Tooling

LM Studio

GUI-first local LLM app, closed-source but free, the easiest path for non-developers in 2026.

LM Studio is the GUI-first answer to "I want to run local LLMs and I don't want a terminal." Closed-source (its one big asterisk) but free for personal use, polished, cross-platform, and the easiest install for someone who doesn't write code. Under the hood it's llama.cpp and (recently) MLX on Apple Silicon, with a Hugging Face model browser and a built-in OpenAI-compatible server.

For full FOSS alternatives see Jan / GPT4All. For the developer-first default see Ollama. For the broader picture see the overview and hardware tier guide.

What it is

  • macOS, Windows, Linux desktop app. Apple Silicon native; CUDA / ROCm GPU on Win/Linux; CPU fallback.
  • Model browser — search Hugging Face for GGUF / MLX models; one-click download.
  • Chat UI — multi-turn chats; system prompt; sampling controls; multiple chats side-by-side.
  • Server mode — built-in OpenAI-compatible REST API on localhost.
  • Offline once installed — no telemetry beyond model downloads.
  • Closed source. This is the catch — the engine inside (llama.cpp / MLX) is FOSS, the wrapper is not.
  • Free for personal use; paid for "Business" licensing (the team has been straightforward about this).

Why people pick it

  • No terminal required. Download the .dmg / .exe, double-click, search for a model, chat.
  • Hugging Face integration. Browse, filter by quant / size, see "this fits your machine" hints.
  • Apple Silicon MLX support. First-class; competitive with mlx-lm directly. Large unified-memory Macs run beautifully.
  • Chat-with-doc — upload a PDF, chat over it (basic RAG, not as deep as Open WebUI).
  • OpenAI-compat server — point Aider / Continue.dev at localhost:1234 and it works.

Why people skip it

  • Closed source core. For OSS purists this disqualifies it; Jan is the FOSS GUI alternative.
  • Heavier than Ollama for headless / server use.
  • No model registry tags — you manage GGUFs as files, not as named tags.
  • Less hackable. Ollama Modelfiles + Open WebUI plugins beat LM Studio for power-user customization.

When to pick LM Studio

  • You want a GUI-first install. Ollama + Open WebUI is two installs and a docker compose; LM Studio is one .dmg.
  • You're handing this to a non-technical family member. "Open this app, click chat" is the win.
  • You're on a big Mac (M3 Max / Ultra). MLX integration is excellent; you can drive 70B+ models with a polished UI.
  • You want a desktop app, not a web app. Open WebUI is web; LM Studio is native.

When to skip LM Studio

  • You're already comfortable with terminals. Just use Ollama.
  • You insist on FOSS. Use Jan or Ollama + Open WebUI instead.
  • You want server / multi-user serving. This isn't LM Studio's lane.

Honest comparison

OllamaLM StudioJanGPT4All
OSS licenseMITClosedAGPLMIT
GUINo (Open WebUI on top)Yes, native desktopYes, nativeYes, native
Apple MLXNo (uses Metal via llama.cpp)YesPartialNo
API serverYesYesYesYes
Model registryYes (ollama.com)HF browserHF browserHF browser
Best forDevs, serversNon-devs on desktopFOSS-only desktopLight FOSS desktop

Pick this if…

  • You want a polished desktop app and don't care about closed-source: LM Studio.
  • You have a big Mac and want MLX: LM Studio.
  • You insist on FOSS: Jan or Ollama + Open WebUI.
  • You're handing the laptop to your spouse: LM Studio.
  • You're a developer: Ollama.

On this page