Maternal care behaviours in rodents are fundamental to early life development but quantifying them requires labour-intensive manual annotation of home cage videos. We present TailOR, a computer vision system that automatically tags and extracts mouse and rat behaviour from side-view recordings. TailOR uses the Segment Anything Model 2 (SAM2) to generate high-quality masks and derives geometric features such as centroid positions, displacements and distances to key objects. These features feed a hierarchical rule-based decision system that applies domain-specific thresholds to classify behaviours including in-nest, off-nest eating, and off-nest non-eating activities. We evaluate TailOR on maternal care videos from mice and rats, achieving frame-level accuracies exceeding 90% across multiple datasets without manual intervention. The system's modular design and reliance on simple geometric features allow it to generalise across species, offering a scalable alternative to manual scoring and accelerating neuroscience studies of maternal care.