MLX research prototype for content-routed subquadratic sparse attention and long-context associative recall. - View it on GitHub