This paper proposes a novel framework for Unsupervised Continual Anomaly Detection (UCAD), which enables continual learning in unsupervised anomaly detection and segmentation. The framework introduces a Continual Prompting Module (CPM) to enable continual learning in unsupervised anomaly detection and a Structure-based Contrastive Learning (SCL) module to extract more compact features across various tasks. The CPM learns a "key-prompt-knowledge" memory space to store auto-selected task queries, task adaptation prompts, and the 'normal' knowledge of different classes. The SCL module uses the Segment Anything Model (SAM) to improve prompt learning and anomaly segmentation results by leveraging the general segmentation ability of SAM. The proposed UCAD framework is evaluated on the MVTec AD and VisA datasets, demonstrating significant improvements in both anomaly detection and segmentation tasks compared to existing methods. The framework is able to handle continual learning scenarios where new tasks are added incrementally, and it outperforms previous state-of-the-art methods by 15.6% on detection and 26.6% on segmentation. The code is available at https://github.com/shirowalker/UCAD.This paper proposes a novel framework for Unsupervised Continual Anomaly Detection (UCAD), which enables continual learning in unsupervised anomaly detection and segmentation. The framework introduces a Continual Prompting Module (CPM) to enable continual learning in unsupervised anomaly detection and a Structure-based Contrastive Learning (SCL) module to extract more compact features across various tasks. The CPM learns a "key-prompt-knowledge" memory space to store auto-selected task queries, task adaptation prompts, and the 'normal' knowledge of different classes. The SCL module uses the Segment Anything Model (SAM) to improve prompt learning and anomaly segmentation results by leveraging the general segmentation ability of SAM. The proposed UCAD framework is evaluated on the MVTec AD and VisA datasets, demonstrating significant improvements in both anomaly detection and segmentation tasks compared to existing methods. The framework is able to handle continual learning scenarios where new tasks are added incrementally, and it outperforms previous state-of-the-art methods by 15.6% on detection and 26.6% on segmentation. The code is available at https://github.com/shirowalker/UCAD.