What you are installing
Ollama runs models and makes them available to compatible applications. It can use local models or hosted cloud models, so installing Ollama does not automatically make every request local. This walkthrough starts with a local model and a small test. Official documentation was checked November 14, 2026; we did not install or benchmark it for this guide.
1. Choose your operating system
- Windows: get the installer from Ollama's download page. It runs natively, without requiring WSL or an administrator install. Current requirements list Windows 10 22H2 or newer. Keep room for the app and separate model downloads. Windows setup and GPU details.
- Mac: download the DMG, open it, and drag Ollama into Applications. Launch it there. Current support is macOS 14 or newer, with Apple Silicon acceleration and CPU-only operation on Intel Macs. Mac installation.
- Linux: the official installer command is below. It downloads and executes a shell script, so read the linked instructions and script before running it on a machine you manage. Linux installation and service setup.
curl -fsSL https://ollama.com/install.sh | sh
A fast GPU helps, but there is no single memory requirement for every model. Start smaller than your maximum available memory: the operating system and the model's working context need room too. If a large model is painfully slow, a smaller one may be more useful than a hardware upgrade.
2. Make the first response work
- Open the Ollama app, or enter
ollamain a new terminal after installation. - Follow the setup prompts and choose a local model. A cloud-model sign-in is not needed for local inference. Allow the download to complete.
- Send a harmless test: Write a three-item checklist for tidying a desk.
- Check that the selected model is the local one you intended. An answer verifies the basic route works; it does not prove every future integration is private or accurate.
Current official quickstart. When this works, add a graphical workspace such as AnythingLLM or Open WebUI only if you need its extra features. Diagnose one layer at a time.
3. Know where requests go
Local inference runs on your hardware; cloud inference sends prompts to a provider and has its own usage limits. Ollama normally binds its local service to 127.0.0.1:11434. Keep that default for personal use rather than exposing the service to a network. For a strictly local configuration, the documented OLLAMA_NO_CLOUD=1 setting disables cloud features after restart. Local privacy and configuration FAQ.
Fixes and maintenance
- Command not found: reopen the terminal after installing and confirm the app is running.
- Another app cannot connect: test Ollama itself first. A Docker app has its own network; its
localhostmay not mean your computer. - Unexpected speed:
ollama psshows loaded models and CPU/GPU placement. Check the official driver guidance before changing anything. - Update: macOS and Windows provide Restart to update; Linux uses the installer again. Before major changes, keep your model names, custom configurations and the separate application's chat backups. Ollama model files alone are not a backup of every connected app's conversations.