What actually runs the model on your machine? Every local AI setup starts there, and three names come up every time: Ollama, LM Studio, Jan. All three let you run LLM locally. All three get you to a working model. They are built for different people, and picking the wrong one is a week of mild regret.
Ollama is the minimalist. Command line first, simple model library, a local API server other apps can talk to. Comfortable in a terminal? It is fast and stays out of your way. It is also the natural backend when your phone acts as client: Ollama serves, a local LLM management app for iOS and Android connects. The tradeoff is the interface, or lack of one. Everything is commands and config. Some of us like it that way. Most people do not.
LM Studio is the opposite bet. Polished desktop app, built-in chat, model discovery with one-click downloads. Friendliest on-ramp for someone who wants local AI without learning infrastructure first. The tradeoff: it is a desktop app, not a server. Great on the machine in front of you, less natural as the always-on backend for the whole house.
Jan sits in the middle with an open-source, local-first philosophy. Desktop app with chat, model management, an API mode for when other tools want to connect. If the open-source part matters to you, Jan wears it most proudly. I respect that.
So: terminal person building a home server, Ollama. Click-and-chat today, LM Studio. Open-source purist wanting both, Jan. And whichever you pick, the model matters more than the runner. Our open source LLM benchmark database search compares 200 open-weight models on benchmarks, hardware needs, and licenses, so you pair the right model with the right runner and watch it all from a mobile admin app for your local AI stack. Want it installed and tuned on your machines? I do private AI setup for businesses at privateaiagent.fyi. Your data never leaves.