This repository implements a dictionary-driven, sheaf-based semantic communication framework for heterogeneous AI agents, learning both the communication topology and sheaf maps for latent-space alignment, supported by a sparse dictionary–based semantic denoising and compression module that builds a shared global semantic space.