Monitoring highly mobile pelagic predators remains a major challenge for ocean ecology and fisheries science. We present Piscivore, an autonomous predator observation system integrated with a Long Range Autonomous Underwater Vehicle (LRAUV) that enables persistent, non-extractive monitoring of marine predators while simultaneously collecting high-resolution oceanographic data. Piscivore combines forward- and aft-facing cameras, a passive visual attractor, and synchronized environmental sensors to detect, identify, and contextualize predator encounters in situ. Surveys have recorded a diverse assemblage of pelagic predators, including fishes, sharks, seabirds, and marine mammals, while concurrently imaging forage communities and measuring the physical and biogeochemical environment. Here, we describe the system design, machine-learning-enabled detection workflow, and case studies illustrating how autonomous observations reveal links among predator occurrence, prey distributions, and fine-scale oceanographic features. We evaluate the strengths and limitations of autonomous visual monitoring, highlighting applications for adaptive monitoring of dynamic marine ecosystems including fisheries and offshore infrastructure. Piscivore demonstrates how autonomous vehicles can unite biological and environmental observations, providing a scalable framework for monitoring marine predators and the ecosystem processes they shape.