Pick the interface, not the hype
Ollama: the engine
Ollama runs quietly in the background and is supported by a very wide range of other apps, from document chat tools to coding assistants and agents. It has a simple app and a command line. It now also sells cloud models; those have names ending in :cloud. Choose Ollama if you'll connect other tools to it.
LM Studio: the workshop
The classic LM Studio app makes it easy to search for models, see whether they fit your memory, download them and tune settings, with a built-in chat and a local server. It is free, including at work. Its website now leads with Bionic, a separate agent product with paid cloud models; if you want the classic app, choose 'LM Studio' on the download page.
Jan: the open-source chat app
Jan looks like ChatGPT, runs offline once a model is downloaded, and is open source under the Apache 2.0 licence. It has Projects with file uploads, optional web search, and tool connections that ask before they run. You can add cloud models with your own API key, so check which model a chat is using.
Installing your first one
Download from the official site only (ollama.com, lmstudio.ai, jan.ai); fake installer pages for popular AI tools do exist. Install, open the app, pick a small model it recommends, wait for the download, and ask a question. Then switch off Wi-Fi and ask another. If it still answers, you are running AI locally.
Do you want a chat window to talk to, or an engine that other apps can use? Or both?