CIANNA provides a framework for creating and training deep-learning models for astronomical data analysis. Functionalities and optimizations are added based on relevance to astrophysical problem-solving. CIANNA allows the building and training of a wide variety of neural network architectures aiming for various tasks through a high-level Python interface. It supports both computing on CPU and GPU acceleration through low-level CUDA programming, taking advantage of AI-dedicated hardware substructures. CIANNA distinguishes itself from other frameworks by its low latency, allowing tight integration with other codes.