Always have been, unless you're using the API meant for apps.
But if you're doing something very basic, you might be able to slop together a tool that does local inferencing based on a small, local model instead, alleviating the need to call Claude entirely.
