About
Praxis AI captures egocentric video, 3D scans, and multimodal data from real industrial and residential environments. It supplies this physical-world training data to frontier AI labs and humanoid robotics companies, differentiating itself through embedded access to 60,000 workers across five continents and more than 150 environment types.
Market
Praxis competes in physical or embodied AI data infrastructure, specifically the capture of real-world demonstrations and environmental data needed to train robots and other physical-AI systems. It positions itself as a source of egocentric video, 3D scans, and multimodal recordings from industrial and residential environments for frontier labs and humanoid companies. Its differentiation is proprietary access to operational environments and workers at scale, combined with dedicated capture hardware and live reconstruction rather than relying only on generic data labeling.
Praxis primarily targets frontier AI labs, humanoid-robot companies, and other robotics developers that need real-world physical-AI training data. On the data-supply side, it works within publicly listed and unicorn-scale industrial and residential enterprises, reaching their workers and operating environments.
At a Glance
Problem
Physical-AI and humanoid-robot developers are bottlenecked by the scarcity of real-world training data. Models need egocentric video, 3D scans, and multimodal observations of people performing work in varied industrial and residential environments, but frontier labs and robotics companies cannot easily source that data at the required scale. The economic pain is the cost and delay of obtaining representative demonstrations and environment data; without it, robot-learning programs remain limited by synthetic or narrow datasets. The clearest use case is capturing how workers actually perform physical tasks so those demonstrations can train embodied-AI systems.
Product / Service
Praxis AI operates as a physical-world data vendor. It embeds its data-capture operation within large companies and uses workers and real environments to collect egocentric video, 3D scans, and other multimodal data, effectively turning participating businesses into sources of training data. Its value proposition is to give frontier AI labs and humanoid-robot companies access to authentic task demonstrations and environment coverage that they would struggle to gather directly, accelerating the development of more capable physical-AI systems.
Market
Praxis competes in the emerging embodied-AI and physical-AI training-data market, at the intersection of robotics data collection, reinforcement learning, and hard-tech infrastructure. Appen is a clear comparable because it markets large-scale egocentric video, environmental scene capture, and interaction data for world models and embodied-AI agents. Other adjacent physical-AI data platforms identified in the research include Encord, Scale AI, Labelbox, V7, Foxglove, and Rerun, although the evidence does not establish that each is a direct substitute for Praxis's embedded field-capture model.
The company appears to be very early but not merely an unvalidated idea. Y Combinator lists Praxis AI as an active Summer 2026 hard-tech company founded in 2026, with three employees in San Francisco, while its company description claims reach across 60,000 workers, five continents, and more than 150 environment types through embedded relationships with publicly listed and unicorn-scale conglomerates. The available evidence does not disclose revenue, funding, or named customers, so the strongest defensible assessment is early commercial traction or access rather than proven revenue scale.
Founders & Leadership
Funding History
Y Combinator
Recent News
A Y Combinator S26 batch directory entry lists Praxis (praxisrobotics.io) and describes its thesis: capturing how real work is done for the machines that will inherit it.
Y Combinator’s company profile describes Praxis AI’s focus on capturing the physical-world data needed by physical AI, including egocentric video, 3D scans, and multimodal capture from industrial and residential settings.
Active Roles
1Business Model
Praxis AI appears to operate as a business-to-business physical-AI data supplier, generating revenue by providing real-world training datasets and multimodal data-capture services to frontier AI labs and humanoid robotics companies. Public pricing and specific contract structures were not disclosed.