diff --git a/api/core/workflow/nodes/llm/node.py b/api/core/workflow/nodes/llm/node.py
index 9bfb402dc8..756cbf2cee 100644
--- a/api/core/workflow/nodes/llm/node.py
+++ b/api/core/workflow/nodes/llm/node.py
@@ -69,6 +69,12 @@ from core.workflow.nodes.event import (
 )
 from core.workflow.utils.variable_template_parser import VariableTemplateParser
 
+# --- New imports for file-usage persistence ---
+from sqlalchemy.orm import Session
+from sqlalchemy import select
+from extensions.ext_database import db
+from models.model import NodeFileUsage
+
 from . import llm_utils
 from .entities import (
     LLMNodeChatModelMessage,
@@ -175,6 +181,46 @@ class LLMNode(BaseNode[LLMNodeData]):
                 else []
             )
 
+            # ------------------------------
+            # Persist current-turn file usage so that subsequent turns/nodes can see them.
+            # A node is considered "using" sys.files when vision is enabled and a variable selector is configured.
+            # ------------------------------
+
+            uses_sys_files_for_node = self.node_data.vision.enabled and bool(
+                self.node_data.vision.configs.variable_selector
+            )
+
+            if uses_sys_files_for_node and files:
+                conv_var = variable_pool.get(["sys", SystemVariableKey.CONVERSATION_ID.value])
+                conversation_id_val = conv_var.value if isinstance(conv_var, StringSegment) else None
+
+                if conversation_id_val:
+                    try:
+                        with Session(db.engine) as session:
+                            for f in files:
+                                upload_id = getattr(f, "related_id", None)
+                                if not upload_id:
+                                    continue
+
+                                exists_stmt = select(NodeFileUsage.id).where(
+                                    NodeFileUsage.conversation_id == conversation_id_val,
+                                    NodeFileUsage.node_id == self.node_id,
+                                    NodeFileUsage.upload_file_id == upload_id,
+                                )
+                                if not session.scalar(exists_stmt):
+                                    session.add(
+                                        NodeFileUsage(
+                                            conversation_id=conversation_id_val,
+                                            node_id=self.node_id,
+                                            upload_file_id=upload_id,
+                                            message_id=None,
+                                        )
+                                    )
+                            session.commit()
+                    except Exception:
+                        # Non-critical – failure to persist should not break node execution
+                        pass
+
             if files:
                 node_inputs["#files#"] = [file.to_dict() for file in files]
 
@@ -219,6 +265,7 @@ class LLMNode(BaseNode[LLMNodeData]):
                 vision_detail=self.node_data.vision.configs.detail,
                 variable_pool=variable_pool,
                 jinja2_variables=self.node_data.prompt_config.jinja2_variables,
+                uses_sys_files_for_node=uses_sys_files_for_node,
             )
 
             # handle invoke result
@@ -570,6 +617,7 @@ class LLMNode(BaseNode[LLMNodeData]):
         vision_detail: ImagePromptMessageContent.DETAIL,
         variable_pool: VariablePool,
         jinja2_variables: Sequence[VariableSelector],
+        uses_sys_files_for_node: bool,
     ) -> tuple[Sequence[PromptMessage], Optional[Sequence[str]]]:
         prompt_messages: list[PromptMessage] = []
 
@@ -590,6 +638,7 @@ class LLMNode(BaseNode[LLMNodeData]):
                 memory=memory,
                 memory_config=memory_config,
                 model_config=model_config,
+                allowed_node_id=None if uses_sys_files_for_node else "",
             )
             # Extend prompt_messages with memory messages
             prompt_messages.extend(memory_messages)
@@ -1075,6 +1124,7 @@ def _handle_memory_chat_mode(
     memory: TokenBufferMemory | None,
     memory_config: MemoryConfig | None,
     model_config: ModelConfigWithCredentialsEntity,
+    allowed_node_id: str | None,
 ) -> Sequence[PromptMessage]:
     memory_messages: Sequence[PromptMessage] = []
     # Get messages from memory for chat model
@@ -1083,6 +1133,7 @@ def _handle_memory_chat_mode(
         memory_messages = memory.get_history_prompt_messages(
             max_token_limit=rest_tokens,
             message_limit=memory_config.window.size if memory_config.window.enabled else None,
+            allowed_node_id=allowed_node_id,
         )
     return memory_messages