diff --git a/api/core/repositories/sqlalchemy_workflow_node_execution_repository.py b/api/core/repositories/sqlalchemy_workflow_node_execution_repository.py
index 79b0c702e0..0a3189f398 100644
--- a/api/core/repositories/sqlalchemy_workflow_node_execution_repository.py
+++ b/api/core/repositories/sqlalchemy_workflow_node_execution_repository.py
@@ -480,9 +480,10 @@ class SQLAlchemyWorkflowNodeExecutionRepository(WorkflowNodeExecutionRepository)
         if reasoning_content:
             # reasoning_content could be a string or already a list
             if isinstance(reasoning_content, str):
-                reasoning_list = [reasoning_content] if reasoning_content else []
+                reasoning_list = [reasoning_content] if reasoning_content.strip() else []
             elif isinstance(reasoning_content, list):
-                reasoning_list = reasoning_content
+                # Filter out empty or whitespace-only strings
+                reasoning_list = [r.strip() for r in reasoning_content if isinstance(r, str) and r.strip()]
 
         # Extract tool_calls from metadata.agent_log
         tool_calls_list: list[dict] = []
@@ -491,16 +492,25 @@ class SQLAlchemyWorkflowNodeExecutionRepository(WorkflowNodeExecutionRepository)
             for log in agent_log:
                 # Each log entry has label, data, status, etc.
                 log_data = log.data if hasattr(log, "data") else log.get("data", {})
-                if log_data.get("tool_name"):
+                tool_name = log_data.get("tool_name")
+                # Only include tool calls with valid tool_name
+                if tool_name and str(tool_name).strip():
                     tool_calls_list.append(
                         {
                             "id": log_data.get("tool_call_id", ""),
-                            "name": log_data.get("tool_name", ""),
+                            "name": tool_name,
                             "arguments": json.dumps(log_data.get("tool_args", {})),
                             "result": str(log_data.get("output", "")),
                         }
                     )
 
+        # Only save if there's meaningful generation detail (reasoning or tool calls)
+        has_valid_reasoning = bool(reasoning_list)
+        has_valid_tool_calls = bool(tool_calls_list)
+
+        if not has_valid_reasoning and not has_valid_tool_calls:
+            return
+
         # Build sequence based on content, reasoning, and tool_calls
         sequence: list[dict] = []
         text = outputs.get("text", "")
@@ -514,10 +524,6 @@ class SQLAlchemyWorkflowNodeExecutionRepository(WorkflowNodeExecutionRepository)
         for i in range(len(tool_calls_list)):
             sequence.append({"type": "tool_call", "index": i})
 
-        # Only save if there's meaningful generation detail
-        if not reasoning_list and not tool_calls_list:
-            return
-
         # Check if generation detail already exists for this node execution
         existing = (
             session.query(LLMGenerationDetail)
diff --git a/api/core/workflow/nodes/llm/node.py b/api/core/workflow/nodes/llm/node.py
index e14dfebe64..4acbe25ae3 100644
--- a/api/core/workflow/nodes/llm/node.py
+++ b/api/core/workflow/nodes/llm/node.py
@@ -1680,9 +1680,6 @@ class LLMNode(Node[LLMNodeData]):
                         "content": text,
                     },
                 },
-                metadata={
-                    WorkflowNodeExecutionMetadataKey.LLM_CONTENT_SEQUENCE: [],
-                },
                 inputs={
                     **node_inputs,
                     "tools": [