See the screenshot below:
Is there a way to prevent this? We handled it when parsing the API result but the additional messages still go toward the token count right?
We tried handling this through the Assistant instruction, but without luck.
Unfortunately, I am not able to share our Assistant instruction due to our NDA.