Forecasting for large numbers of related time series is a common need across many domains. E.g., in supply chains it may be necessary to forecast demand for millions of products at thousands of locations, amounting to billions of time series. Similarly in IoT and manufacturing there can be thousands of devices at thousands of locations. Handling the scale for fine-grained forecasting in such applications while supporting state-of-the-art forecasting components and their selection and evaluation is a challenging task that has not been fully addressed in existing systems. We present a toolkit for this purpose and describe the components, design, challenges, and results applying it to different data.