All Local AI
Pick VRAM for fit estimates
Models Tools Hardware Best Of Compare Calculators Stacks Guides Blog
desktop app MIT linux macos windows

LM Studio

Desktop app for discovering, downloading, and chatting with local models, with a built-in OpenAI-compatible server.

GitHub stars

5,353

License

MIT

Install

binary

Latest release

—

Source on GitHub Documentation Official site

Best for

People who want a graphical interface, model search, and a local API without touching a terminal.

Model formats

  • GGUF
  • MLX

Hardware behaviour

Shows an estimated fit indicator against your detected VRAM before you download, and exposes a GPU-offload slider that maps to llama.cpp's layer offload.

Not sure what your card can hold? Run the fit check — the entry point in our listings is the RTX 4060 8GB at 8.00 GB.

Setup

Install LM Studio

Copy-paste starting points. Always read the project's own instructions before running anything with elevated permissions.

Download

Download

# Get the installer for your OS from https://lmstudio.ai

Closed-source desktop app; the CLI is open source.

CLI

CLI

npx lmstudio install-cli && lms ls

Lists locally downloaded models.

Start the local server

LM Studio

lms server start --port 1234

Call the API

LM Studio

curl http://localhost:1234/v1/models

Quick start

Getting the first model running

Search a model in the app, check the fit badge against your GPU, download, and chat. The Local Server tab turns the same model into an OpenAI-compatible endpoint on port 1234, which is the fastest route from "I want to try a local model" to "my script is calling it".

On Apple Silicon it can load MLX builds as well as GGUF, which is usually the faster option on those machines.

FAQ

LM Studio questions people actually ask

Short answers to the problems that come up most often.

Is LM Studio free, and is it open source?

The desktop app is free to use, including for work under its current terms, but the app itself is closed source. The lms command-line tool is open source. If a fully open stack matters to you, use llama.cpp or Ollama instead.

What does the GPU offload slider change?

It sets how many model layers go to the GPU, the same idea as llama.cpp's -ngl. Push it to maximum for speed; pull it back when the model will not load or the app falls back to CPU mid-generation.

Can other apps use LM Studio as a backend?

Yes. Start the local server and point any OpenAI-compatible client at http://localhost:1234/v1 with any placeholder API key.

Why does a model download but fail to load?

Almost always memory. The fit indicator is based on detected VRAM, but the context length you select adds KV-cache on top of the weights. Reduce context, choose a smaller quantization, or lower GPU offload so part of the model sits in system RAM.

GGUF or MLX on a Mac?

On Apple Silicon, MLX builds generally run faster because they target the unified-memory architecture directly. GGUF remains the safer default for compatibility and has far wider model coverage.

Alternatives

Compared with other desktop app options

Same job, different tradeoffs.

inference engine

Ollama

One-command local model runner with an OpenAI-compatible API and automatic GPU offload.

 inference engine

llama.cpp

The C/C++ engine that made CPU and mixed CPU/GPU inference practical, and the origin of the GGUF format.

 inference server

vLLM

High-throughput serving engine with PagedAttention and continuous batching, built for concurrent requests.

Next step

A runner is only half the decision. Check which models fit your GPU with the VRAM calculator, then follow a setup guide or pick a ready-made stack.