[Local LLM 02] Install Ollama on Windows 11 and Run a Model

Building a Local LLM · 02/30 · Introduction

Full series and learning sequence · Practice code ZIP

For a first response on Windows 11, begin with Ollama and one small chat model. Python and Docker can wait until the exercises that need them. Install the runtime, download the example model and verify that the same short input produces a response.

Check the PC specifications listed in Part 1 and prepare space to store the model. An Internet connection is required while downloading the installation file and model.

Install Ollama for Windows

Ollama Obtain the installation file from Windows Download. After confirming the domain and publisher, install it and run Ollama from the Start menu. Also open PowerShell in a new window. The changed PATH may not be reflected in terminals that are already open.

ollama --version
Get-Command ollama
Invoke-RestMethod http://127.0.0.1:11434/api/version

If there is a version entry in the response of the last command, the server is receiving requests. If you receive a message that the command cannot be found, check the installation path in a new terminal. If the connection is refused, first check if the Ollama app is running. If you run ollama serve again while the tray app is already running, it may crash if it tries to use the same port.

Download and run your first model

The example uses qwen3:4b-instruct. The download and memory burden is relatively small, and since it is a modification for general responses, it is easy to use for first practice. I didn’t choose this model based on its performance. Comparisons with other candidates are conducted by asking the same questions in Parts 3 and 18.

ollama pull qwen3:4b-instruct
ollama run qwen3:4b-instruct

When prompted, paste the virtual inspection notes you prepared earlier. The first answer may take a little longer to load the model into memory. When exiting, enter /bye. After you have finished downloading your model, you can run regular inference on your local model without a network.

Disk list and run list

ollama list
ollama show qwen3:4b-instruct
ollama ps

list shows the model saved on disk, and ps shows the model currently in memory. Note the PROCESSOR and CONTEXT in the run list. With this output, you can check whether GPU is being used and the execution context. If the list is empty after you stop talking, it is possible that the model has expired and has been removed.

File size and running memory usage vary. Even if the download is 2.5GB, it requires more space for the context cache, buffers, and Windows to use. If your disk usage is skyrocketing and your entire PC is slowing down while you’re answering, check if you’re using the page file due to insufficient RAM.

Check for installation errors

symptoms	check first	next action
ollama not found	Get-Command ollama	Check if new terminal/installation is complete
11434 Connection refused	Whether to run tray app	After starting the app, recheck the version API
Model not found	Tag spelling·ollama list	Check the official tag page and pull
Response comes but very slow	ollama ps·RAM usage	Comparison with smaller models and shorter context

When a problem arises, we do not first turn off the antivirus and firewall. Check whether it responds to the version request, whether the model has finished downloading, and whether it responds to the same input. If you separate these three, it will be easier to find whether you are stuck in the installation or inference stage.

reference material

• Windows installation/log location
• Qwen3 official distribution tag
• Ollama CLI

Documentation and model tags checked: October 18, 2026. Primary lab environment: Windows 11 x64.

---