Ollama
Paying for AI chat, or worried about privacy? Ollama downloads free AI models and runs them on your own computer, so you can chat without a subscription.
How you use it: Download it for Mac, Windows or Linux, open a terminal, type one command, and start chatting.
In plain English
A language model is the kind of AI that reads and writes text. Ollama is the program that fetches those models and starts them for you. Its README says you can use recent open models locally or in the cloud, and the project description names Kimi, GLM, MiniMax, DeepSeek, gpt-oss, Qwen, and Gemma.
You drive it from a terminal, the text window on a computer. After you type ollama, the README says you are prompted to run a model or connect tools you may already use, including Claude Code, Codex, Copilot CLI, and OpenCode. It also documents an assistant called OpenClaw for WhatsApp, Telegram, Slack, and Discord.
Other programs on the same computer can talk to Ollama through a local web API. The README shows that address as http://localhost:11434 and lists Python and JavaScript libraries.
Practical ways to use it
Example 1
Chat with a model on your own machine
The README's own example is ollama run gemma4. You type a question in the terminal and the reply comes from the model you just downloaded.
Example 2
Try an open model without building a server
The project description points people at models such as DeepSeek, Qwen, and Gemma. The README sends you to ollama.com/library for the full list.
Example 3
Point a coding tool at a local model
The README shows ollama launch claude, and it names the same style of launch for Codex, Copilot CLI, and OpenCode.
Example 4
Let a personal assistant reach your chat apps
The README says ollama launch openclaw turns Ollama into a personal assistant across WhatsApp, Telegram, Slack, and Discord.
Who it is for
Some tech comfort. You do not have to write a program, but you do use a terminal or an installer. A friend who has installed an app from a download page can follow this.
Difficulty: Easier start. Mac, Windows, and Linux each have an install path in the README, including manual downloads for Mac and Windows.
Getting started
- 1
Pick the install that matches your computer
The README links a Mac disk image, a Windows setup file, and an install script for Mac and Linux. Choose the manual download if you would rather not run a script from the internet.
- 2
Install Ollama
If you use the script, the README shows curl -fsSL https://ollama.com/install.sh | sh on Mac and Linux, and irm https://ollama.com/install.ps1 | iex in Windows PowerShell. Docker is also listed, as the image ollama/ollama.
- 3
Open a terminal and start it
Type ollama and press Enter. The README says you will be prompted to run a model or connect an existing agent.
- 4
Chat with one model
Type ollama run gemma4. When the model is ready, type a normal sentence. Other model names live at ollama.com/library.
What you need first
- Accounts
- A model that runs on your computer does not need an account. The README also describes cloud use. Check ollama.com if that path asks you to sign in.
- Hardware
- The README does not list a minimum computer. Larger models need more memory. If the machine strains, pick a different model from the library.
- Cost
- Running a model on your own computer does not require a paid key. The README does not list cloud prices, so check the Ollama site before you pay.
Cautions
- An install script downloads and runs software. The README also offers manual Mac and Windows downloads.
- A chat stays on your machine only when you are not using a cloud model or a connected agent. Read the prompt before you paste private writing.
- The coding integrations can change project files. Learn the plain chat before you launch one.
Questions people ask
What does Ollama do?
It downloads open language models and runs them on your computer, and it can also connect you to models in Ollama's cloud. You start it from a terminal.
Who is Ollama for?
People who are comfortable with an installer or a short terminal command, and who want a model running on a computer they control.
What do I need first?
A Mac, Windows, or Linux computer, and the installer from the README. You do not need an account for a local model. The README does not publish a minimum memory size.
What should I watch out for?
Prefer the manual download if you do not want to run an install script. Do not connect a coding agent until you are happy with a normal chat, because those tools can edit files.