Love it! Which local model do you use to transcribe the screenshots? I once built a similiar tool and found it a hard trade-off in terms of quality so ended up using Antrophic API...