fixed token streaming

chatxbt-assistant.py

@@ -94,14 +94,7 @@ async def start():
 
     # Initialize the assistant
     cxbt_assistant = Assistant(
-        llm=OpenAIChat(
-            api_key=os.getenv("PORTKEY_API_KEY"),
-            base_url=os.getenv("PORTKEY_LOCAL_GATEWAY_URL") or PORTKEY_GATEWAY_URL,
-            default_headers=createHeaders(
-                api_key=os.getenv("PORTKEY_API_KEY") or None,
-                config=portkey_config
-            ),
-        ),
+        llm=OpenAIChat(),
         tools=[
             DuckDuckGo(),
             CryptoDataTools(),
@@ -130,16 +123,21 @@ async def start():
 @cl.on_message
 async def main(message: cl.Message):
 
+    msg = cl.Message(content="")
+    await msg.send()
+
     # Retrieve the assistant from the user session
     agent = cl.user_session.get("agent")
 
     # Process the user message using the assistant
-    response = ""
     for delta in agent.run(message.content, stream=True):
-        response += delta
-    
-    # Send the response back to the user
-    await cl.Message(content=response).send()
+        for part in delta:
+            if token := part or "":
+                # Send the response back to the user
+                print(token)
+                await msg.stream_token(token)
+
+    await msg.update()
 
 # Run the Chainlit application
 if __name__ == "__main__":