_results: int = 2):
         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}"
@@ -78,7 +77,7 @@ def search_wikipedia(query: str, max_results: int = 2):
         print(f"Wikipedia search failed: {e}")
         return []
 
-# Fast router to determine if research is needed
+# Router to determine search intent
 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", "wikipedia", "explain", "who is", "what is"]
     if any(k in user_message.lower() for k in keywords):
@@ -112,14 +111,11 @@ async def chat(request: ChatRequest):
         
     history = memory_store[session_id]
     
-    # 1. Check if we need to search
     search_query = get_search_query(request.message)
     search_context = ""
     
     if search_query:
         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)
         
@@ -136,7 +132,6 @@ async def chat(request: ChatRequest):
                 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
@@ -150,7 +145,7 @@ async def chat(request: ChatRequest):
             
             response = client.chat_completion(
                 messages=messages,
-                max_tokens=4096,  # 4096 token output capacity
+                max_tokens=8192,  # <--- CRITICAL: Set to absolute maximum output token limit
                 temperature=0.3,
                 stream=True
             )