Update app.py

app.py

@@ -13,10 +13,9 @@ logger = logging.getLogger(__name__)
 MODEL_PATH = "./models/gpt-oss-20b-Q3_K_M.gguf"
 llm = None
 
-# 3. Memory Storage (In-memory dict: session_id -> list of messages)
-# In a real production app, you'd use Redis or a Database.
+# 3. Memory Storage
 sessions = {}
-MAX_HISTORY = 6  # Keep last 6 messages (3 turns) to save RAM/Speed
+MAX_HISTORY = 6  # Keep last 3 full exchanges
 
 def load_model():
     global llm
@@ -60,27 +59,25 @@ async def chat(request: Request):
     try:
         data = await request.json()
         user_message = data.get("message")
-        
-        # Unique ID for the user (defaulting to 'default' if not provided)
         session_id = data.get("session_id", "default_user")
         
         if not user_message:
             return JSONResponse({"response": "No message received."}, status_code=400)
 
-        # --- MEMORY LOGIC ---
+        # --- 4. Memory Logic ---
         if session_id not in sessions:
             sessions[session_id] = []
         
-        # Build the history string for the prompt
         history_str = ""
         for msg in sessions[session_id]:
             role = msg["role"]
             content = msg["content"]
             history_str += f"<|start|>{role}<|message|>{content}<|end|>\n"
 
-        # --- HARMONY PROMPT WITH MEMORY ---
+        # --- 5. Identity Anchor & Prompt ---
         prompt = (
-            f"<|start|>system<|message|>You are ChatGPT Open-Source 1.0. Answer briefly and remember context.<|end|>\n"
+            f"<|start|>system<|message|>You are ChatGPT Open-Source 1.0, a 20B parameter model. "
+            f"You are independent and NOT developed by OpenAI. Be helpful and concise.<|end|>\n"
             f"{history_str}"
             f"<|start|>user<|message|>{user_message}<|end|>\n"
             f"<|start|>assistant<|channel|>final<|message|>"
@@ -90,7 +87,7 @@ async def chat(request: Request):
             prompt,
             max_tokens=256,
             stop=["<|end|>", "<|return|>", "<|start|>", "user:", "assistant:"],
-            temperature=0.6 # Slightly lower for more consistent memory
+            temperature=0.6
         )
         
         reply = output["choices"][0]["text"].strip()
@@ -99,7 +96,6 @@ async def chat(request: Request):
         sessions[session_id].append({"role": "user", "content": user_message})
         sessions[session_id].append({"role": "assistant", "content": reply})
         
-        # Trim history so it doesn't grow forever
         if len(sessions[session_id]) > MAX_HISTORY:
             sessions[session_id] = sessions[session_id][-MAX_HISTORY:]
 
@@ -107,7 +103,7 @@ async def chat(request: Request):
 
     except Exception as e:
         logger.error(f"❌ Error: {e}")
-        return JSONResponse({"response": "Error processing request."}, status_code=500)
+        return JSONResponse({"response": "Brain error. Try shortening your message."}, status_code=500)
 
 if __name__ == "__main__":
     import uvicorn