diff --git a/ra_aid/agent_utils.py b/ra_aid/agent_utils.py
index 0f67623..f467e5f 100644
--- a/ra_aid/agent_utils.py
+++ b/ra_aid/agent_utils.py
@@ -874,6 +874,7 @@ def run_agent_with_retry(
     _max_test_retries = config.get("max_test_cmd_retries", DEFAULT_MAX_TEST_CMD_RETRIES)
     auto_test = config.get("auto_test", False)
     original_prompt = prompt
+    agent_type = get_agent_type(agent)
     msg_list = [HumanMessage(content=prompt)]
 
     with InterruptibleSection():
@@ -898,12 +899,19 @@ def run_agent_with_retry(
                     return "Agent run completed successfully"
                 except ToolExecutionError as e:
                     print("except ToolExecutionError in AGENT UTILS")
-                    if not isinstance(agent, CiaynAgent):
-                        logger.debug("AGENT UTILS ToolExecutionError called!")
-                        fallback_response = fallback_handler.handle_failure(e, agent)
-                        if fallback_response:
+                    logger.debug("AGENT UTILS ToolExecutionError called!")
+                    fallback_response = fallback_handler.handle_failure(e, agent)
+                    if fallback_response:
+                        if agent_type == "React":
                             msg_list.extend(fallback_response)
-                            continue
+                        else:
+                            agent.chat_history.extend(fallback_response)
+                            agent.chat_history.append(
+                                HumanMessage(
+                                    content="Fallback tool handler successfully ran your tool call. See last message for result."
+                                )
+                            )
+                        continue
                 except (KeyboardInterrupt, AgentInterrupt):
                     raise
                 except (
diff --git a/ra_aid/agents/ciayn_agent.py b/ra_aid/agents/ciayn_agent.py
index 59e4e2d..211877c 100644
--- a/ra_aid/agents/ciayn_agent.py
+++ b/ra_aid/agents/ciayn_agent.py
@@ -98,6 +98,7 @@ class CiaynAgent:
         self.tools = tools
         self.max_history_messages = max_history_messages
         self.max_tokens = max_tokens
+        self.chat_history = []
         self.available_functions = []
         for t in tools:
             self.available_functions.append(get_function_info(t.func))
@@ -105,6 +106,9 @@ class CiaynAgent:
         self.tool_failure_current_provider = None
         self.tool_failure_current_model = None
         self.fallback_handler = FallbackHandler(config, tools)
+        self.sys_message = SystemMessage(
+            "Execute efficiently yet completely as a fully autonomous agent."
+        )
 
     def _build_prompt(self, last_result: Optional[str] = None) -> str:
         """Build the prompt for the agent including available tools and context."""
@@ -348,51 +352,47 @@ Output **ONLY THE CODE** and **NO MARKDOWN BACKTICKS**"""
     ) -> Generator[Dict[str, Any], None, None]:
         """Stream agent responses in a format compatible with print_agent_output."""
         initial_messages = messages_dict.get("messages", [])
-        chat_history = []
+        # self.chat_history = []
         last_result = None
         first_iteration = True
 
         while True:
             base_prompt = self._build_prompt(None if first_iteration else last_result)
-            chat_history.append(HumanMessage(content=base_prompt))
+            self.chat_history.append(HumanMessage(content=base_prompt))
 
-            full_history = self._trim_chat_history(initial_messages, chat_history)
-            response = self.model.invoke(
-                [
-                    SystemMessage(
-                        "Execute efficiently yet completely as a fully autonomous agent."
-                    )
-                ]
-                + full_history
-            )
+            full_history = self._trim_chat_history(initial_messages, self.chat_history)
+            response = self.model.invoke([self.sys_message] + full_history)
 
             try:
                 logger.debug(f"Code generated by agent: {response.content}")
                 last_result = self._execute_tool(response)
-                chat_history.append(response)
+                self.chat_history.append(response)
                 first_iteration = False
                 yield {}
 
             except ToolExecutionError as e:
-                fallback_response = self.fallback_handler.handle_failure(e, self)
-                print(f"fallback_response={fallback_response}")
-                if fallback_response:
-                    hm = HumanMessage(
-                        content="The fallback handler has fixed your tool call results are in the last System message."
-                    )
-                    chat_history.extend(fallback_response)
-                    chat_history.append(hm)
-                    logger.debug("Appended fallback response to chat history.")
-                    yield {}
-                else:
-                    yield self._create_error_chunk(str(e))
-                    # yield {"messages": [fallback_response[-1]]}
-
-                # chat_history.append(
+                # self.chat_history.append(
                 #     HumanMessage(
                 #         content=f"Your tool call caused an error: {e}\n\nPlease correct your tool call and try again."
                 #     )
                 # )
+                raise e
+                # yield self._create_error_chunk(str(e))
+                yield {}
+
+                # fallback_response = self.fallback_handler.handle_failure(e, self)
+                # print(f"fallback_response={fallback_response}")
+                # if fallback_response:
+                #     hm = HumanMessage(
+                #         content="The fallback handler has fixed your tool call results are in the last System message."
+                #     )
+                #     self.chat_history.extend(fallback_response)
+                #     self.chat_history.append(hm)
+                #     logger.debug("Appended fallback response to chat history.")
+                #     yield {}
+                # else:
+                #     yield self._create_error_chunk(str(e))
+                # yield {"messages": [fallback_response[-1]]}
 
 
 def _extract_tool_call(code: str, functions_list: str) -> str: