OmniSci is designed for big data but for custom computations the data needs to be retrieved to a client machine. This puts a ceiling on data sizes based on both network speed and local hardware resources. What if we could send the algorithm to the data instead?