We present a new technique for designing and analyzing non-Markovian couplings for bounding the convergence rate of Markov chains. As an example, we use our theorem to improve the threshold for efficiently sampling colorings of bounded-degree graphs.