The article presents OK-Robot, an open and modular framework for home robots that can perform pick-and-drop tasks without requiring training. The system uses vision-language models, navigation primitives, and grasping primitives to achieve a 58.5% success rate in 10 real-world home environments. The authors analyze the performance of OK-Robot and identify the leading causes of failures, including difficulties in retrieving the right object and hardware issues. The framework is open-source and available on GitHub.