We describe the functionality of pyATF – a new, generic, publicly available open-source auto-tuning tool for optimizing programs written in arbitrary programming languages and whose performance-critical parameters may be constrained (e.g., the value of one parameter has to divide the value of another parameter). The user interface of pyATF is designed with a particular focus on usability for real-world demands, and it is offered in the Python programming language which is becoming increasingly attractive due to its ease of use even for non-expert users. We experimentally confirm the practicality of pyATF using real-world studies from the areas of quantum chemistry, image processing, data mining, and deep learning: we show that pyATF is able to auto-tune the complex parallel implementations of our studies to higher performance than achieved by state-of-practice approaches, including hand-optimized vendor libraries.