Partitioning is the easy bit I think, and falls out naturally from Postgres.

But how are people doing recency biasing with vector databases?

All of our approaches are either expensive or super hacky.