diff --git a/backend/open_webui/utils/chat.py b/backend/open_webui/utils/chat.py
index 5d5f6f60c4..c97985dde4 100644
--- a/backend/open_webui/utils/chat.py
+++ b/backend/open_webui/utils/chat.py
@@ -33,7 +33,7 @@ from open_webui.utils.filter import (
 )
 from open_webui.utils.json_codec import JSONCodec
 from open_webui.utils.models import check_model_access, get_all_models
-from open_webui.utils.payload import convert_payload_openai_to_ollama
+from open_webui.utils.payload import apply_system_prompt_to_body, convert_payload_openai_to_ollama
 from open_webui.utils.response import (
     convert_response_ollama_to_openai,
     convert_streaming_response_ollama_to_openai,
@@ -281,6 +281,14 @@ async def generate_chat_completion(
             # Below does not require bypass_filter because this is the only route the uses this function and it is already bypassing the filter
             return await generate_function_chat_completion(request, form_data, user=user, models=models)
         if model.get('owned_by') == 'ollama':
+            # Apply before Ollama conversion so tool follow-ups keep the model system prompt
+            if not bypass_system_prompt:
+                model_info = await Models.get_model_by_id(form_data['model'])
+                if model_info:
+                    system = model_info.params.model_dump().get('system')
+                    form_data = await apply_system_prompt_to_body(system, form_data, metadata, user)
+                request.state.bypass_system_prompt = True
+
             # Using /ollama/api/chat endpoint
             form_data = convert_payload_openai_to_ollama(form_data)
             response = await generate_ollama_chat_completion(