In March, through the Great Claude Token Drought™, I realised I needed a fallback. I'm becoming dependent on big players with their SOTA models, and getting used to the velocity they allow for. I needed to make sure I had an alternative that was (mostly) my own. So I spun up an ai. subdomain, gave it a chat GUI, and let it switch between claude -p and an ollama model. It's intended to be slow and mostly backgroundable. Here's how it went.

Read the original article