Install Ollama on Linux, Mac, and Windows: Complete 2025 Guide
Ollama is an open-source tool that lets you download, manage, and run large language models locally — entirely on your own hardware, with no internet connection required after the initial model download. Think of it as a package manager for AI models: one command to pull, one command to run.
This guide walks you through a complete installation on macOS, Linux, and Windows, followed by running your first model and solving the most common issues newcomers hit.
Requirements
Ollama runs on any modern 64-bit machine. GPU acceleration is optional but dramatically improves speed.
| Model size | Min RAM | Recommended | Example models |
|---|---|---|---|
| 1–3B | 4 GB | 8 GB | Llama 3.2 3B, Phi-4 Mini |
| 7–8B | 8 GB | 16 GB | Llama 3.1 8B, Mistral 7B |
| 13–14B | 16 GB | 24 GB | Qwen3 14B, Phi-4 14B |
| 30–34B | 32 GB | 48 GB | Qwen3 32B, Llama 3.3 70B Q4 |
| 70B+ | 64 GB | 128 GB | Llama 3.3 70B FP16, Qwen3 72B |
Install on macOS
The easiest path on macOS is the official app, which bundles everything and adds a menu-bar icon. Alternatively, install via Homebrew if you prefer a CLI-only setup.
Option A — Official app (recommended)
Download Ollama.dmg from
ollama.com/download, open it, and drag Ollama to
your Applications folder. Launch it once — the server starts automatically on port 11434.
Option B — Homebrew
# Install Homebrew first if you don't have it
brew install ollama
# Start the server (runs in background)
brew services start ollamaInstall on Linux
Ollama provides a single install script that detects your distro, downloads the correct
binary, and registers a systemd service
so the server starts on boot.
curl -fsSL https://ollama.com/install.sh | shThe script handles Ubuntu, Debian, Fedora, Arch, and most other mainstream distros. After installation, the service starts automatically:
# Check service status
systemctl status ollama
# Enable NVIDIA GPU support (if you have one)
ollama serve # CUDA is auto-detected if drivers are installedsudo.
For NVIDIA GPU acceleration, install CUDA 12+ drivers before running Ollama.
AMD GPUs are supported via ROCm on Linux only.
Install on Windows
Ollama on Windows ships as a standard installer — no WSL or admin rights required. It runs as a background process and adds an icon to the system tray.
- 01 Go to ollama.com/download and click Download for Windows.
- 02 Run OllamaSetup.exe. The installer completes in under 30 seconds.
- 03 Ollama launches automatically and appears in the system tray (bottom-right).
- 04 Open PowerShell or Command Prompt — the ollama command is now available.
Run your first model
Once installed, open a terminal and pull and run Llama 3.2 — Meta's compact 3B model that runs comfortably on 8 GB of RAM:
# Pull the model and start an interactive chat
ollama run llama3.2
# Output (first run downloads the model ~2 GB):
pulling manifest... done
pulling 966de95ca8a6... 100% ▕████████████████▏ 2.0 GB
verifying sha256 digest... done
>>> Send a message (/? for help)Type your message and press Enter.
To exit, type /bye or press
Ctrl+D.
To list downloaded models and manage them:
ollama list # show downloaded models
ollama ps # show models currently loaded in memory
ollama rm llama3.2 # delete a modelTroubleshooting
ollama: command not found The binary isn't in your shell's PATH. On Linux/macOS, close and reopen your terminal after installation. If that doesn't work:
# Add to ~/.bashrc or ~/.zshrc
export PATH=$PATH:/usr/local/bin
source ~/.zshrcOn Windows, restart PowerShell or open a new terminal window.
bind: address already in use :11434 Another Ollama process (or something else) is holding port 11434. Kill the existing process and restart:
# Linux / macOS
pkill -f ollama
ollama serve
# Windows (PowerShell)
taskkill /F /IM ollama.exemodel crashes / killed after loading Your system ran out of RAM while loading the model weights. The OS OOM killer terminated the process. Switch to a smaller model or a more aggressive quantization:
# If llama3.2 (3B) crashes, try the 1B variant
ollama run llama3.2:1b
# Or a heavily quantized 7B (uses ~4.5 GB)
ollama run llama3.1:8b-instruct-q4_K_SWhat model should you run next?
Now that Ollama is running, the next step is picking a model that fits your hardware and use case. Our Model Picker guide compares Qwen3, Llama 3.3, Mistral, and more with real benchmarks.
Go to Model Picker →