Update app.py

app.py

@@ -11,7 +11,15 @@ from langchain_core.messages import HumanMessage, SystemMessage
 # --- Configuration ---
 NEBIUS_API_KEY = os.getenv("NEBIUS_API_KEY") 
 NEBIUS_BASE_URL = "https://api.studio.nebius.ai/v1/"
-MODEL_NAME = "openai/gpt-oss-120b"
+
+# OPTION 1: Best Balance (Speed + Intelligence) -> RECOMMENDED
+MODEL_NAME = "Qwen/Qwen3-30B-A3B-Thinking-2507"
+
+# OPTION 2: Maximum Speed (Good for simple tasks)
+# MODEL_NAME = "Qwen/Qwen3-30B-A3B-Thinking-2507"
+
+# OPTION 3: Maximum Intelligence (Slowest)
+# MODEL_NAME = "openai/gpt-oss-120b"
 
 # --- System Prompt ---
 SYSTEM_PROMPT = """You are a 'Vibe Coding' Python Tutor. 
@@ -77,9 +85,9 @@ def parse_agent_response(full_text):
 
     return chat_content, videos, articles, quiz
 
-async def run_tutor_dashboard(user_message):
+async def run_tutor_dashboard_stream(user_message):
     """
-    Main function to run the agent loop.
+    Runs the agent loop and yields streaming text updates.
     """
     server_params = StdioServerParameters(
         command=sys.executable, 
@@ -96,22 +104,33 @@ async def run_tutor_dashboard(user_message):
                 api_key=NEBIUS_API_KEY,
                 base_url=NEBIUS_BASE_URL,
                 model=MODEL_NAME,
-                temperature=0.7
+                temperature=0.7,
+                streaming=True # Critical for speed
             )
 
             agent_executor = create_react_agent(llm, tools)
             
-            # Prepend the SystemMessage to ensure the agent follows instructions
             inputs = {
                 "messages": [
                     SystemMessage(content=SYSTEM_PROMPT),
                     HumanMessage(content=user_message)
                 ]
             }
-            response = await agent_executor.ainvoke(inputs)
-            final_text = response["messages"][-1].content
-            
-            return parse_agent_response(final_text)
+
+            # Stream the response token by token
+            final_text = ""
+            async for event in agent_executor.astream_events(inputs, version="v1"):
+                kind = event["event"]
+                
+                # Stream actual tokens from the LLM
+                if kind == "on_chat_model_stream":
+                    content = event["data"]["chunk"].content
+                    if content:
+                        final_text += content
+                        yield final_text
+
+            # Once done, yield the final text one last time to ensure completeness
+            yield final_text
 
 # --- Gradio Dashboard UI ---
 # Professional "Slate" theme
@@ -144,7 +163,7 @@ with gr.Blocks(title="AI Python Tutor", theme=theme, fill_height=True) as demo:
 
             chatbot = gr.Chatbot(
                 height=600, 
-                show_label=False, # Removed built-in label to match custom header
+                show_label=False, 
                 type="messages",
                 bubble_full_width=False,
                 show_copy_button=True,
@@ -211,17 +230,26 @@ with gr.Blocks(title="AI Python Tutor", theme=theme, fill_height=True) as demo:
         # Placeholder for AI
         history.append({"role": "assistant", "content": "Thinking..."}) 
         
-        # Yield placeholders to ALL output boxes (Side AND Bottom)
-        # Structure: history, msg, 3x Side Boxes, 3x Bottom Boxes
+        # Initial yield
         yield history, "", "", "", "", "", "", ""
         
-        # Run Agent
-        chat_text, video_text, article_text, quiz_text = await run_tutor_dashboard(user_message)
+        full_response = ""
         
-        # Update AI response
-        history[-1]["content"] = chat_text
+        # STREAMING LOOP
+        async for chunk in run_tutor_dashboard_stream(user_message):
+            full_response = chunk
+            
+            # Update chat history LIVE
+            history[-1]["content"] = full_response
+            
+            # Yield update just for chat, keeping others blank for now
+            yield history, "", "", "", "", "", "", ""
+
+        # Final Parse (Split into 4 components)
+        chat_text, video_text, article_text, quiz_text = parse_agent_response(full_response)
         
-        # Yield final content to ALL output boxes
+        # Update everything one last time
+        history[-1]["content"] = chat_text
         yield history, "", video_text, article_text, quiz_text, video_text, article_text, quiz_text
 
     # --- Focus Mode Logic ---
@@ -229,14 +257,9 @@ with gr.Blocks(title="AI Python Tutor", theme=theme, fill_height=True) as demo:
 
     def toggle_fullscreen(current_state):
         new_state = not current_state
-        # If new_state is True (Fullscreen): Hide side col, Show bottom row
-        # If False (Normal): Show side col, Hide bottom row
-        
         side_visible = not new_state
         bottom_visible = new_state
-        
         btn_text = "↩ Exit Focus" if new_state else "⛶ Focus Mode"
-        
         return new_state, gr.Column(visible=side_visible), gr.Row(visible=bottom_visible), btn_text
 
     fullscreen_btn.click(
@@ -245,7 +268,7 @@ with gr.Blocks(title="AI Python Tutor", theme=theme, fill_height=True) as demo:
         outputs=[is_fullscreen, right_col, bottom_dashboard, fullscreen_btn]
     )
 
-    # Actions - We must map outputs to BOTH side and bottom components
+    # Actions
     outputs_list = [
         chatbot, msg, 
         video_box_side, article_box_side, quiz_box_side,