If you were trying to fold a symmetric homo-oligomer, there’s the application FoldAndDock, but that’s not for heteroligomers, though.
I am not aware of a protocol in Rosetta which will ab-initio fold a protein in complex with another protein (even one where the interacting partner is already know). What I would suggest to try instead is to do a regular ab inito run with the unknown protein, and then in addition to the regular post processing, do protein-protein docking runs with the putative models (probably limited to cluster centers, to keep the computational costs down). Hopefully, one of the clusters will stand out as having a good docking energy. You can then further explore members of that cluster to see which model best balances folded structure and binding energy.
.
