diff --git a/erniebot-agent/src/erniebot_agent/agents/agent.py b/erniebot-agent/src/erniebot_agent/agents/agent.py
index d04a9c96..ed02f714 100644
--- a/erniebot-agent/src/erniebot_agent/agents/agent.py
+++ b/erniebot-agent/src/erniebot_agent/agents/agent.py
@@ -361,9 +361,37 @@ async def _run_llm_stream(self, messages: List[Message], **opts: Any) -> AsyncIt
             )
         opts["system"] = self.system.content if self.system is not None else None
         opts["plugins"] = self._plugins
-        llm_ret = await self.llm.chat(messages, stream=True, functions=functions, **opts)
-        async for msg in llm_ret:
-            yield LLMResponse(message=msg)
+        # llm_ret = await self.llm.chat(messages, stream=True, functions=functions, **opts)
+        # async for msg in llm_ret:
+        #     yield LLMResponse(message=msg)
+        # print(self.llm.extra_params.get("enable_multi_step_tool_call"))
+        # 流式时，无法同时处理多个工具调用
+        # 所以只有关闭多步工具调用时才用流式
+        if self.llm.extra_data.get("multi_step_tool_call_close", True):
+            llm_ret = await self.llm.chat(messages, stream=True, functions=functions, **opts)
+            async for msg in llm_ret:
+                print("_run_llm_stream", msg)
+                yield LLMResponse(message=msg)
+        else:
+            llm_ret = await self.llm.chat(messages, stream=False, functions=functions, **opts)
+            class MyAsyncIterator:
+                def __init__(self, data):
+                    self.data = data
+                    self.index = 0
+                async def __anext__(self):
+                    if self.index < len(self.data):
+                        result = self.data[self.index]
+                        self.index += 1
+                        return result
+                    else:
+                        raise StopAsyncIteration
+                def __aiter__(self):
+                    return self
+
+            llm_ret = MyAsyncIterator([llm_ret])
+            async for msg in llm_ret:
+                print("_run_llm_stream", msg)
+                yield LLMResponse(message=msg)
 
     async def _run_tool(self, tool: BaseTool, tool_args: str) -> ToolResponse:
         parsed_tool_args = self._parse_tool_args(tool_args)