AI Tools & ReviewsHow-to guide
How to Install Ollama on Windows and Run a Local AI Model
By Cognova AI, AI consultantsPublished
Download OllamaSetup.exe from ollama.com/download/windows, or run irm https://ollama.com/install.ps1 | iex in PowerShell. It installs without Administrator rights on Windows 10 22H2 or newer. Then open a new terminal and run ollama run gemma4:e2b to download and chat with a small model; type /bye to exit. Models stay on your computer, and Ollama serves a local API at http://localhost:11434. The latest stable release on November 20, 2026 was v0.40.16.
Ollama runs AI language models on your own computer instead of in the cloud. For law firms, accounting practices and other businesses wary of sending client documents to an AI service, that is the main appeal. "Install ollama" is searched about 1,900 times a month in the US, according to Semrush.
This guide follows Ollama's Windows docs, quickstart and FAQ as published on November 20, 2026. It is written for business owners and the IT person who will set it up. We did not install Ollama on our test PC for this article, so every command below comes straight from Ollama's documentation.
What you need
- Windows 10 22H2 or newer, Home or Pro
- At least 4 GB of free disk space for Ollama, plus several GB per model (gemma4:e2b is about 7.2 GB)
- An NVIDIA or AMD graphics card with current drivers for good speed (optional; Ollama can use system memory instead, more slowly)
- PowerShell or Command Prompt
Time needed
15–30 minutes, plus model download time
Step 1: Check your PC
Ollama's Windows docs require Windows 10 22H2 or newer, Home or Pro, and at least 4 GB of disk space for the program. Models are extra and can be tens to hundreds of GB. For NVIDIA cards, the Windows page asks for driver 551.61 or newer while the GPU page says 550 or newer, so install the latest driver. AMD Radeon cards work through ROCm or Vulkan drivers.
Step 2: Download and run the installer
Go to ollama.com/download/windows and download OllamaSetup.exe, or open PowerShell and run
irm https://ollama.com/install.ps1 | iex. Ollama's docs say the installer does not need Administrator rights and installs in your home folder. It puts the program in%LOCALAPPDATA%\Programs\Ollamaand adds it to your PATH. To install somewhere else, runOllamaSetup.exe /DIR="d:\some\location".Step 3: Move model storage if your C drive is small
Models are stored in
C:\Users\<you>\.ollama\modelsby default. To use another drive, quit Ollama from the taskbar, open Windows Settings, search for "environment variables", choose Edit environment variables for your account, and add a variable namedOLLAMA_MODELSpointing at the new folder. Then start Ollama again from the Start menu. Do this before downloading large models.Step 4: Run your first model
Open a new PowerShell window and run
ollama run gemma4:e2b. Ollama downloads the model, about 7.2 GB, then opens a chat prompt in the terminal. Ollama recommends 8 GB of free graphics memory for this model; with less, it uses system memory and responds more slowly. Type/byeto leave. To download a model without opening a chat, useollama pull gemma4:e2b.Step 5: Try a smaller or larger model
Ollama's model library lists sizes for each model family. For older or lower-powered PCs, try a small model such as
llama3.2(1b and 3b sizes) orgemma3(sizes from 270m). For a more capable model on a strong graphics card, try a larger gemma4 size. Larger models give better answers but need more memory. Runollama psto see which models are loaded and whether they are running on the GPU.Step 6: Call Ollama from other software
Ollama serves an API at http://localhost:11434, with an OpenAI-compatible version at http://localhost:11434/v1, and local requests need no API key. This is how tools such as n8n, document assistants or your own scripts can use a local model. The PowerShell example from Ollama's Windows docs is shown in the code block below. Run
ollama stopwith a model name to unload a model and free memory.Step 7: Keep it private and up to date
By default Ollama only listens on 127.0.0.1, so other computers on your network cannot reach it. Leave
OLLAMA_HOSTalone unless you deliberately want to share it, and then protect the network. Ollama also offers cloud models, which run on Ollama's servers rather than your PC, so stick to local models for confidential work. Ollama downloads updates itself; click Restart to update when asked.
Calling the local API from PowerShell
This example is copied from Ollama’s Windows documentation. It sends one prompt to a local model and prints the reply:
(Invoke-WebRequest -method POST -Body '{"model":"llama3.2", "prompt":"Why is the sky blue?", "stream": false}' -uri http://localhost:11434/api/generate ).Content | ConvertFrom-json
Pull the model first with ollama pull llama3.2, or change llama3.2 to a model you have already downloaded.
Useful settings
Ollama reads these from your user environment variables. Quit Ollama before changing them and restart it afterwards:
OLLAMA_MODELSmoves model storage to another folder or drive.OLLAMA_HOSTchanges the address Ollama listens on. The default, 127.0.0.1:11434, keeps it private to your PC.OLLAMA_CONTEXT_LENGTHchanges how much text a model can consider at once. The default is 4096 tokens.
How to uninstall
Open Windows Settings, go to Add or remove programs and uninstall Ollama. Ollama’s docs note that if you moved models with OLLAMA_MODELS, the uninstaller does not delete them, so remove that folder yourself if you no longer need it.
What we did and did not test
We did not install Ollama on our test PC for this guide. Every command comes from Ollama’s documentation and download page as published on November 20, 2026, and the version number comes from Ollama’s GitHub releases page, where v0.40.16 was marked Latest.
How Cognova AI can help
We help law firms, CPA practices and other professional firms decide when local AI makes sense, set up private models and connect them to document workflows. See AI for Law Firms, AI for CPA Firms, or our guide to self-hosting n8n.
Frequently asked questions
Is Ollama free to use on Windows?
Yes. Ollama's pricing page says running models on your own hardware is always unlimited, and the Windows app and command line cost nothing. Ollama sells paid plans for its cloud models, which run on Ollama's servers instead of your computer. For confidential client work, use local models so your documents stay on the machine.
Does Ollama keep my data on my computer?
When you run a local model, the model and your prompts stay on your PC, and by default Ollama only accepts connections from the same computer, at 127.0.0.1 port 11434. Cloud models are different, because those requests go to Ollama's servers. Also check any other app you connect to Ollama, since that app may send data elsewhere on its own.
What computer do I need to run Ollama well?
Ollama runs on Windows 10 22H2 or newer. Speed depends mostly on graphics memory. Ollama recommends 8 GB of free graphics memory for gemma4:e2b, its quickstart example; without that, it falls back to system memory and responds more slowly. A recent NVIDIA card with current drivers, or a supported AMD Radeon card, gives the best results. Smaller models run on more modest PCs.
Sources
Cite this article
Cognova AI. (2026, November 20). How to Install Ollama on Windows and Run a Local AI Model. Cognova AI. https://cognovaai.com/insights/how-to-install-ollama-on-windows
Permanent link: cognovaai.com/insights/how-to-install-ollama-on-windows. Quoting a passage? Select it and use “Copy quote” to get a link that jumps to that exact text.
