I am wondering: Shouldn’t there be an api call that allows us to remove a message?
The assistant gave a response that was caught by some eval function and was found to be a bad response. Best thing would be: Remove the message, and restart the run.
Currently you need to try to explain the assistant what was wrong with this previous message, which is not a very good way of handling it. Also it keeps this “bad response” in the thread history.
Actually the only proper way of handling this currently might even be to retrieve all messages of the thread (but the last), create a new thread and try again.