The consumer AI app where prompts stop at your server

You run an open-weights model as a daemon program in a container on a GPU-category server from the marketplace. Your app's exec handlers call it at localhost:11434. No provider key, no gateway, no token leaving the box. Entries go in, reflections come out, and the path is a loopback interface.

Hoody's API sees that a container exists. It doesn't see prompts or responses; the proxy is on your server. Hosted inference would have been metered per token and read by a third party. This is a flat-rate box and a model you chose.

Your users' prompts have exactly one reader, and you rent the rack it lives in.