The tools people use

The same job, four ways

All four tools here do the same underlying thing. They load a model file and run it on your computer.

They differ in what they ask of you, what they hand you, and how much they decide on your behalf.

One of them is Pip, made by the company publishing this course. Its limits are named here beside the others. A tool that suits you matters more than a tool that suits us.

Three of the four are free. That is worth knowing before you read any further.

Backed by llama.cpp LICENSE (MIT), Ollama LICENSE (MIT), LM Studio is free for use at work, Pip.cpp: use AI freely, keep control.

What they share underneath

Before the four descriptions, one fact that changes how you read them. Three of these four share a single engine underneath, and that engine is llama.cpp.

When the engine is the same, the tool is rarely what decides your speed. The model you picked and the machine you own decide it.

So choosing between those three is mostly a choice about the wrapper. How you find a model, how you start it, and how much is arranged for you.

Two caveats. LM Studio has a second engine on Apple Macs, named below. And this course has not verified which engine Pip uses, so Pip stays out of this comparison.

The useful conclusion is a saving. If a tool feels slow, switching tools is rarely the fix. Lesson 14 and Track 5 go after the causes that actually move it.

Backed by llama.cpp: LLM inference in C/C++, Ollama: start building with open models, Welcome to LM Studio Docs.

llama.cpp: the engine

llama.cpp is the engine underneath. Ollama lists it as a supported backend, and LM Studio runs models with it. It is free and open source under the MIT licence.

Its stated goal is to run models with minimal setup and strong performance on a wide range of hardware.

It is plain C and C++ with no dependencies. It supports Apple chips, Nvidia and AMD graphics cards, Intel and AMD processors, and a long list of others.

It can split a model between your graphics card and your main memory. A model too large for the card alone will still run, more slowly.

It is not terminal only, which surprises people. Running llama serve starts a small web server, and its documentation says the web interface is switched on by default. You open 127.0.0.1 port 8080 in a browser and there is a chat window.

What it asks of you: the command line to start it. You get every setting and every flag, and nothing is arranged for you.

Backed by llama.cpp: LLM inference in C/C++, llama.cpp LICENSE (MIT), llama.cpp HTTP server: REST endpoints and the built-in web UI, Welcome to LM Studio Docs, Ollama: start building with open models.

Ollama: one command

Ollama is free and open source under the MIT licence. It runs on macOS, Windows and Linux.

You type a single command, such as ollama run gemma4. It fetches the model and drops you into a chat.

It also runs a small server on your own machine, at localhost port 11434, so other programs on your computer can send it questions.

That second job is the one it is really built for. Its own front page offers to connect it to coding tools such as Claude Code, Codex and Copilot, one command each.

Ollama installs as a desktop app as well, on macOS and on Windows. So a terminal is not the only way in.

• Better than Pip at: price. It is free, and MIT licensed.
• Better than Pip at: feeding your other software. It answers at a fixed address on your machine, and ships one-command hookups for coding tools. Pip's site describes an app you talk to, and lists nothing of that kind.
• Costs you: picking the model name yourself, and checking each time whether it is the local one or the cloud one.

Backed by Ollama: start building with open models, Ollama LICENSE (MIT), Ollama quickstart, Ollama cloud models, Ollama on macOS: the desktop application and its files, Ollama on Windows: a native Windows application, Pip.cpp: use AI freely, keep control.

LM Studio: a window, not a terminal

LM Studio is a desktop app with a chat window. It runs on macOS, Windows and Linux.

You search for models inside the app, download them from Hugging Face, and chat without ever opening a terminal.

It runs models through llama.cpp. On Apple Silicon Macs it can also use Apple's own MLX.

It does the serving job too. Its documentation lists serving local models on OpenAI-like endpoints, and connecting to tools such as Codex and Claude Code.

Its documentation says nothing you type when chatting with a model leaves your device, and that the app works offline once a model is downloaded.

It is free at home and at work. That changed in July 2025; before then, use at a company needed a separate licence.

• Better than Pip at: choice. You can reach any model on Hugging Face rather than a tested shortlist.
• Better than Pip at: price. It costs nothing, at home or at work.
• Costs you: you are still the one choosing the model, and there are a great many settings.

Backed by Welcome to LM Studio Docs, LM Studio offline operation, LM Studio is free for use at work.

Pip: a narrower path

Pip is a desktop app from RedRooster Technologies, who publish this course.

Its idea is different. Rather than offer every model, it checks your computer and recommends a model it has already tested.

What it asks of you is money. Pip is paid software and the other three are free.

That is the honest trade. You are paying for a shorter decision and a tested pairing, not for a better engine underneath.

This course does not print what Pip costs, for the same reason it prints no cloud prices: a number in a lesson goes stale and then misleads. The current price is on the Pip site, next to the download.

If you enjoy choosing, or you want a model Pip has not tested, one of the other three will serve you better.

• You want every setting, and a terminal does not scare you: llama.cpp.
• You want one command and a library of models: Ollama.
• You want a window, a search box and no terminal: LM Studio.
• You would rather not choose a model at all, and will pay not to: Pip.

Backed by Pip.cpp: use AI freely, keep control.

Your actual first five minutes

The difference between these tools is clearest in what you do first. Here is the real first move with each.

• llama.cpp: one command, llama cli, then -hf and a model's name on Hugging Face. Or llama serve, then open 127.0.0.1:8080 in a browser.
• Ollama: one command, ollama run and a model name. It downloads the model and starts chatting.
• LM Studio: install the app, type in its search box, press download, then chat. No terminal at any point.
• Pip: check first. Its site runs a compatibility check on your computer and names a model it has already tested for it.

Backed by llama.cpp: LLM inference in C/C++, llama.cpp HTTP server: REST endpoints and the built-in web UI, Ollama quickstart, Welcome to LM Studio Docs, Pip.cpp: use AI freely, keep control.