I built a free local LLM workflow with my 10-year-old-GPU, and it's reliable enough to replace the cloud

It was a bit of a hassle to set up, though