diff --git a/src/services/chatbot.py b/src/services/chatbot.py
index ca9d733..d113080 100644
--- a/src/services/chatbot.py
+++ b/src/services/chatbot.py
@@ -1,6 +1,7 @@
 import os
 import json
 import math
+import asyncio
 from uuid import UUID
 from datetime import date, datetime, timedelta, UTC
 from fastapi import HTTPException
@@ -202,7 +203,14 @@ def chatbot_workflow(bot, now):
                              )
                          ] + trimmed_messages
 
-        response = await model.ainvoke(input=input_messages, **kwargs)
+        # KIA API returns choices=null in streaming chunks which breaks LangChain.
+        # Even with streaming=False, ainvoke() still uses _astream() internally in async path.
+        # asyncio.to_thread forces sync model.invoke() which calls the real non-streaming endpoint.
+        if is_kia:
+            print(f"[KiaAI] Calling model via asyncio.to_thread (non-streaming) | model='{bot.model}'")
+            response = await asyncio.to_thread(model.invoke, input_messages)
+        else:
+            response = await model.ainvoke(input=input_messages, **kwargs)
 
         if is_kia:
             print(f"[KiaAI] Response received | type={type(response).__name__}")