Yes. Ollama is free and open source. No subscription or API key required.

Download Ollama for Windows — Official Installer & Setup Guide 2026

Q: Does Ollama work offline?

Yes. After downloading the installer and pulling a model, Ollama works completely offline. Prompts never leave your machine.

Q: Do I need a GPU to run Ollama?

No, but a GPU significantly improves speed. NVIDIA CUDA and AMD DirectML are both supported on Windows.

This is an unofficial Windows guide site for Ollama. The official Ollama project is at ollama.com. Download links are partner links.

What is Ollama?

Run AI models locally on Windows — free and private

Ollama is a free, open-source tool that lets you run large language models (LLMs) like Llama 3, Mistral and Qwen 2.5 directly on your Windows PC. No cloud subscription, no API key, no data leaving your machine. After the initial model download, everything works completely offline.

On Windows, Ollama installs as a background service that exposes a local API at localhost:11434. You interact with it via the command line or through any compatible chat UI.

Quick start

Get Ollama running on Windows in 4 steps

1

Download OllamaSetup.exe

Use the download button above. The installer is ~5 MB and has no dependencies. Works on Windows 10 and 11 (64-bit). Verify the file is genuine before running.
2

Run the installer

Double-click the .exe and follow the one-page wizard. Ollama installs in under a minute and starts a background service automatically. See the full install guide for tips on GPU detection and firewall prompts.
3

Pull your first model

cmd.exe or PowerShell
C:\> ollama pull llama3
pulling manifest...
pulling 8934d96d3f08... 100% ████████████ 4.7 GB
success

The first pull downloads the model weights (~4–8 GB depending on the model). After that it runs fully offline.
4

Start chatting

cmd.exe
C:\> ollama run llama3
>>> Send a message (/? for help)
>>> Hello! What can you do?
I can help with writing, coding, analysis, Q&A...

All guides

Everything you need for Ollama on Windows

Install on Windows

Step-by-step install guide with GPU tips and common pitfalls.

→ install guide

GPU Acceleration

Enable NVIDIA CUDA or AMD DirectML for 5-10x faster inference.

→ GPU guide

Models Hub

Llama 3, Mistral, Qwen 2.5, Gemma 2 — copy the pull command.

→ browse models

Offline & Privacy

Is Ollama fully offline? Where is data stored? How to block network access.

→ privacy guide

Troubleshooting

Fix CLI not recognized, port conflicts, slow downloads and GPU issues.

→ fix issues

FAQ

All common Ollama Windows questions answered in one place.

→ see FAQ

System requirements

Minimum requirements for Ollama on Windows

Component	Minimum	Recommended
OS	Windows 10 64-bit	Windows 11 64-bit
RAM	8 GB	16 GB or more
Disk	10 GB free (for one model)	50+ GB SSD
CPU	Any x64 with AVX2	Modern multi-core (Ryzen 5/i5+)
GPU	Optional	NVIDIA (8+ GB VRAM) or AMD (DirectML)

Without a GPU, Ollama runs models on the CPU — functional but slower. A GPU with 8+ GB VRAM dramatically improves speed. See GPU Acceleration.

FAQ

Frequently asked questions

Is Ollama free?

Yes. Ollama is free and open source (MIT license). The models it runs are also free to download and use. There is no subscription, API key or usage limit.

Does Ollama work offline?

Yes, completely. After downloading the installer and pulling a model, Ollama works with no internet connection. Prompts and responses never leave your machine. See Offline & Privacy guide.

Which models can I run on Ollama for Windows?

Ollama supports Llama 3, Mistral, Qwen 2.5, Gemma 2, Phi-4, DeepSeek, CodeLlama and many more. Browse the full list at Models Hub or at ollama.com/library.

Do I need a GPU to run Ollama?

No. Ollama runs on CPU only, but it is significantly slower without a GPU. For usable response speeds, a mid-range NVIDIA GPU with 8+ GB VRAM is recommended. AMD GPUs are supported via DirectML. See GPU Acceleration.

Is 'ollama' not recognized after installing?

This usually means the terminal was open before Ollama was installed, so PATH was not refreshed. Close all terminal windows and open a new one. If it still fails, see the not recognized fix guide.

Related guides

browse available models