How to Install Ollama on Windows (Step by Step)

Q: Where does Ollama store models on Windows?

Models are stored under %USERPROFILE%\.ollama\models by default. You can change the location by setting the OLLAMA_MODELS environment variable before starting the service.

Step 1: Download and Install

Go to ollama.com and click Download — Windows is detected automatically.
Run OllamaSetup.exe. No admin prompt is required; the installer places Ollama in your user profile.
When it finishes, open PowerShell or Windows Terminal and type:

ollama --version

You should see a version number. If PowerShell says "ollama is not recognized," close and reopen the terminal — the installer adds it to your PATH, and the new PATH only applies to new terminal windows.

Step 2: Run Your First Model

One command downloads, quantizes, and starts a model:

ollama run qwen3:8b

The first run downloads the model (about 5GB for the 8B Q4 file), which can take a few minutes. After that you're in an interactive chat:

>>> Write a Python function to check if a number is prime
def is_prime(n):
    if n < 2: return False
    for i in range(2, int(n ** 0.5) + 1):
        if n % i == 0: return False
    return True

Type /bye to exit. The model stays downloaded, so ollama run qwen3:8b starts instantly next time. For hardware-appropriate picks, see Top 10 GGUF Models by RAM — an 8GB laptop should start with qwen3:4b or phi4-mini instead.

Step 3: Manage Models

Command	What it does
`ollama list`	Show installed models
`ollama pull <model>`	Download a model without running it
`ollama run <model>`	Download (if needed) and chat
`ollama rm <model>`	Delete a model to free disk space
`ollama ps`	Show what's loaded in memory right now
`ollama show <model>`	Show model details (size, context, params)

💡 Disk space check: models are stored in %USERPROFILE%\.ollama\models. Each model is 1–10GB, so watch your free space — ollama rm anything you're not using. To store models on another drive, set the OLLAMA_MODELS environment variable to the new path, restart Ollama (the tray icon → Quit, then relaunch), and re-pull.

GPU vs CPU on Windows

NVIDIA GPU: used automatically if your driver supports CUDA (11.8+). You'll see GPU offload in the logs, and generation will be much faster.
AMD GPU: CPU inference on Windows for most setups; check the Ollama docs for the current support matrix.
CPU only: works fine for models up to ~8B with 16GB RAM — just slower (a few tokens/sec vs 30–60 on GPU).
Intel Arc: supported in recent Ollama releases; verify your driver version.

Want to know what your hardware can actually run? The local model ranking by hardware tier has the full picture, and Ollama vs llama.cpp explains when you'd want the lower-level runtime instead.

Step 4: Use the Local API

Ollama starts a local server at http://localhost:11434 with an OpenAI-compatible endpoint. Test it from PowerShell:

curl http://localhost:11434/v1/chat/completions ^
  -H "Content-Type: application/json" ^
  -d "{\"model\":\"qwen3:8b\",\"messages\":[{\"role\":\"user\",\"content\":\"Say hi in one word\"}]}"

This means any tool that supports a custom OpenAI base URL — VS Code extensions, Cline, Continue, custom scripts — can point at http://localhost:11434/v1 and use your private local model. That's the same API pattern our local RAG tutorial uses, and the same privacy architecture as Lawyer Assistant: your data never leaves the machine.

Troubleshooting Common Windows Issues

"ollama is not recognized" — close and reopen the terminal after install (PATH refresh).
Model downloads stall — check your connection and free disk space; resume with ollama pull <model> again.
Slow generation — confirm GPU usage (ollama ps shows "GPU" in the processor column); if CPU, pick a smaller model or lower quant.
Out of memory — your model is too big for RAM/VRAM; see Q4_K_M vs Q8_0 and drop to a smaller quant or size.
Firewall prompt — allow Ollama to listen on localhost; it's your own local server.

🚀 Beyond the command line

Prefer a GUI? GGUF Loader runs the same GGUF models with drag-and-drop on Windows, and LM Studio offers a visual alternative. For model discovery with direct download links, Local AI Zone keeps a daily-updated directory. And for the full beginner journey, our beginner's guide to local models covers the whole stack.

Frequently Asked Questions (FAQ)

How do I install Ollama on Windows?

Download the OllamaSetup.exe installer from ollama.com, run it (no admin prompt needed), then open PowerShell and run ollama run qwen3:8b (or any model name). The first run downloads the model automatically.

Does Ollama use my NVIDIA GPU on Windows?

Yes, automatically if you have a recent NVIDIA driver (CUDA 11.8+). Ollama uses your GPU when VRAM is sufficient and falls back to CPU when it isn't. AMD users on Windows currently get CPU inference; check the Ollama docs for the latest GPU support matrix.

Where does Ollama store models on Windows?

Models are stored under %USERPROFILE%\.ollama\models by default. You can change the location by setting the OLLAMA_MODELS environment variable before starting the service.

How do I use Ollama's API on Windows?

After installation, Ollama runs a local OpenAI-compatible server at http://localhost:11434. Send POST requests to /api/generate or /api/chat, or point any OpenAI-compatible tool at it by setting base_url to http://localhost:11434/v1.

Why is my Ollama model slow on Windows?

Check three things: is it using your GPU (look for GPU offload in the logs)? Is the model quantized (use a GGUF quant like Q4_K_M)? And is the model size right for your RAM/VRAM? An 8B model at Q4 needs ~5GB — see our RAM guide for the right size per machine.

How do I uninstall Ollama from Windows?

Use Windows Settings > Apps > Ollama > Uninstall, or uninstall via the Control Panel. Models stored in .ollama\models are kept unless you delete the folder manually.