diff --git a/backend/open_webui/utils/middleware.py b/backend/open_webui/utils/middleware.py
index 64adaef668..6a3e4b6f5b 100644
--- a/backend/open_webui/utils/middleware.py
+++ b/backend/open_webui/utils/middleware.py
@@ -2504,6 +2504,10 @@ async def process_chat_payload(request, form_data, user, metadata, model):
         except Exception:
             log.exception('Context compaction failed; continuing with full chat history')
 
+        metadata['context_start_message_id'] = next(
+            (message.get('id') for message in form_data.get('messages', []) if message.get('role') != 'system'), None
+        )
+
     # Process messages with OR-aligned output items for clean LLM messages
     for message in form_data.get('messages', []):
         output = message.get('output')
@@ -3523,6 +3527,11 @@ async def drain_approved_tool_calls(request, form_data, user, model, metadata) -
             assistant_message = await Chats.get_message_by_id_and_message_id(chat_id, message_id)
             if assistant_message:
                 db_messages.append({k: v for k, v in assistant_message.items() if k in MESSAGE_REPLAY_KEYS})
+            context_start_message_id = metadata.get('context_start_message_id')
+            start_index = next(
+                (index for index, message in enumerate(db_messages) if message.get('id') == context_start_message_id), 0
+            )
+            db_messages = db_messages[start_index:]
             for message in db_messages:
                 output = message.get('output')
                 # reasoning_details can be model/provider-bound, so only replay them
@@ -3534,8 +3543,9 @@ async def drain_approved_tool_calls(request, form_data, user, model, metadata) -
                 ):
                     message['output'] = strip_reasoning_details(output)
 
+            system_message = get_system_message(form_data.get('messages', []))
             form_data['messages'] = process_messages_with_output(
-                db_messages,
+                [system_message, *db_messages] if system_message else db_messages,
                 reasoning_format=get_reasoning_format(model),
                 include_file_context=metadata.get('include_file_context', False),
             )
diff --git a/backend/open_webui/utils/tool_approval.py b/backend/open_webui/utils/tool_approval.py
index e2c33e9a74..7d5376bb01 100644
--- a/backend/open_webui/utils/tool_approval.py
+++ b/backend/open_webui/utils/tool_approval.py
@@ -7,6 +7,8 @@ from sqlalchemy.ext.asyncio import AsyncSession
 from open_webui.constants import ERROR_MESSAGES
 from open_webui.env import ENABLE_ADMIN_CHAT_ACCESS
 from open_webui.models.chats import Chats
+from open_webui.models.config import Config
+from open_webui.models.users import Users
 from open_webui.socket.main import get_event_emitter
 from open_webui.utils.json_codec import JSONCodec
 
@@ -161,8 +163,17 @@ async def build_tool_approval_resume_payload(chat_id: str, message_id: str, chat
         raise HTTPException(status_code=status.HTTP_409_CONFLICT, detail='Tool call message model is missing.')
 
     messages = []
-    if params.get('system'):
-        messages.append({'role': 'system', 'content': params.get('system')})
+    system_prompt = params.get('system')
+    if not system_prompt:
+        # Mirror the chat UI's system prompt fallback
+        user = await Users.get_user_by_id(chat.user_id)
+        ui_settings = (user.settings.ui if user and user.settings else None) or {}
+        system_prompt = ui_settings.get('system')
+        if system_prompt is None:
+            default_interface_settings = await Config.get('ui.default_interface_settings') or {}
+            system_prompt = default_interface_settings.get('system')
+    if system_prompt:
+        messages.append({'role': 'system', 'content': system_prompt})
 
     return {
         'stream': params.get('stream_response', True),