We propose DEAL, a continual low-rank fine-tuning framework that enables efficient and privacy-preserving adaptation of large language models.