A research PyTorch framework for accelerating diffusion planning in decision‑making tasks. It “habitizes” slow diffusion planners by transferring them into a fast habitual policy suitable for real‑time inference. The authors report 800+ Hz action frequency even on a laptop CPU and evaluate the approach on offline‑RL benchmarks such as D4RL (Mujoco, Kitchen, AntMaze, Maze2D). We use it to explore efficient diffusion‑based planning and deployment‑friendly decision models.