Currently SpamBrainz is not designed to allow training new models entirely "in the cloud" or more concretely on the Hetzner-Based infrastructure.

Training new models should be triggered manually for now but still rely on the existing webservice. If necessary, ancillary workers for training could also be spun up.