
LM Studio is a desktop app for running open-source large language models directly on your own computer. It is built for developers and privacy-conscious users who want models like gpt-oss, Llama, Gemma, Qwen, and DeepSeek without sending data to the cloud. You download a model once, then chat with it or serve it to your apps, fully offline.
LM Studio combines a graphical app with real developer tooling. Most ways to run local models are command-line only, while LM Studio gives you a point-and-click model browser, a chat window, and a server you start with one toggle. On Apple Silicon it runs both GGUF models (via llama.cpp) and MLX models, which use Apple's framework and GPU cores for faster inference than llama.cpp on Metal.
You search for a model inside the app, download it from Hugging Face, and start chatting in seconds. The same model can be exposed through a local, OpenAI-compatible API server, so you swap the endpoint in your existing SDK calls and run against a model that never leaves your machine.
For automation, LM Studio ships JavaScript (@lmstudio/sdk) and Python (lmstudio) SDKs, an lms CLI, and Model Context Protocol support. The headless llmster build runs the same core without a desktop interface, for Linux servers, cloud instances, and CI.
LM Studio is widely regarded as one of the easiest ways to run local LLMs, praised for its clean interface, simple model downloads, and the drop-in OpenAI-compatible server. Common criticisms are that large models demand a lot of RAM and a capable GPU, and that performance and output quality depend heavily on your hardware and the model.
The core app is free for personal and commercial use, so most individuals and developers pay nothing; teams and enterprises pay only for shared access and admin controls.
Ask specific questions about this tool.