A modular stack that cleanly separates visual expression from persistent memory, so AI-driven worlds can both look cinematic and reliably remember/respond