At scale you definitely want a separate transactor, but for development or for deploying an early prototype , it would be very handy to be able to have the transactor running in the same process as the peer. This would trade reliability and predictability for resource efficiency, and allow deploying on an inexpensive instance somewhere in a public cloud. This would probably lower the bar for trying out datomic in many settings.
In fact, it seems that datomic local allows having an embedded H2 database and a transactor into the same JVM process as your client.
You mentioned that it is possible to do the same for datomic on-prem, could you detail how one might do it?