diff --git a/src/config.toml b/src/config.toml
index 669a31c..eae94ff 100644
--- a/src/config.toml
+++ b/src/config.toml
@@ -1,3 +1,5 @@
+[base]
+maxcount = 20
 [siloconflow]
 api_key = ""
 modules = ["Qwen/Qwen2.5-7B-Instruct", "Qwen/Qwen2.5-Coder-7B-Instruct"]
diff --git a/src/process.py b/src/process.py
index 51f7830..0250a1f 100644
--- a/src/process.py
+++ b/src/process.py
@@ -6,44 +6,48 @@ import time
 
 class SiloconFlowAPI:
     def __init__(self, api_key,modules,base_url):
-        self.client = OpenAI(api_key=api_key, base_url=base_url)
         self.api_key = api_key
         self.module = modules
-
-    def get_ai_message(self, message,i=0):
-        try:
-            response = self.client.chat.completions.create(  
-                model=self.module[i],  
-                messages=message,
-                temperature=0.7,  
-                max_tokens=4096
-            )
-            print("token usage {0}".format(response.usage.total_tokens))
-            return response.choices[0].message.content
-        except:
-            if i<len(self.module)-1:
-                i+=1
-                self.get_ai_message(message,i)
-            else:
-                print("server busy,waiting")
-                time.sleep(10)
-                i = 0
-                self.get_ai_message(message,i)
+        self.url = base_url
+        self.count = 0
+    def get_ai_message(self, message,maxcount,i=0):
+        with OpenAI(api_key=self.api_key, base_url=self.base_url) as client:
+            try:
+                response = client.chat.completions.create(  
+                    model=self.module[i],  
+                    messages=message,
+                    temperature=0.7,  
+                    max_tokens=4096
+                )
+                print("token usage {0}".format(response.usage.total_tokens))
+                return response.choices[0].message.content
+            except:
+                if i<len(self.module)-1:
+                    i+=1
+                    self.count = self.count+1
+                    return self.get_ai_message(message,i)
+                else:
+                    print("server busy,waiting")
+                    time.sleep(10)
+                    i = 0
+                    if(self.count>maxcount)
+                    return self.get_ai_message(message,i)
 
 
 
 class Ai_reply(BasePlugin):
     def after_save(self):
         config = self.config.get("siloconflow")
+        maxcount = self.config.get("base").get("maxcount")
         ai_server = SiloconFlowAPI(base_url= config.get("base_url"), api_key= config.get("api_key"), modules= config.get("modules"))
         if self.ctx.group is None:
             self.ctx.user.messages.append({'role': 'user', 
-            'content': f"当前用户为:{self.ctx.user.nickname}，其uid为:{self.ctx.user.user_id}，以上为背景信息，根据背景信息回复用户消息。用户消息:{self.ctx.raw_message}"})
-            message = ai_server.get_ai_message(self.ctx.user.messages)
+            'content': 
+            f"当前用户为:{self.ctx.user.nickname}，其uid为:{self.ctx.user.user_id}，以上为背景信息，根据背景信息回复用户消息。用户消息:{self.ctx.raw_message}"})
             typing_thread = threading.Thread(target=self.ctx.user.set_input_status, args=(1,))
             typing_thread.start()
             #获取ai返回值
-            message = ai_server.get_ai_message(self.ctx.user.messages)
+            message = ai_server.get_ai_message(self.ctx.user.messages,maxcount)
             #结束刷新状态
             self.ctx.user.signal = False
             #回收线程
@@ -57,7 +61,7 @@ class Ai_reply(BasePlugin):
                 self.ctx.group.messages.append({'role': 'user', 
                                                 'content': 
                                                         f"群聊名称为:{self.ctx.group.nickname},用户在群中的近十条消息为:{self.ctx.group.current_user.messages},用户名叫:{self.ctx.group.current_user.nickname},以上为背景信息，根据背景信息回复消息。消息为：{self.ctx.raw_message}"})
-                message = ai_server.get_ai_message(self.ctx.group.messages)
+                message = ai_server.get_ai_message(self.ctx.group.messages,maxcount)
            
                 self.ctx.group.send_message(message)
         return "ok"
\ No newline at end of file