Yes, this! I do similar things with local models. Small tasks at volume. I also use local models for things that demand more privacy ... managing or dealing with content I wouldn't want to send to the cloud.
I've been playing with local LLMs on a 4 year old Mac Studio and have been quite impressed. I keep remembering that the LLM I have available to me today is the worst it will ever be. It will just keep getting better. I'm super interested in and curious about the path forward for local inference.