This week we filed a patent application on how Numa remembers. It covers the memory method behind the results in Claude Code vs Codex: forget, or pay to remember?, where Numa kept 90 to 96% of the details we planted across months-long conversations, at about a tenth of the cost on the same models.

It is the part of Numa that lets an intern stay with you. Memory is why you stop re-explaining, and why the jobs you teach Numa keep getting easier to hand over.

There is nothing you have to do. Numa is already built on it, and it keeps improving.