This paper develops a feedback linearization framework for constrained optimization, with convergence guarantees, a connection to SQP, and a momentum-accelerated variant.