diff --git a/.vscode/launch.json b/.vscode/launch.json
index b4b1d6b..7a37906 100644
--- a/.vscode/launch.json
+++ b/.vscode/launch.json
@@ -20,7 +20,7 @@
                 "PYTHONPATH": "${workspaceFolder}"
             },
             "args": [
-                "--debug", "--max_plan_iterations", "1", "--max_step_num", "3"
+                "--debug", "--max_plan_iterations", "1", "--max_step_num", "1"
             ]
         },
         {
diff --git a/src/graph/nodes.py b/src/graph/nodes.py
index d3397eb..30f83ba 100644
--- a/src/graph/nodes.py
+++ b/src/graph/nodes.py
@@ -199,6 +199,7 @@ def _execute_agent_step(
 ) -> Command[Literal["research_team"]]:
     """Helper function to execute a step using the specified agent."""
     current_plan = state.get("current_plan")
+    observations = state.get("observations", [])
 
     # Find the first unexecuted step
     for step in current_plan.steps:
@@ -244,7 +245,7 @@ def _execute_agent_step(
                     name=agent_name,
                 )
             ],
-            "observations": [response_content],
+            "observations": observations + [response_content],
         },
         goto="research_team",
     )
diff --git a/src/graph/types.py b/src/graph/types.py
index 128bc67..ccbbedf 100644
--- a/src/graph/types.py
+++ b/src/graph/types.py
@@ -14,7 +14,7 @@ class State(MessagesState):
 
     # Runtime Variables
     locale: str = "en-US"
-    observations: Annotated[list[str], operator.add] = []
+    observations: list[str] = []
     plan_iterations: int = 0
     current_plan: Plan | str = None
     final_report: str = ""
diff --git a/src/server/app.py b/src/server/app.py
index a2f4ccf..e7036f4 100644
--- a/src/server/app.py
+++ b/src/server/app.py
@@ -72,7 +72,14 @@ async def _astream_workflow_generator(
     auto_accepted_plan: bool,
     interrupt_feedback: str,
 ):
-    input_ = {"messages": messages, "auto_accepted_plan": auto_accepted_plan}
+    input_ = {
+        "messages": messages,
+        "plan_iterations": 0,
+        "final_report": "",
+        "current_plan": None,
+        "observations": [],
+        "auto_accepted_plan": auto_accepted_plan,
+    }
     if not auto_accepted_plan and interrupt_feedback:
         resume_msg = f"[{interrupt_feedback}]"
         # add the last message to the resume message