Kriging and geostatistical simulation underpin environmental modeling, yet practitioners must choose between aging file-driven Fortran codes such as GSLIB and convenient but slower high-level packages. We present KrigeKit, an open-source (MIT) library pairing an OpenMP-parallel Fortran engine with a dependency-light Python layer. A single engine provides ordinary, simple, and co-kriging, kriging with external drift, sequential Gaussian and indicator simulation, and space-time kriging. Its hallmark is the breadth of advanced capabilities it exposes together &mdash; multiplicative covariance structures, gradient constraints, spatially varying anisotropy, and positive-semidefinite co-kriging fitting &mdash; seldom available in one package. Performance comes from parallel block estimation, tabular variogram evaluation, and a per-thread Cholesky cache that cuts the per-block cost from <em>O</em>(<em>n</em>³) to <em>O</em>(<em>n</em>²) on repeated neighborhoods. We verify KrigeKit against analytic results and established packages, characterize its parallel scaling, and demonstrate it on three environmental case studies. KrigeKit ships as pre-built wheels on PyPI for Linux, macOS, and Windows.