fix: use custom cache

app.py

@@ -65,11 +65,14 @@ def get_model_response(messages):
         print(messages)
         return None
 
+conversation_cache = {}
+
 def respond(message, history, system_message):
-    if not hasattr(history, 'conversation'):
-        history.conversation = [{"role": "system", "content": system_message}]
+    history_id = id(history)
+    if history_id not in conversation_cache:
+        conversation_cache[history_id] = [{"role": "system", "content": system_message}]
 
-    history.conversation.append({"role": "user", "content": message})
+    conversation_cache[history_id].append({"role": "user", "content": message})
 
     available_functions = {
         "evaluate_math_expression": evaluate_math_expression,
@@ -77,9 +80,9 @@ def respond(message, history, system_message):
 
     function_calls = []
     while True:
-        response = get_model_response(history.conversation)
+        response = get_model_response(conversation_cache[history_id])
         response_message = response.choices[0].message
-        history.conversation.append(response_message)
+        conversation_cache[history_id].append(response_message)
 
         if not response_message.tool_calls and response_message.content is not None:
             break
@@ -92,7 +95,7 @@ def respond(message, history, system_message):
                 }
                 function_calls.append(function_call)
                 function_response = call_function(tool_call, available_functions)
-                history.conversation.append(function_response)
+                conversation_cache[history_id].append(function_response)
                 function_calls.append({
                     "name": function_response["name"],
                     "result": json.loads(function_response["content"])