Or run the whole engine inside your own walls.

Same engine, same API, same model addresses — on your machines. We stop metering your calls: you pay for a licence, not for tokens.