How to use Ollama, in short: download it from ollama.com/download, open a terminal and type ollama run followed by a model name. Ollama downloads the model and starts a chat that runs on your own computer. It is available for macOS, Windows and Linux, and local models need no account and no API key. (Ollama docs: Quickstart (opens in a new tab); Ollama: Download (opens in a new tab))

The appeal is privacy and control. A local model runs on your own hardware, so your prompts do not have to leave it.

Ollama also sells cloud models, and it can connect models to tools such as Claude Desktop, Claude Code and Codex. (Ollama docs: Quickstart (opens in a new tab); Ollama blog: Transparent pricing (opens in a new tab))

How to install Ollama

1. Download it

Go to ollama.com/download and pick macOS, Windows or Linux. On Linux the page gives a one-line install command that you paste into a terminal.

Linux install command

curl -fsSL https://ollama.com/install.sh | sh

2. Run the setup

Open the app, or type ollama in a terminal and follow the prompts. During setup you either sign in to use cloud models or choose a local model.

Tip: You can skip the sign-in if you only want local models. The docs say an account and API key are needed only for Ollama's cloud models.

Run your first model in Ollama

Ollama's own quickstart uses gemma4:e2b, a small version of Gemma 4. The download is about 7.2 GB, and Ollama recommends about 8 GB of free graphics memory, or unified memory on a Mac. With less, Ollama falls back to system memory and answers get slower. (Ollama docs: Quickstart (opens in a new tab))

Command 1: Download and chat

ollama run gemma4:e2b

3. Chat and leave

Type your question after the prompt and press Enter. Type /bye when you are done. To paste several lines as one message, start and end the block with three quote marks.

Command 2: Ask about an image

ollama run gemma4 "What's in this image? [path to your picture]"

Tip: Put a prompt right after the model name, as in the image example, to get a one-off answer without opening a chat.

The Ollama commands you will actually use

Ollama's CLI reference lists a short set of commands: ollama pull downloads a model without starting it, ollama ls lists downloaded models, ollama ps shows what is running, ollama stop unloads a model and ollama rm deletes it from disk. (Ollama docs: CLI reference (opens in a new tab))

Command 3: Clean up disk space

ollama ls
ollama rm [model name]

LLM Hardware's cheat sheet adds a useful check: ollama ps shows how much of a model sits on the graphics card. Below 100 percent means part of it spilled to the processor, and speed can drop to roughly 1 to 5 tokens per second. (LLM Hardware: Ollama cheat sheet (opens in a new tab))

Tip: Ollama only listens on your own computer by default. LLM Hardware notes that opening it to other devices means changing the OLLAMA_HOST setting, which exposes the service on your network, so only do that on a network you trust.

How much memory you need

The same guide estimates graphics memory at the common 4-bit setting: about 6.8 GB for an 8 billion parameter model, about 10.4 GB for 14 billion, about 21 GB for 32 billion and about 44 GB for a 70 billion model. Longer conversations need more on top of that. (LLM Hardware: Ollama cheat sheet (opens in a new tab))

For most laptops that means small models. They are fine for summaries, rewriting and simple questions, but they are not as sharp as the big cloud assistants. Start small and move up only if your machine keeps up.

Connect Ollama to Claude, Codex and other apps

Claude Desktop

In the Ollama app on a Mac, open Apps and turn on Claude. Ollama sets up the connection, and you choose the model in Ollama. Turn Claude off to restore your previous setup.

Command 4: Start a coding agent on a local model

ollama launch claude

Run that from your project folder. The quickstart says ollama launch also works with codex and opencode, and Ollama offers to install Claude Code or OpenCode if they are missing. Codex has to be installed first. (Ollama docs: Quickstart (opens in a new tab); Ollama docs: CLI reference (opens in a new tab))

Ollama free plan and cloud prices

Local use is free. For cloud models, Ollama's free plan includes a small monthly amount of usage on a set of starter models. Pro costs $20 a month with $60 of usage included, Max costs $100 a month with $300, and Team starts at $500 a month. Usage is billed per token, and unused amounts do not roll over. (Ollama blog: Transparent pricing (opens in a new tab))

Ollama is the easiest way to try local AI, but it is a terminal tool at heart. If you never want to type a command, it will feel bare. If you care about keeping your data on your own machine, it is the first thing to install.