Update app.py

app.py

@@ -197,6 +197,34 @@ After calling a tool, I will provide you with the results to interpret for the u
         
         return tool_calls
     
+    def truncate_tool_result(self, result: Dict[str, Any], max_chars: int = 2000) -> Dict[str, Any]:
+        """Truncate tool results to prevent context overflow"""
+        if not isinstance(result, dict):
+            return result
+            
+        result_copy = result.copy()
+        result_str = json.dumps(result_copy, indent=2)
+        
+        if len(result_str) > max_chars:
+            # Try to truncate data arrays/lists first
+            for key, value in result_copy.items():
+                if isinstance(value, list) and len(value) > 3:
+                    result_copy[key] = value[:3] + [f"... ({len(value) - 3} more items truncated)"]
+                elif isinstance(value, str) and len(value) > 500:
+                    result_copy[key] = value[:500] + "... (truncated)"
+            
+            # If still too long, add truncation notice
+            result_str = json.dumps(result_copy, indent=2)
+            if len(result_str) > max_chars:
+                result_copy = {
+                    "success": result.get("success", False),
+                    "truncated": True,
+                    "message": f"Result truncated due to size. Original had {len(result_str)} characters.",
+                    "sample_data": str(result)[:1000] + "..." if len(str(result)) > 1000 else str(result)
+                }
+        
+        return result_copy
+
     def process_message(self, user_message: str) -> Tuple[str, str]:
         """Process user message and handle tool calls"""
         tool_info = ""
@@ -213,7 +241,7 @@ After calling a tool, I will provide you with the results to interpret for the u
                     model="gpt-3.5-turbo",
                     messages=messages,
                     temperature=0.7,
-                    max_tokens=1000
+                    max_tokens=800  # Reduced to leave more room for context
                 )
                 ai_response = response.choices[0].message.content
             else:
@@ -222,7 +250,7 @@ After calling a tool, I will provide you with the results to interpret for the u
                     model="gpt-3.5-turbo",
                     messages=messages,
                     temperature=0.7,
-                    max_tokens=1000
+                    max_tokens=800
                 )
                 ai_response = response.choices[0].message.content
             tool_calls = self.extract_tool_calls(ai_response)
@@ -243,9 +271,12 @@ After calling a tool, I will provide you with the results to interpret for the u
                         tool_call['arguments']
                     )
                     
+                    # Truncate large results to prevent context overflow
+                    truncated_result = self.truncate_tool_result(result)
+                    
                     tool_results.append({
                         'tool': tool_call['name'],
-                        'result': result
+                        'result': truncated_result
                     })
                     
                     if result.get('success'):
@@ -253,8 +284,9 @@ After calling a tool, I will provide you with the results to interpret for the u
                     else:
                         tool_info += f"❌ {tool_call['name']} failed: {result.get('error', 'Unknown error')}\n"
                 
+                # Create concise tool results summary
                 tool_results_text = "\n\n".join([
-                    f"Tool: {tr['tool']}\nResult: {json.dumps(tr['result'], indent=2)}"
+                    f"Tool: {tr['tool']}\nResult: {json.dumps(tr['result'], indent=2)[:1500]}{'...(truncated)' if len(json.dumps(tr['result'], indent=2)) > 1500 else ''}"
                     for tr in tool_results
                 ])
                 
@@ -269,7 +301,7 @@ After calling a tool, I will provide you with the results to interpret for the u
                         model="gpt-3.5-turbo",
                         messages=final_messages,
                         temperature=0.7,
-                        max_tokens=1000
+                        max_tokens=800  # Reduced max tokens
                     )
                     return final_response.choices[0].message.content, tool_info
                 else:
@@ -277,7 +309,7 @@ After calling a tool, I will provide you with the results to interpret for the u
                         model="gpt-3.5-turbo",
                         messages=final_messages,
                         temperature=0.7,
-                        max_tokens=1000
+                        max_tokens=800
                     )
                     return final_response.choices[0].message.content, tool_info
             else:
@@ -362,6 +394,11 @@ def chat_interface(message, history, openai_key, mcp_url):
     try:
         print(f"Calling process_message with: {message}")
         
+        # Limit conversation history to prevent context overflow
+        # Keep only the last 5 exchanges (10 messages total)
+        if len(history) > 10:
+            history = history[-10:]
+        
         # Make sure we call the synchronous method
         result = assistant.process_message(message)
         print(f"process_message returned: {type(result)} - {result}")