With cloud agents I also process stuff overnight. But my gpu is limited. If I want to OCR a high number of documents or vectorize data, I can't run an LLM at the same time. I use remote models and use my compute for running commands on my desktop machine.