[sig_kill]

This is actually how I develop and use the mesh at home. Rather than splitting models, I aggregate disparate compute behind one endpoint, without having separate inference providers on each host and a gateway like LiteLLM