Converting outputs from external tools into formats that the model can understand and reason over, like structured text or annotated images.