Run it locally
A model that fits your machine still needs a runtime to load it. These are the ones worth knowing — every single one loads the same GGUF files our memory figures are measured from, so what the finder says fits will fit here too.
Never touched a terminal?
LM Studio — a desktop app with a built-in model browser. Search, download, chat.
Comfortable with a terminal?
Ollama — one command per model, runs as a service, and most other tools can attach to it.
Want every knob?
llama.cpp — the engine the others build on, with the most control and the newest features first.
01 /
One command to download and run a model.
Best for
The easiest serious start — and what most other tools build on.
Runs as a background service: `ollama run <model>` pulls the weights and starts chatting; every model on our device pages with an Ollama link works this way. Also exposes a local API that GUI tools can attach to.
02 /
The polished desktop app: browse, download, chat.
Best for
People who never want to see a terminal.
Free desktop app with a built-in model browser — search a model from our board, pick a GGUF quantization, and it shows whether it fits your machine before downloading. Can also serve a local API.
03 /
The engine underneath almost everything else.
Best for
Maximum control and the newest features first.
The project that created the GGUF format; Ollama, LM Studio and Jan all build on it. Compile or download a release, then run models with every knob exposed — including the KV-cache quantization options our finder lets you plan with.
04 /
Open-source desktop chat app.
Best for
A GUI where you can read the source of every part.
Desktop app in the LM Studio mold, fully open source. Downloads GGUF models, chats offline, and can connect to cloud APIs when you want both in one window.
05 /
A self-hosted web interface on top of Ollama.
Best for
Serving models to a household or a small team.
Runs in your browser against an Ollama or OpenAI-compatible backend: accounts, chat history, document upload — the pieces a shared setup needs. The natural next step once a model runs on one machine.
Tool installed? Then find out which model your machine actually runs — open the Local AI Finder →
All five are free to download. None of these links carries a referral — they route through our /go layer only so outbound clicks stay countable, like every external link on this site.