Right now, each nu->nu pipeline collects all the results before moving to the next step. Ideally, all pipelines would be streaming so we could seamlessly integrate with external commands and also handle commands with large amounts of data.