Adaptive AI execution substrate. Distributed inference and elastic compute for real-time workloads. Open weights, one endpoint, no lock-in.