diff --git a/platform/backend/app/core/nvidia_client.py b/platform/backend/app/core/nvidia_client.py
index 5fdf561..aa50666 100644
--- a/platform/backend/app/core/nvidia_client.py
+++ b/platform/backend/app/core/nvidia_client.py
@@ -79,15 +79,19 @@ def _get_db_defaults(provider: Optional[str] = None) -> dict:
         cfg = db.query(LLMConfig).filter(LLMConfig.provider == p).order_by(LLMConfig.is_active.desc()).first()
         db.close()
         if cfg:
-            return {"temperature": cfg.temperature, "max_tokens": cfg.max_tokens}
+            return {
+                "temperature": cfg.temperature,
+                "max_tokens": cfg.max_tokens,
+                "system_prompt": cfg.system_prompt or "你是一个专业的内容创作助手。",
+            }
     except Exception:
         pass
-    return {"temperature": 0.20, "max_tokens": 2048}
+    return {"temperature": 0.20, "max_tokens": 2048, "system_prompt": "你是一个专业的内容创作助手。"}
 
 def call_llm(
     prompt: str,
     model: Optional[str] = None,
-    system_prompt: str = "你是一个专业的内容创作助手。",
+    system_prompt: Optional[str] = None,
     temperature: Optional[float] = None,
     max_tokens: Optional[int] = None,
     top_p: float = 0.70,
@@ -100,6 +104,7 @@ def call_llm(
     defaults = _get_db_defaults(provider)
     temperature = temperature if temperature is not None else defaults["temperature"]
     max_tokens = max_tokens if max_tokens is not None else defaults["max_tokens"]
+    system_prompt = system_prompt if system_prompt is not None else defaults["system_prompt"]
     cfg = _get_provider_config(provider)
     endpoint = f"{cfg['base_url'].rstrip('/')}/chat/completions"
     headers = {