Update app.py

app.py

@@ -597,7 +597,7 @@ class SearchRouter:
 # Initialize the router globally
 search_router = SearchRouter()
 
-def build_smart_prompt(conversation_history: List[Dict[str, str]], context: str = "", original_prompt: str = "") -> str:
+def build_smart_prompt(conversation_history: List[Dict[str, str]], context: str = "", original_prompt: str = "", enable_thinking: bool = False) -> str:
     """
     Builds an intelligent prompt that defines the 'ToolBoxesAI Assistant' persona
     and enforces strict adherence to provided context (Web/RAG) to prevent hallucinations.
@@ -652,7 +652,7 @@ def build_smart_prompt(conversation_history: List[Dict[str, str]], context: str
         final_messages,
         tokenize=False,
         add_generation_prompt=True,
-        enable_thinking=False
+        enable_thinking=enable_thinking
     )
     
     return prompt_str
@@ -804,6 +804,7 @@ class PromptRequest(BaseModel):
     temperature: float = 0.7
     enable_code_execution: bool = True
     enable_web_search: bool = True
+    enable_thinking: bool
 
 @app.get("/")
 async def root():
@@ -865,7 +866,7 @@ async def chat_with_model_async(request: PromptRequest):
         logger.info(f"🛠 Tool selected: {tool_result['tool_name'] or 'None'}")
 
         # Step 3: Build prompt and prepare streaming response
-        prompt_str = build_smart_prompt(conversation_history, context, request.prompt)
+        prompt_str = build_smart_prompt(conversation_history, context, request.prompt, enable_thinking=request.enable_thinking)
         
         # Model generation (still needs to run in thread due to PyTorch limitations)
         inputs = tokenizer(prompt_str, return_tensors="pt").to(model.device)