Quick answer
LM Studio is the easiest desktop app for running open-weight models, with hardware guidance and a polished interface, but it is closed source. Jan is a fully open-source desktop alternative that also connects to cloud providers. Open WebUI is a self-hosted web interface, usually run with Docker over Ollama, with users, roles, document chat, and web search, which makes it the pick for teams and home labs. All three are free.
Running a model on your own machine means no subscription, no data leaving the building, and no rate limits. The models are now good enough for a lot of everyday work. What used to be the hard part, getting a friendly interface on top, is now a choice between three good free options.
LM Studio: easiest start
LM Studio searches Hugging Face for models, tells you which ones will fit your RAM and GPU, downloads them, and gives you a chat window and a local server. It is the app to recommend to someone who has never run a local model. Its limitations are that the code is not open, and its multi-user and document features are thin.
Jan: open source, local plus cloud
Jan does most of what LM Studio does with an open-source licence, and it lets you add cloud providers alongside local models in the same app, so one window can hold a local Llama and a cloud Claude. It is a little less polished in model discovery and hardware advice, and its feature set is smaller than Open WebUI's.
Open WebUI: self-hosted for teams
Open WebUI is a web app you run on a server or your own machine, usually alongside Ollama. It has accounts and roles, document upload with retrieval, web search, image-generation hooks, and admin controls, which is why small companies use it as a private ChatGPT. The price is setup: Docker, a reverse proxy if you expose it, and updates. Its licence for large enterprise deployments changed in 2025, so check the terms if you are over the user threshold.
Hardware, briefly
- 8 GB of RAM or VRAM runs small models (around 7B parameters, quantised) at usable speed
- 16 to 24 GB runs mid-sized models comfortably; Apple Silicon with unified memory is well suited
- Anything larger needs a dedicated GPU or a rented one, at which point a cloud provider may be cheaper
Related reading
Bottom line
LM Studio for your first local model. Jan if open source matters or you want cloud and local in one app. Open WebUI when more than one person needs access. They all run the same models, so the choice is about interface and who is using it.
