The Toolkit for Advanced Optimization is distributed as part of the PETSc and supports unconstrained, bound-constrained, and PDE-constrained optimization problems and variational inequalities on high-performance architectures. It contains many methods, including derivative-free methods such as POUNDERS, nonlinear conjugate gradient and limited-memory variable-metric methods that require only first derivatives, and trust-region and line-search methods that can utilize second derivatives.