Great read on the local-first architecture! One thing I'm wondering about though: doesn't shifting to locally hosted OCR VLMs and serving models locally significantly increase your infrastructure and inference costs compared to lightweight APIs, especially if you have to provision dedicated GPUs?
Muhammad Azlaan Zubair
Software Engineer | Applied AI Researcher | AI Workflows & Automation
Great read on the local-first architecture! One thing I'm wondering about though: doesn't shifting to locally hosted OCR VLMs and serving models locally significantly increase your infrastructure and inference costs compared to lightweight APIs, especially if you have to provision dedicated GPUs?