We use a custom in-house sync engine (similar to how Git works) to propagate messages between peers, but the "source of truth" is local first and a sync server.

This gives you full ownership of your data, instant search (even offline), and naturally supports the kind of privacy/custody requirements you're describing.