How Jev computes in parallel

How does Jev — TypeSafe AI's "System One" model — achieve parallel computation, and what is the most defensible speculation about its underlying architecture?