Calorine is a Python package that simplifies the construction, analysis, and use of neuroevolution potentials (NEPs) via GPUMD. NEPs are a class of machine-learned interatomic potentials that combine the speed of heuristic force fields with the accuracy of ab-initio methods. They are highly accurate and efficient, and have been used to study various properties in a range of materials, including radiation damage in tungsten, phase transitions, and dynamics of halide perovskites. GPUMD is a C++/CUDA package that enables MD simulations and NEP model construction, with all computations running on a discrete GPU. Calorine provides a Python interface that makes it easy to access GPUMD's functionality and integrate it into Python-based workflows. This includes managing NEP model construction and setting up and analyzing MD simulations. Calorine also exposes two ASE Calculator objects, one using the CPU and one using the GPU, enabling NEP models to be used outside of GPUMD and on machines without discrete GPUs. Calorine differs from other software packages like PyNEP and GPYUMD by having a broader scope, encompassing both NEP construction and sampling with MD simulations. It also provides an interface for modifying potential files, improving the transferability of NEP models. Examples of recently published work supported by Calorine include studies on the through-plane lattice thermal conductivity in van-der-Waals structures and dynamic modes in halide perovskites under a continuous-order phase transition. Calorine is supported by the Swedish Research Council and the Swedish Foundation for Strategic Research, and enabled by computational resources provided by the National Academic Infrastructure for Supercomputing in Sweden and the Swedish National Infrastructure for Computing. The full documentation, examples, and tutorials can be found at https://calorine.materialsmodeling.org/.Calorine is a Python package that simplifies the construction, analysis, and use of neuroevolution potentials (NEPs) via GPUMD. NEPs are a class of machine-learned interatomic potentials that combine the speed of heuristic force fields with the accuracy of ab-initio methods. They are highly accurate and efficient, and have been used to study various properties in a range of materials, including radiation damage in tungsten, phase transitions, and dynamics of halide perovskites. GPUMD is a C++/CUDA package that enables MD simulations and NEP model construction, with all computations running on a discrete GPU. Calorine provides a Python interface that makes it easy to access GPUMD's functionality and integrate it into Python-based workflows. This includes managing NEP model construction and setting up and analyzing MD simulations. Calorine also exposes two ASE Calculator objects, one using the CPU and one using the GPU, enabling NEP models to be used outside of GPUMD and on machines without discrete GPUs. Calorine differs from other software packages like PyNEP and GPYUMD by having a broader scope, encompassing both NEP construction and sampling with MD simulations. It also provides an interface for modifying potential files, improving the transferability of NEP models. Examples of recently published work supported by Calorine include studies on the through-plane lattice thermal conductivity in van-der-Waals structures and dynamic modes in halide perovskites under a continuous-order phase transition. Calorine is supported by the Swedish Research Council and the Swedish Foundation for Strategic Research, and enabled by computational resources provided by the National Academic Infrastructure for Supercomputing in Sweden and the Swedish National Infrastructure for Computing. The full documentation, examples, and tutorials can be found at https://calorine.materialsmodeling.org/.