So, is the only way to use this feature to setup your own instances of the models (which need resource heavy GPU instances that would be costly)? This feature looks very useful, but it seems like it would cost me more to set up than my actual Discourse hosting.