also i’m currently using the stack of opencode + kimi 2.5 + fireworks but i’m realizing i could probably expose ollama small models and direct claude to different models locally when i get the mac mini to handle tasks that don’t need a bunch of horsepower – what’s your stack currently look like?
@settee figuring out model routing is a difficult problem to solve, such that, its never really solved
@llbbl really not trying to “solve” just trying to be cost effective when it comes to simple ai stuff running locally vs the cloud spend of using dedicated services like replicate and fireworks, that’s all. just costings and what in the house/homelab can be kept local.
@settee Lookup or research Model routing. It’s complicated enough to turn side project into full time. From my understanding, the hard part is building and maintaining the “rules” around how to classify and decide where to route based on request. IE… “the harness”
@llbbl yeah i’ve been looking into that a bit with claude (claude code) but i’m starting to use opencode so i need to definately lock down my .md files correctly for each little project but at the moment they are a mess and i know i can be doing a lot more there to get better results! +1 on blogpost