Automated lesion segmentation and tracking in longitudinal medical imaging remain critical challenges in oncology, particularly as cancer incidence and imaging volumes continue to rise. Current promptable segmentation models are predominantly designed for 2D or single-timepoint 3D data, neglecting the temporal dimension essential for disease monitoring. We present LesionLocator, the first end-to-end framework unifying zero-shot 3D lesion segmentation and 4D tracking across longitudinal medical images, originally published in the Proceedings of CVPR 2025 [1]. Our model leverages extensive pretraining and introduces a novel synthetic longitudinal data generation technique to address the scarcity of multi-timepoint datasets. LesionLocator achieves human-level performance in universal lesion segmentation, outperforming existing promptable models by nearly 10 Dice points across diverse tumor types. Our autoregressive mask propagation achieves 86% retrieval accuracy with 79% Dice for longitudinal tracking.We provide the first open-access solution for promptable lesion tracking, releasing both synthetic 4D dataset and model weights.