Refresh is building infrastructure for RL environments and datasets designed for rigor and reproducibility.
We create rigorous, sandboxed environments that enable AI systems to learn and execute real-world tasks, from navigating complex software interfaces to writing production-grade code.
Our environments provide verifiable reward signals for reinforcement learning, helping labs train models that can reliably automate computer use and coding workflows.
We believe rigorous environments and curated datasets are essential to advancing reinforcement learning.