added jenkins

backend/app/chat/service.py

@@ -292,10 +292,14 @@ class ChatService:
             reasoning_details: list[Any] | None = None
             finish_reason = ""
 
+            # После tool-раунда стримим вживую; до tools — буфер (иначе текст «переписывает» notice).
+            stream_live = tools_executed > 0
+
             async for event in self.llm.stream_chat(messages, tools=TOOL_DEFINITIONS):
                 if event["type"] == "content":
                     content_parts.append(event["content"])
-                    yield self._sse("token", {"content": event["content"]})
+                    if stream_live:
+                        yield self._sse("token", {"content": event["content"]})
                 elif event["type"] == "reasoning":
                     reasoning = event.get("reasoning", "") or reasoning
                     if event.get("reasoning_details"):
@@ -390,8 +394,12 @@ class ChatService:
 
                 continue
 
+            if content_parts and not stream_live:
+                for part in content_parts:
+                    yield self._sse("token", {"content": part})
+
             final_content = "".join(content_parts).strip()
-            if not final_content and streamed_reply_parts:
+            if not final_content and streamed_reply_parts and tools_executed == 0:
                 final_content = "".join(streamed_reply_parts).strip()
             if not final_content and reasoning:
                 final_content = reasoning.strip()