Xhxag persona fixes (BasedHardware#1855)

backend/utils/llm.py

@@ -645,15 +645,14 @@ def answer_omi_question_stream(messages: List[Message], context: str, callbacks:
 
 
 def answer_persona_question_stream(app: App, messages: List[Message], callbacks: []) -> str:
-    prompt = f"""
-    {app.persona_prompt}
+    chat_messages = [{"role": "system", "content": app.persona_prompt}]
+    for msg in messages:
+        if msg.sender == "ai":
+            chat_messages.append({"role": "ai", "content": msg.text})
+        else:
+            chat_messages.append({"role": "user", "content": msg.text})
 
-    ---
-    CHAT MESSAGES:
-
-    {Message.get_messages_as_string(messages)}
-    """
-    return llm_medium_stream.invoke(prompt, {'callbacks':callbacks}).content
+    return llm_medium_stream.invoke(chat_messages, {'callbacks':callbacks}).content
 
 def _get_qa_rag_prompt(uid: str, question: str, context: str, plugin: Optional[Plugin] = None,
                        cited: Optional[bool] = False,
@@ -2131,6 +2130,6 @@ def generate_persona_intro_message(prompt: str, name: str):
         {"role": "system", "content": prompt},
         {"role": "user", "content": f"Generate a short, funny 5-8 word message that would make someone want to chat with you. Be casual and witty, but don't mention being AI or a clone. Just be {name}. The message should feel natural and make people curious to chat with you."}
     ]
-    
+
     response = llm_medium.invoke(messages)
-    return response.content.strip('"').strip()
+    return response.content.strip('"').strip()