OnDeck AI is a vision infrastructure platform that uses Vision Language Models to let enterprises analyze video footage without training custom models, instantly finding any object, behavior, or event in any footage with no training data. The platform deploys in minutes rather than months, generalizes across diverse camera setups and environments, and understands qualitative behavior and event sequences. Applications span autonomous vessels, robotics research, security, port monitoring, fisheries electronic monitoring, and offshore energy operations. Founded in 2025 in Vancouver, OnDeck AI was part of YC's Summer 2025 batch and published a NeurIPS workshop paper showing its VLM methods beat traditional computer vision.