Power in Your Hands: Running Local AI on Windows
Author: Net Equity Alliance (S. Ajideh) Research Initiative
Date: DEC, 2025
To run Ollama locally and offline on Windows, follow these steps. Once you have downloaded your models, Ollama requires zero internet connection to function, ensuring 100% privacy for your data.
1. Initial Setup (Requires Internet)
You must be online to install the software and download the model "weights" to your hard drive.
-
Download: Go to ollama.com and download the Windows installer.
-
Install: Run the file. It will install as a background service and appear in your system tray (bottom right corner). .exe
-
Download your Model: Open your Command Prompt (CMD) or PowerShell and run the following command to download your chosen model (e.g., ): qwen3.5:4b
DOS
ollama pull qwen3.5:4b
-
Note: You can pull as many models as your storage space allows.
2. Verify Local Storage
Once the download finishes, the model is saved permanently on your computer. You can verify this by running:
DOS
ollama list
This confirms the model is ready for offline use.
3. Going Fully Offline
Once the models are pulled, you can disconnect from the internet:
-
Disable Wi-Fi or unplug your Ethernet cable.
-
Test: Open a new Command Prompt window and run your model while offline:
DOS
ollama run qwen3.5:4b
-
The model will load instantly from your local disk and allow you to chat without any network connectivity.
4. Advanced: Hardening Privacy (Optional)
If you want to ensure the Ollama executable never attempts to make a network connection (e.g., for extreme security environments), you can add a Windows Firewall outbound rule:
-
Open PowerShell as Administrator.
-
Run this command to block Ollama's internet access: PowerShell New-NetFirewallRule -DisplayName "Block Ollama Outbound" -Direction Outbound -Program "$env:LOCALAPPDATA\Programs\Ollama\ollama.exe" -Action Block
Note: To download new models in the future, you will need to delete this rule or disable it.
Summary of Tips for Windows Users:
-
GPU Acceleration: If you have an NVIDIA graphics card, Ollama will automatically use it for much faster inference. If you have an integrated GPU (like Intel Iris or AMD Radeon), it will run on your CPU, which is slower but still completely functional.
-
Storage Path: If your C: drive is low on space, you can change where models are saved by setting a System Environment Variable named to the path of a different folder (e.g., ).OLLAMA_MODELSD:\OllamaModels
-
Exiting: When chatting in the CLI, always use to cleanly close the session and free up your system memory./bye