Local models speak the same language as the cloud — so wiring one into your code is mostly changing a single line.