What is Ollama? Run AI models on your computer

In plain words

Ollama is not an AI model and not a company's chatbot. It is a free program you install on a Mac, Windows or Linux computer. Its job is to fetch AI models and run them for you.

Many companies publish models with open weights: the trained model itself, free to download. Normally, getting one running takes technical setup. With Ollama, one short command downloads a model and starts a chat, for example with Meta's Llama, Google's Gemma or Alibaba's Qwen.

Because the model runs on your own machine, your words never have to leave it, and it works offline. The trade-off is power: a laptop can only run smaller models, which are usually less capable than the biggest ones in the cloud. Developers also use Ollama to build their own apps on top of local models.

Everyday example

A music player for AI models

A music player app does not write songs. It downloads tracks and plays them on your phone, even in flight mode. Ollama does the same for AI models: it fetches a model once and plays it on your computer. The models are the songs, made by Meta, Google, Mistral and others.

Where the example breaks: A song plays the same on any phone. A model needs memory and a good chip: a big model may not fit on an ordinary laptop, or may answer very slowly.

How it actually works

Ollama wraps a few hard steps into simple commands such as ollama run.

Pull the model

Ollama downloads the model's weights from its online library. Most come already shrunk by quantization: each number stored with fewer bits, so the model takes a fraction of the memory it would otherwise need.

Run it on your chips

Under the hood Ollama uses open-source engines such as llama.cpp to do inference, the work of producing answers. It uses your GPU or Apple's M-series chips when it can, and falls back to the ordinary processor when it cannot.

Talk to it

You can chat in a terminal window or in Ollama's own desktop app. It also starts a small local server with an API, so other programs on your computer can send it questions, just as they would to an online AI service.

Optional cloud models

For models too big for your machine, Ollama also offers cloud models that run on its servers after you sign in. Local models are unaffected, and cloud features can be switched off for fully on-device use.

Made by

Ollama, a start-up founded by Jeffrey Morgan and Michael Chiang

First release

8 July 2023

Licence

Open source, MIT licence

Runs on

macOS, Windows and Linux

Ollama is one of several tools for running models locally. Hugging Face is the main place where open models are shared.

Questions people ask

Is Ollama free?

Yes, the software is free and open source under the MIT licence. Running local models costs nothing beyond your own electricity and hardware. Its optional cloud models are tied to an account and plan.

Does Ollama need the internet?

Only to download a model. After that, local models run fully offline on your computer, and your prompts stay on your machine.

Is Ollama the same as Llama?

No. Llama is a family of models made by Meta. Ollama is a program that runs models, Llama among many others, including Gemma, Qwen, Mistral and DeepSeek models. The names sound alike, but they are different things.

What computer do I need for Ollama?

Small models of a few billion parameters run on most recent laptops. Bigger models need much more memory and a strong GPU or an Apple M-series chip to answer at a usable speed.

Is Ollama safe?

The software runs on your own computer. One known risk: do not expose its local server to the internet. In January 2026 security researchers found many Ollama servers left open online, letting strangers use them.