We introduce Forecasting in Non-stationary Offline RL (FORL), a novel framework designed to be robust to passive non-stationarities, leveraging diffusion probabilistic models and time-series forecasting foundation models.