Deepo is an open framework for painlessly assembling specialized Docker images for deep learning research. It provides modular Lego-like components for customizing deep learning toolchains and automatically resolves dependencies between CUDA, cuDNN, Python, and frameworks like PyTorch and TensorFlow through topological sorting. The project includes pre-built images supporting widely used frameworks including Caffe, Caffe2, Chainer, Keras, MXNet, ONNX, PyTorch, Sonnet, TensorFlow, Theano, and Torch.