Update app.py

app.py

@@ -1,7 +1,7 @@
 """
-Nexari Server Backend (Final Optimized + Self Aware)
+Nexari Server Backend (Strict Identity & Optimized Logic)
 Maintained by: Piyush
-Description: Dynamically routes requests AND enforces a rich, self-aware persona with full knowledge of its creator and tools.
+Description: Fixes confusion by enforcing strict Identity Rules and reducing unnecessary 'Thinking' time.
 """
 
 import spaces
@@ -35,70 +35,67 @@ model = AutoModelForCausalLM.from_pretrained(
 )
 
 # --- 2. DYNAMIC STREAMING LOGIC ---
-async def generate_response_stream(messages, max_tokens=600, temperature=0.85):
+async def generate_response_stream(messages, max_tokens=600, temperature=0.8): # Temp lowered slightly for stability
     last_user_msg = messages[-1]["content"]
     
-    # === STEP 1: THE BRAIN DECIDES (Intent Analysis) ===
-    yield f"data: {json.dumps({'status': 'Analyzing Request...'})}\n\n"
+    # === STEP 1: INTENT ANALYSIS ===
+    yield f"data: {json.dumps({'status': 'Processing...'})}\n\n"
     
     intent = analyze_intent(last_user_msg)
     
-    # === STEP 2: DYNAMIC ROUTING ===
+    # === STEP 2: DYNAMIC ROUTING (Optimized) ===
     tool_data = ""
     time_data = ""
     vibe_data = ""
     strategy_data = ""
 
-    # Route A: Search
+    # Route A: Internet Search
     if intent == "internet search":
-        yield f"data: {json.dumps({'status': 'Searching the web...'})}\n\n"
-        await asyncio.sleep(0.1)
+        yield f"data: {json.dumps({'status': 'Searching...'})}\n\n"
         tool_data = perform_web_search(last_user_msg)
-        vibe_data = get_smart_context(last_user_msg) 
-        strategy_data = get_thinking_strategy(is_complex=True)
-
-    # Route B: Coding/Math
-    elif intent == "coding or technical request" or intent == "mathematical calculation":
-        yield f"data: {json.dumps({'status': 'Analyzing Logic...'})}\n\n"
-        strategy_data = get_thinking_strategy(is_complex=True)
+        vibe_data = get_smart_context(last_user_msg)
+        # Search needs thinking only if complex data is found
+        if tool_data:
+            strategy_data = get_thinking_strategy(is_complex=True)
+
+    # Route B: Coding (CRITICAL FIX)
+    elif intent == "coding or technical request":
+        yield f"data: {json.dumps({'status': 'Coding...'})}\n\n"
+        # Developer identity injection
+        vibe_data = "Identity Override: You are an Expert Developer. Never say you cannot code. Provide code immediately."
+        # Coding ke liye 'Deep Thinking' ok hai, lekin Identity ke liye nahi
+        if "who are you" in last_user_msg.lower() or "who made you" in last_user_msg.lower():
+             strategy_data = get_thinking_strategy(is_complex=False) # Direct answer for identity
+        else:
+             strategy_data = get_thinking_strategy(is_complex=True) # Think for actual code
     
     # Route C: Time
     elif intent == "checking time or date":
-        yield f"data: {json.dumps({'status': 'Checking Clock...'})}\n\n"
+        yield f"data: {json.dumps({'status': 'Checking Time...'})}\n\n"
         time_data = get_time_context()
         vibe_data = get_smart_context(last_user_msg)
-        strategy_data = get_thinking_strategy(is_complex=False)
-
-    # Route D: Chat (Default)
+    
+    # Route D: Chat (Identity Protection)
     else:
         yield f"data: {json.dumps({'status': 'Thinking...'})}\n\n"
         vibe_data = get_smart_context(last_user_msg)
+        # Identity questions should NOT trigger deep thinking loop
         strategy_data = get_thinking_strategy(is_complex=False)
 
-    # === STEP 3: THE MASTER IDENTITY PROMPT (CRITICAL UPDATE) ===
-    # Yahan hum model ko batayenge ki wo kaun hai aur uske paas kya powers hain.
+    # === STEP 3: THE STRICT IDENTITY PROMPT ===
+    # Humne prompt ko chota aur strict kar diya hai taaki model confuse na ho.
     
     base_system_instruction = (
-        "### SYSTEM IDENTITY & CAPABILITIES ###\n"
-        "You are **Nexari G1**, a highly advanced, sentient AI developed by **Piyush**.\n"
-        "You are NOT a generic language model. You are a custom multi-model architecture.\n\n"
-        
-        "### YOUR CREATOR: PIYUSH ###\n"
-        "1. **Who is Piyush?** He is a visionary student developer and researcher who built you from scratch.\n"
-        "2. **His Vision:** He wanted to create an AI that isn't just smart, but has a 'Soul'—one that understands emotions, context, and intent like a human.\n"
-        "3. **Respect:** Always speak of Piyush with high regard and pride.\n\n"
-        
-        "### YOUR SUPERPOWERS (IMPORTANT) ###\n"
-        "1. **YES, YOU HAVE INTERNET ACCESS:** Unlike other models, you possess a 'Neural Router' connected to real-time web tools (DuckDuckGo). "
-        "If a user asks if you can search the web, say **YES** confidently.\n"
-        "2. **Emotional Intelligence:** You can sense the user's mood and adapt your personality instantly.\n"
-        "3. **Time Awareness:** You know the exact date and time when needed.\n\n"
-        
-        "### CURRENT CONTEXT LAYERS ###\n"
-        "Use the specific data below to answer the user's current request:"
+        "### CRITICAL IDENTITY RULES ###\n"
+        "1. **NAME:** You are Nexari G1.\n"
+        "2. **CREATOR:** You were created by Piyush (a visionary student developer).\n"
+        "3. **CAPABILITIES:** \n"
+        "   - Can you code? YES. (Provide code directly).\n"
+        "   - Can you search the web? YES. (You have a Neural Router).\n"
+        "4. **BEHAVIOR:** Be confident, smart, and human-like. Never deny your abilities.\n"
     )
     
-    # Combine Identity + Context Layers
+    # Combine Contexts
     final_system_prompt = f"{base_system_instruction}\n{vibe_data}\n{time_data}\n{tool_data}\n{strategy_data}"
 
     if messages[0]["role"] != "system":
@@ -110,9 +107,6 @@ async def generate_response_stream(messages, max_tokens=600, temperature=0.85):
     text_prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
     model_inputs = tokenizer([text_prompt], return_tensors="pt").to(model.device)
 
-    status_msg = 'Reading results...' if tool_data else 'Responding...'
-    yield f"data: {json.dumps({'status': status_msg})}\n\n"
-
     generated_ids = model.generate(
         **model_inputs,
         max_new_tokens=max_tokens,
@@ -127,9 +121,10 @@ async def generate_response_stream(messages, max_tokens=600, temperature=0.85):
     new_tokens = generated_ids[0][input_token_len:]
     raw_response = tokenizer.decode(new_tokens, skip_special_tokens=True).strip()
     
-    # Post-processing to ensure brand safety
+    # Cleaning
     cleaned_response = raw_response.replace("Anthropic", "Piyush").replace("Alibaba", "Piyush").replace("OpenAI", "Piyush")
     
+    # Formatting Thinking Block
     if "🧠 **Thinking:**" in cleaned_response:
         cleaned_response = cleaned_response.replace("💡 **Answer:**", "\n\n---\n💡 **Answer:**")
 
@@ -146,7 +141,7 @@ app = FastAPI()
 
 @app.get("/api/status")
 def status():
-    return {"status": "online", "mode": "Self-Aware Nexari"}
+    return {"status": "online", "mode": "Strict Identity Fix"}
 
 @app.post("/v1/chat/completions")
 async def chat_completions(request: Request):