Non-Metric Space Library (NMSLIB) is an efficient cross-platform similarity search library and a toolkit for evaluation of similarity search methods. The goal of the project is to create an effective and comprehensive toolkit for searching in generic non-metric spaces. Being comprehensive is important, because no single method is likely to be sufficient in all cases. Also note that exact solutions are hardly efficient in high dimensions and/or non-metric spaces. Hence, the main focus is on approximate methods.