Initially, I did not include in the module the capacity to capture output directly in order to simplify its implementation and also because I was not able to imagine an API supporting it that would make sense while still being simple to learn an use...

Over time, I have come to appreciate that way of working. Using the file-system to store the generated data escalates quite well, allows easier debugging and in most cases, the performance penalty is negligible (usually the bottleneck is in the network). Also, the mechanism is easy to understand and to use.