Discretize VAR(1) of arbitrary size, with arbitrary covariance matrix for innovations. Support for VAR(1) with covariance matrix perturbed by common AR(1) volatility shock, e.g. "volatility regime," like baseline Bansal-Yaron process. Allows the elimination of support points with low probability in the ergodic distribution (non-tensor grid). Uses the Armadillo library for C++, with HDF5 support for I-O.
Looking instead for a MATLAB library? Consider the code repository for "Discretizing Nonlinear, Non-Gaussian Markov Processes with Exact Conditional Moments" by Leland E. Farmer & Alexis Akira Toda, in QE, or the refinement of Grey Gordon's "Efficient VAR Discretization" in EL.