Follow-up messages were starting a new agent every time, so each reply waited through process startup and a new session before the model was called.
Follow-up messages were starting a new agent every time, so each reply waited through process startup and a new session before the model was called.