Model Routing & Local AI is a working prototype at 65% readiness. The principle is simple: your hardware does the everyday work, and expensive cloud models are called only when the task truly needs them. Working now: local model execution, including GPU, and the routing foundation that sends each subtask to the right model. Every routing decision weighs privacy, latency, cost, capability, and available compute — inside the data boundaries you set. Still ahead: load-aware scheduling, benchmarking across models, and production reliability. The Private AI page documents the four deployment modes: local-only, hybrid, cloud, and private infrastructure.