Run an LLM locally without the command line using LM Studio

By Mouhamed BANKOLE, CEO at ITNET Technologies · Updated November 6, 2026 · Reading time: 9 min

You want to try an AI model on your computer, but the command line puts you off. LM Studio is built for exactly that: an app with a graphical interface to find a model, download it and chat with it, without sending anything over the Internet. Your machine still has to handle the load, you need to pick a suitable model, and you should know when the tool reaches its limits. Here are the system requirements, installation, model choice, the OpenAI-compatible server, the comparison with Ollama and moving to a GPU server.

In short

• LM Studio is a desktop app for macOS, Windows and Linux that downloads and runs language models locally, with a graphical interface.
• The recommended setup is at least 16 GB of RAM and, on Windows, 4 GB of dedicated VRAM; on a Mac you need Apple Silicon.
• It includes a local server compatible with the OpenAI API, which your scripts and tools can call.
• LM Studio suits a workstation, while Ollama is easier to automate on a server: the two complement each other.
• When the model outgrows your PC's VRAM, a rented GPU server takes over.

Contents

1. What is LM Studio?
2. What setup do you need for LM Studio?
3. How do you install LM Studio and run a first model?
4. Which model should you choose in LM Studio?
5. How do you use LM Studio as an OpenAI-compatible server?
6. LM Studio or Ollama: which one should you choose?
7. Can you run LM Studio on a server?
8. When should you move from a PC to a rented GPU?
9. How do you secure access to an LM Studio server?

What is LM Studio?

LM Studio is a desktop application for finding, downloading and running open language models on your own computer, with no command line. Everything happens in a graphical interface: model catalog, chat window, settings.

What you can do with it:

• chat with a model as you would with an online assistant, but offline;
• ask questions about your documents (PDF, text), as shown in the DataCamp LM Studio tutorial;
• compare several models on the same question;
• expose a local API for your own development.

For the command-line take on the same idea, see our Ollama guide.

What setup do you need for LM Studio?

At least 16 GB of RAM is recommended, a CPU with AVX2 on a PC, and Apple Silicon on a Mac: Intel Macs are not supported. These are the requirements published on the LM Studio system requirements page.

• VRAM is the real limit: the more you have, the larger and faster the model can be.
• Without a dedicated GPU, LM Studio still works, but only with small models and slower answers.

How do you install LM Studio and run a first model?

Download the app from the official website, open the model search tab, download a small model, then load it in the chat window. Allow ten minutes, model download included.

1. Install LM Studio from lmstudio.ai for your system.
2. Open model search (Discover tab) and type, for example, "Mistral" or "Qwen".
3. Pick a quantized version (a 4-bit GGUF file, labelled Q4) that fits in your memory.
4. Load the model in the chat tab and ask a first question.
5. Adjust context and temperature in the settings if answers are cut off or too random.

Pro tip: start with a 7B to 8B model in Q4, which weighs about 4 to 5 GB according to the Ollama library. It gives a good feel for quality without overloading the machine; then move up in size if your VRAM allows.

Which model should you choose in LM Studio?

Choose the largest quantized version that fits entirely in your card's VRAM, with room left for context: below that, the model spills into RAM and slows down sharply. The most common format is GGUF, available in several quantization levels (Q4, Q5, Q8…), as the DataCamp tutorial explains.

• Q4: the best size-to-quality trade-off for most uses.
• Q8: more faithful, but heavier, because each weight is stored on 8 bits instead of 4.
• Model size: our VRAM guidelines per model are in our Ollama guide; the orders of magnitude are the same.
• Use case: a coding model for development, a general model for writing, a multilingual model for non-English text.

How do you use LM Studio as an OpenAI-compatible server?

Turn on the local server from the interface or run lms server start: LM Studio then exposes endpoints compatible with the OpenAI API, callable from any OpenAI SDK. The LM Studio developer documentation lists Chat Completions, Embeddings, Responses and model listing, among others.

Commands of the lms command-line tool:

• In your scripts, replace the OpenAI API address with http://localhost:1234/v1.
• The documentation examples use a token passed in the Authorization header.

LM Studio or Ollama: which one should you choose?

LM Studio is the most comfortable on a workstation thanks to its graphical interface; Ollama is easier to install and automate on a Linux server. Both run open models locally and expose an API.

• To test and compare models on your PC: LM Studio.
• For a shared service on a GPU server: Ollama, possibly with a web interface.

Can you run LM Studio on a server?

Yes, LM Studio offers llmster, its engine packaged as a headless service for servers and cloud instances. The developer documentation says it installs on Linux with a script, curl -fsSL https://lmstudio.ai/install.sh | bash.

On a server, the process is:

1. Connect over SSH to the server.
2. Install llmster with the official script.
3. Download a model with lms get.
4. Start the server with lms server start, without exposing it directly to the Internet.

If you prefer a tool designed for servers from the start, our Ollama guide covers installation on a rented GPU. A VPS is enough to host a web interface that calls the model remotely.

When should you move from a PC to a rented GPU?

Move to a rented GPU when the model you want no longer fits in your card's VRAM, when several people need to use it, or when the service must stay on all the time. The PC remains ideal for testing.

Typical situations:

• a 20B to 30B model to run without spilling over on a consumer card, when the 27B version weighs 17 GB according to the Ollama library;
• a team that wants a shared internal assistant;
• long batch jobs (summarizing documents in bulk) that tie up the workstation.

Wayhost GPU rental offers a dedicated NVIDIA V100 32 GB at €0.60 excl. VAT per hour or €199.99 excl. VAT per month, hosted in Cergy with root access by SSH key (checked on November 6, 2026). For a whole machine of your own, also see the physical server offer and our dedicated server guide.

How do you secure access to an LM Studio server?

Keep the LM Studio server reachable only locally and go through an SSH tunnel or an HTTPS reverse proxy with authentication. An AI server left open without protection can be used by anyone, at your expense.

• SSH tunnel: ssh -L 1234:localhost:1234 root@server, then call http://localhost:1234 from your workstation.
• API token: turn on the authentication described in the documentation and never put the token in your code.
• Firewall: block the server port from outside.
• Updates: keep LM Studio and the system up to date.

Our guide to hosting an AI application describes the full architecture of an AI service exposed to users.

Frequently asked questions

Is LM Studio free?

Yes, LM Studio can be downloaded for free. It is not open-source software, unlike Ollama. Each downloaded model has its own license.

Does LM Studio work without the Internet?

Yes, once the model is downloaded, everything works offline. Your questions and documents stay on your computer. The Internet is only needed to download models and updates.

Which graphics card do you need for LM Studio?

LM Studio recommends at least 4 GB of dedicated VRAM on Windows. The more memory the card has, the larger the models you can load. Without a GPU, only small models stay comfortable.

Does LM Studio run on an Intel Mac?

No, LM Studio requires an Apple Silicon Mac (M1 to M4) running macOS 14 or later. Intel Macs are not supported. On those machines, Ollama remains an option.

What is a GGUF file?

It is a file format for quantized models, used by LM Studio and other local tools. Quantization shrinks the model so that it fits in memory. Q4 is the most common trade-off.

Can you use LM Studio with your own programs?

Yes, its local server is compatible with the OpenAI API. You only need to change the API address in your code. Existing OpenAI SDKs work with no other change.

LM Studio or Ollama: which one is faster?

Both rely on comparable inference engines, and speed depends mostly on the model and the graphics card. The difference is in how you use them: a graphical interface for LM Studio, automation for Ollama. Test both with the same model.

Key takeaways

• LM Studio runs LLMs locally with a graphical interface and no command line.
• 16 GB of RAM and a dedicated GPU are recommended; Intel Macs are not supported.
• Choose a quantized GGUF model that fits entirely in VRAM.
• The OpenAI-compatible local server connects LM Studio to your scripts and tools.
• For a team or a large model, move to a GPU server with protected access.

To run your models on an NVIDIA V100 32 GB hosted in France, discover Wayhost GPU rental.

Article written by Mouhamed BANKOLE, CEO at ITNET Technologies.