Problem

Organizations with large video and image archives often cannot extract timely intelligence from them. Traditional computer-vision development can require months of engineering to collect and label data, train a model, and deploy it; for niche tasks, enough data may be impossible to obtain, while models can fail to generalize across different cameras, environments, and workflows. The economic pain is therefore a combination of high R&D cost, slow deployment, manual review, and poor coverage of previously unseen events or behaviors.

The clearest killer use case is operational and defense intelligence: finding previously unseen targets, activities, or behaviors in footage from autonomous vessels and other surveillance systems. OnDeck cites Singapore’s Ministry of Defence using it to improve threat detection and situational awareness across autonomous assets, while other reported applications include port monitoring, security cameras, offshore oil and gas, and robotics research.

Product / Service

OnDeck is an enterprise infrastructure layer for vision-language models. It lets organizations search and analyze essentially any footage to find objects, events, and qualitative behaviors without collecting labeled data or training a task-specific model. Its vision engine is designed to generalize across tasks, understand temporal sequences and interactions, and identify things that were not present in the original training set. The platform can generate human- or machine-readable reports and adapt to a customer’s workflow through agentic systems that recursively improve.

The delivery model is enterprise-oriented rather than a simple consumer application: OnDeck says it supports cloud, on-premises, private-cloud, and air-gapped deployments, with SOC 2 compliance for sensitive environments. The intended benefit is to move from footage to searchable intelligence quickly, preserve data sovereignty, avoid training on customer data, and replace months of custom computer-vision R&D with a deployable, customizable system.

Market

OnDeck competes in enterprise computer vision, video intelligence, and vision-language-model infrastructure, with a focus on B2B customers handling large volumes of unstructured footage. Its target users include defense agencies, robotics companies, universities, public safety organizations, conservation teams, and industrial operators. The closest named adjacent competitor in the evidence is TwelveLabs, whose platform also provides enterprise video search and analysis across vision, audio, and language; traditional computer-vision systems and internally built ML pipelines are additional alternatives, although OnDeck differentiates itself by emphasizing no labeled data and open-ended behavioral understanding.

The company is not pre-revenue. OnDeck’s YC launch says it analyzed thousands of hours across autonomous surface vessels, robotics, security, port monitoring, and offshore oil and gas, and reports six-figure revenue before its pivot. Creative Destruction Lab later reported that it had grown from zero to six-figure ARR since May 2025 and had a $160,000 pilot with the Singaporean military; OnDeck’s current site also says it is deployed with defense agencies, universities, robotics companies, and public safety organizations. It is a Y Combinator Summer 2025 company and remains an early-stage, active entrant in a rapidly developing market.