I am working on the a memory system governance gateway that sits between long-running AI agents and their persistent state. Every candidate fact is scored for confidence and blast radius, the safe majority is auto-written, the ambiguous middle is routed to a human, and every decision is captured in a hash-chained audit trail.
