Enact

Enact describes itself as the post training layer for physical AI. It produces the data that makes robotics models useful outside the lab. The company rolls out a policy on real tasks, identifies the specific states in which that policy fails, and then generates targeted demonstration and recovery datasets aimed at those failures rather than delivering a generic collection of trajectories. It is aimed at robotics teams training and refining control policies who have a model that works in demonstrations and breaks on contact with the real world. The loop is closed in house: Enact evaluates model performance itself to verify that the data it generated actually addresses the failures it was collected for, so the customer receives evidence about the effect of the data rather than only the data. The offering is real world robotics data, collected against the tasks and policies the customer brings. No pricing is published. There are no plans, rates or a dataset catalog on the site. Buyers start by sending a data request to the founders by email, and scope and terms are agreed directly. The site is a single page announcing the company's launch, with a link to its launch post on X. There is no product documentation, team page or customer list, so anyone evaluating Enact will need a direct conversation to establish what it can deliver and on what timeline.