Helpful DoggyBot is an indoor mobility operating system for quadrupedal robots that manipulates objects using a front-end gripper and leverages low-level controllers trained in simulated environments to achieve agility skills like climbing and full-body inclination. Additionally, it incorporates pre-trained visual-language models (VLMs) for semantic understanding and command generation. The system can generalize to unseen environments to accomplish tasks with zero-shot learning, such as retrieving a randomly placed toy from beside a bed after climbing, with a success rate of 60%, without any actual data collection or training.