dont send full traces to endpoint

app.py

@@ -83,7 +83,7 @@ def execute_jupyter_agent(
         )
     message_history.append({"role": "user", "content": user_input})
 
-    print("history:", message_history)
+    #print("history:", message_history)
 
     for notebook_html, notebook_data, messages in run_interactive_notebook(
         client, model, message_history, sbx, max_new_tokens=max_new_tokens

utils.py

@@ -81,6 +81,19 @@ def parse_exec_result_llm(execution, max_code_output=1000):
         output.append(truncate_if_needed(execution.error.traceback))
     return "\n".join(output)
 
+def clean_messages_for_api(messages):
+    """
+    Create a clean copy of messages without raw_execution fields for API calls.
+    This prevents 413 errors caused by large execution data.
+    """
+    cleaned_messages = []
+    for message in messages:
+        cleaned_message = message.copy()
+        if "raw_execution" in cleaned_message:
+            cleaned_message.pop("raw_execution")
+        cleaned_messages.append(cleaned_message)
+    return cleaned_messages
+
 
 def run_interactive_notebook(client, model, messages, sbx, max_new_tokens=512):
     notebook = JupyterNotebook(messages)
@@ -91,21 +104,17 @@ def run_interactive_notebook(client, model, messages, sbx, max_new_tokens=512):
     max_code_output = 1000
     turns = 0
     done = False
-    
-    print("SBX INFO", sbx.get_info())
 
     while not done and (turns <= MAX_TURNS):
         turns += 1
-        
         try:
             # Inference client call - might fail
             response = client.chat.completions.create(
-                messages=messages,
+                messages=clean_messages_for_api(messages),
                 model=model,
                 tools=TOOLS,
                 tool_choice="auto",
             )
-            
         except Exception as e:
             # Handle inference client errors
             notebook.add_error(f"Inference failed: {str(e)}")