Anyone else slowly simplifying their local AI setup?
retained text
I feel like there's a point where the AI stack stops being fun and just becomes another thing to maintain. I had Ollama in a container, a web UI in another, a proxy so my coding agent could reach it, and a little script to unload models when RAM got tight. Four moving parts for one chatbot. Lately I've been leaning toward fewer moving parts. Moved to Magnitude on the Mac, which is one app that picks the model for my RAM and has a one click connect for the coding agent, and deleted the rest. The thing I noticed most was the wait on long prompts, the same small Gemma model now starts answering in about 2 seconds instead of 6. Anyone else gone through a phase of ripping out half their setup after years of adding to it?
document fetched 2026-11-16 00:27 · canonical https://reddit.com/r/selfhosted/comments/1xwzudx/anyone_else_slowly_simplifying_their_local_ai