# How to use Ollama: run AI models on your own computer

> Ollama is a free, open source program that downloads open AI models and runs them on your own computer. You pull a model with one command, chat with it in a terminal and let other apps call it through a local address. Local models need no account, and it runs on Windows, macOS and Linux.

Mindket editors · Published 2026-11-06 · Updated 2026-11-06 · https://mindket.com/learn/tools/ollama

## In short

- Two commands get a model talking: `ollama pull` downloads it and `ollama run` opens a chat.
- Local models need no account and nothing leaves your computer. Cloud models are the exception.
- A graphics card helps but is not required, and `ollama ps` shows whether it is doing the work.
- The same local server feeds your own programs and tools such as Claude Code.

## What Ollama is

Ollama is an open source program, under the MIT license according to [its GitHub page](https://github.com/ollama/ollama), that downloads open AI models and runs them on your own machine. An open model is one whose weights you can download, the large files that hold what the model learned. Ollama is built on llama.cpp, the open source engine that also runs GGUF model files inside LM Studio.

It runs on Windows, macOS and Linux, from a terminal or from a desktop app. Behind both sits a small server on your computer at `127.0.0.1:11434`, and the `ollama` command talks to it. Running local models is free and needs no account. [The FAQ](https://docs.ollama.com/faq) says Ollama does not see your prompts or data for them. Cloud models are the exception: they run on ollama.com and need a sign-in.

## What Ollama is good for

- **A private chat with an open model**: `ollama run gemma4:e2b`, the model [the quickstart](https://docs.ollama.com/quickstart) starts with, opens a chat in the terminal.
- **One question, one answer**: put a prompt in quotes after the model name and Ollama prints a single answer, which suits scripts.
- **Reading pictures**: with a model that accepts images, write the file path inside the prompt and ask what the picture shows.
- **A local API for your own programs**: the server at `127.0.0.1:11434` answers requests from curl, Python or any app that can call it.
- **Text turned into numbers**: `ollama run embeddinggemma "Hello world"` prints the list of numbers a search tool compares.
- **Local models for coding agents**: `ollama launch` sets up OpenCode, Claude Code, Codex, VS Code or Droid against your own server, as [the CLI reference](https://docs.ollama.com/cli) lists.
- **Models from Hugging Face**: `ollama run hf.co/{username}/{repository}` runs a GGUF repository from the Hub, as [Hugging Face's Ollama page](https://huggingface.co/docs/hub/ollama) describes.

## When to use Ollama, and when to pick another tool

Ollama fits when you want a model on your own hardware. When you want the strongest model available, a hosted assistant wins, because the largest models do not fit on a typical computer.

| Your job | Use Ollama? | Or try |
| --- | --- | --- |
| Chat with an open model on your own computer, with no account | Yes, it needs one pull and one run | LM Studio, if you prefer an app with its own window |
| Keep prompts and files off a cloud service | Yes, for local models | Nothing else needed |
| Get the best answers from the largest models | Not on a modest machine, memory runs out first | ChatGPT, Claude, Gemini or DeepSeek |
| Try a model you found on Hugging Face | Yes, when the repository holds GGUF files | Hugging Face, to browse models and read their cards |
| Give your own program a model to call | Yes, through the local server | Hugging Face inference providers, for models too big to run at home |
| Run a coding agent on a local model | Yes, `ollama launch` sets it up | Claude Code with Anthropic's own models |

## How to use Ollama, step by step

Install Ollama first with the one-line command in [lesson 1](https://mindket.com/learn/local-ai/ollama-tutorial). The five pictures below show what the commands printed on a Windows PC, in order.

1. Open a new terminal and run `ollama -v`. A line that starts with ollama version means Ollama is installed and the terminal can find the command.
   ![A dark terminal frame. The command ollama -v is ringed in gold, and the line under it reads: ollama version is 0.40.14.](https://mindket.com/blog/screens/ollama-tutorial-1-version.webp?v=0ca8bfd6f412)
2. Run `ollama pull gemma4:e2b` to download a first model, about 7.2 GB, then `ollama ls` to list the models on your disk. The computer in the pictures already held `qwen2.5:7b`, so they show that model. Type `gemma4:e2b` wherever they show it.
   ![The command ollama ls, ringed in gold, above a table of two models: gemma4:26b at 18 GB and qwen2.5:7b at 4.7 GB.](https://mindket.com/blog/screens/ollama-tutorial-2-list-models.webp?v=b4613f77374e)
3. Ask one question without opening a chat: `ollama run gemma4:e2b "Why is the sky blue?"`. The first answer waits while Ollama loads the model. Leave out the quoted prompt and you get a `>>>` chat prompt instead, where `/bye` ends the chat.
   ![The command ollama run qwen2.5:7b with the prompt Why is the sky blue?, ringed in gold, and the reply about Rayleigh scattering printed under it.](https://mindket.com/blog/screens/ollama-tutorial-3-run-a-prompt.webp?v=feb2f47cefaf)
4. Run `ollama ps` straight afterwards. It lists the models held in memory. The PROCESSOR column says where each one runs, and here it reads `96%/4% CPU/GPU`, so most of the model sits in system RAM. The last column counts down to when Ollama unloads it.
   ![The output of ollama ps: one row for qwen2.5:7b at 5.2 GB, with 96%/4% CPU/GPU under PROCESSOR ringed in gold.](https://mindket.com/blog/screens/ollama-tutorial-4-ps.webp?v=553328af61ab)
5. Free the memory with `ollama stop gemma4:e2b`, which prints nothing. A second `ollama ps` then shows only its header row, so no model is loaded.
   ![Two commands ringed in gold: ollama stop qwen2.5:7b, which prints nothing, and ollama ps, which prints only its header row.](https://mindket.com/blog/screens/ollama-tutorial-5-stop.webp?v=06f3ea3c27e7)

*Real output of Ollama 0.40.14 on Windows, shown in a terminal frame, taken on 6 November 2026.*

> The tutorials at the top of this page show these steps on screen. The Windows and macOS installs also include a desktop app, which the Sam Witteveen video covers.

## An example to try

These four lines download the quickstart's first model, ask it one question, show where it sits in memory and unload it. The download is large, so start it on a connection you can leave running.

````bash
ollama pull gemma4:e2b
ollama run gemma4:e2b "Why is the sky blue?"
ollama ps
ollama stop gemma4:e2b
````

- The answer is a short explanation in plain text. Which model wrote it decides the wording, so yours will differ from the picture.
- The first `ollama ps` lists the model with a PROCESSOR value such as `100% GPU`, or a split like `48%/52% CPU/GPU` when part of it did not fit on the card.
- The second `ollama ps`, after the stop, prints only its header row.

## Tips for better results

- Read the size column in Ollama's library before you pull. [The gemma4 page](https://ollama.com/library/gemma4) lists tags from `e2b` up to `31b`, and the larger ones need far more memory.
- Start with a small tag and move up while `ollama ps` still shows `100% GPU`.
- Closing the chat does not free memory. Ollama keeps a model loaded for 5 minutes, so run `ollama stop` before a game or a video export.
- To move the model files to a bigger drive, set `OLLAMA_MODELS` to the new folder and restart Ollama.
- For a long document, raise the context with `OLLAMA_CONTEXT_LENGTH` or the slider in the app settings.

## Limits to know

- The quickstart's first model wants about 8 GB of free VRAM. With less, part of it runs from system RAM and answers slow down. [The hardware lesson](https://mindket.com/learn/local-ai/local-llm-hardware) works out what fits your machine.
- Cloud models are not local. They need `ollama signin`, they run on Ollama's servers and the FAQ says prompts sent to them are processed there.
- Local requests need no API key. Setting `OLLAMA_HOST=0.0.0.0:11434` with no firewall rule lets anyone who reaches the port use your models.
- Deleting a model with `ollama rm` is final. It has to be pulled again, download and all.

## YouTube tutorials

- [Learn Ollama in 15 Minutes - Run LLM Models Locally for FREE](https://www.youtube.com/watch?v=UtSSMs6ObqY) (Tech With Tim)
- [How to Use Ollama (Free Local AI)](https://www.youtube.com/watch?v=btMoPRHz8eM) (Kevin Stratvert)
- [Ollama Gets a New App](https://www.youtube.com/watch?v=xN4Q98CbqqE) (Sam Witteveen)
- [What is Ollama? Running Local LLMs Made Simple](https://www.youtube.com/watch?v=5RIOQuHOihY) (IBM Technology)

## Learn more about Ollama

- [Ollama tutorial: run AI models on your own computer](https://mindket.com/learn/local-ai/ollama-tutorial)
- [Ollama API: call a local model from your code](https://mindket.com/learn/local-ai/ollama-api)
- [Local LLM hardware requirements: how much VRAM a model needs](https://mindket.com/learn/local-ai/local-llm-hardware)

## Frequently asked questions

### Is Ollama free?

Yes for local models. The program is open source under the MIT license, and running a model on your own computer costs nothing and needs no account. Cloud models run on ollama.com and need a sign-in, so check [Ollama's own pages](https://ollama.com) for their terms.

### Does Ollama work offline?

Yes, once a model is on disk. Pulling a model needs the internet, but a local model answers with the network cable out. Cloud models are the exception, since they run on ollama.com.

### Do I need a graphics card to run Ollama?

No. A graphics card helps, but without enough video memory Ollama keeps part of the model in system RAM and answers arrive more slowly. Run `ollama ps` and read the PROCESSOR column to see the split.

### Where does Ollama store its models?

On macOS in `~/.ollama/models`, on Linux in `/usr/share/ollama/.ollama/models` and on Windows in `C:\Users\%username%\.ollama\models`. Set `OLLAMA_MODELS` to a roomier folder and restart Ollama to move them.

### How do I stop Ollama using my memory?

Run `ollama stop` with the model name. Ollama otherwise keeps a model loaded for 5 minutes after its last request, and `OLLAMA_KEEP_ALIVE` changes that wait for every model.

### Can Ollama run a model from Hugging Face?

Yes, when the repository holds GGUF files. Run `ollama run hf.co/{username}/{repository}` and add a tag such as `:Q8_0` to pick a quantization. Without a tag, Ollama takes `Q4_K_M` when the repository has it.

### What is the difference between Ollama and LM Studio?

Both run open models on your own computer, and both can run GGUF files through llama.cpp. Ollama is driven by commands and a small local server, with an app on Windows and macOS. LM Studio is an app with its own window. [The LM Studio page](https://mindket.com/learn/tools/lm-studio) covers it.

## Sources

- [Ollama Docs: Quickstart](https://docs.ollama.com/quickstart) (read 2026-11-05)
- [Ollama Docs: CLI reference](https://docs.ollama.com/cli) (read 2026-11-05)
- [Ollama Docs: FAQ](https://docs.ollama.com/faq) (read 2026-11-05)
- [GitHub: ollama/ollama](https://github.com/ollama/ollama) (read 2026-11-06)
- [Ollama library: gemma4](https://ollama.com/library/gemma4) (read 2026-11-05)
- [Hugging Face Docs: Use Ollama with any GGUF model on the Hub](https://huggingface.co/docs/hub/ollama) (read 2026-11-05)

[Start the lesson](https://mindket.com/learn/local-ai/ollama-tutorial)