Open WebUI Windows: Setup Guide for Desktop, PowerShell & GPU

Yes, Open WebUI runs on Windows. The easiest path is the Desktop app. If that doesn't work, PowerShell or Docker are fallbacks. This guide covers all three methods plus Windows-specific GPU setup and troubleshooting.

Open WebUI is a self-hosted AI chat interface with 150.9K GitHub stars. It runs local language models (Ollama, Llama, Mistral) without cloud vendor lock-in. Windows users have three ways to get it running. We'll walk you through each one, then dive into the Windows-specific stumbling blocks that trip people up.

Can Open WebUI Actually Run on Windows?

Yes. Open WebUI works natively on Windows 10 and Windows 11. You do not need Linux, a virtual machine, or special developer tools. The team released a standalone Desktop app (2.7K GitHub stars as of 2026) that runs like any other Windows program. No Docker. No terminal required.

Three installation paths exist: Desktop app (recommended for most), PowerShell/pip (for Python developers), or Docker (for production setups). The Desktop app is the simplest. It handles model downloads, API management, and web server startup in one click.

The app is still early-stage (v0.0.20 in early alpha), which means features ship fast but expect occasional instability. That's fine for evaluation. If you need production uptime or team features, managed hosting is the next step.

System Requirements for Windows

Open WebUI itself is lightweight. The bottleneck is the language models you run inside it.

For the application alone: 4GB RAM, 20GB disk, any modern CPU. Windows 10 or Windows 11. Antivirus software is optional but often causes false positives (we'll cover that below).

For running local LLMs alongside Open WebUI: 16GB RAM minimum. Larger models (70B parameters) need 24GB+. This is system RAM, not VRAM. If you have an NVIDIA RTX card (any series: 3060, 4070, RTX 4000, etc.), you can offload model computations to GPU memory, which frees up system RAM and speeds up inference dramatically.

Disk space depends on which models you download. A typical 7B model takes 4GB. A 13B model takes 8GB. A 70B model takes 40GB. SSD storage is faster than HDD, but HDD works.

Windows 10 vs Windows 11 difference: Both work identically. Windows 11 has better GPU driver support in some cases, but not a blocker either way. If you're on Windows 10, you're fine.

GPU support on Windows is NVIDIA-only (as of 2026). AMD cards are not supported by Open WebUI or the underlying Ollama engine. Intel integrated graphics do not work. If you do not have an NVIDIA card, you'll run models on CPU, which is slower but still functional.

Method 1: Desktop App (Recommended for Most Users)

This is the path for you if you want the fastest setup with no terminal or Docker knowledge.

Step 1: Download the Installer

Visit https://github.com/open-webui/desktop and look for the releases section. Download the .exe file for Windows (not the Linux or macOS versions). File size is roughly 200MB. Save it anywhere on your computer.

Step 2: Run the Installer

Double-click the .exe file. Windows Defender or your antivirus may show a warning: "Windows protected your PC." This is normal for unsigned executables. Click "More info" and then "Run anyway." The SmartScreen filter is just being cautious; the Open WebUI team is legitimate.

The installer will extract files and start the application. This takes 30 seconds to 2 minutes depending on your disk speed.

Step 3: Set a Password

After install, Open WebUI will open in your browser at localhost:8080. You'll see a login screen. Create an admin account: username and password. This password encrypts your settings. Do not forget it.

Step 4: Download a Model

After login, you'll see a models dropdown at the top left. Click it and select "Pull a model from Ollama." A dialog opens with popular models: Mistral (7B, 4GB), Llama 2 (7B, 4GB), Neural Chat (7B, 4GB). Pick one. The download starts immediately.

First run takes 5-15 minutes depending on your internet and disk speed. Do not close the app. You'll see a progress bar.

Step 5: Chat

Once the model finishes downloading, type a prompt in the chat box. Hit Enter. The model runs locally on your machine. No cloud, no API calls, no data leaving Windows.

Windows-Specific Gotchas

SmartScreen warnings are harmless. Microsoft flags unsigned apps. Open WebUI's desktop binary is not code-signed, so Windows shows the warning on first run. This is a certificate cost issue, not a security issue.

Windows Defender may quarantine the app on first launch. If this happens: open Windows Defender (search "Virus & threat protection"). Look for quarantined items. Restore Open WebUI. Then add it to the exclusion list so Defender doesn't bother you again.

Third-party antivirus (Norton, McAfee, Kaspersky) sometimes treats the app as suspicious. This is overcautious. Add C:\Users\[YourUsername]\AppData\Local\Programs\Open WebUI to your antivirus exclusion list (or wherever you installed it).

Some users report port 8080 already in use on their machine (usually Docker, Kubernetes, or another service using it). The app will fail to start. We'll cover how to fix this in the troubleshooting section.

Method 2: PowerShell/Pip (For Python Developers)

Use this path if the Desktop app fails, or if you want to control the installation via command line.

Prerequisites

You need Python 3.11 installed. Visit https://www.python.org/downloads/ to download Python 3.11 and run the installer. Important: check the box "Add Python to PATH" during install. This lets you run Python from PowerShell.

Verify it worked: open PowerShell (search "PowerShell" in the Start menu) and run:

python --version

You should see "Python 3.11.x" or later. If you get "command not found," Python is not in PATH. Reinstall and check the PATH box.

Installation Command

In PowerShell, run:

$env:DATA_DIR="C:\open-webui\data"; uvx --python 3.11 open-webui@latest serve

This command does three things: (1) sets a data directory on your C: drive to store models and settings, (2) downloads the latest Open WebUI via uvx (a Python package runner), and (3) starts the server.

First run takes 2-5 minutes as it downloads dependencies. You'll see a lot of text in PowerShell. When you see "Application startup complete" or "Uvicorn running," the app is ready. Open your browser to localhost:8000 to begin.

Why This Method

It gives you version control. You can specify open-webui@0.1.0 instead of @latest to pin a specific release. It also lets you customize the startup command: add --port 9090 to run on a different port, or --host 0.0.0.0 to allow network access from other machines.

Stopping the Server

Press Ctrl+C in PowerShell. The server shuts down. Your models are saved in C:\open-webui\data, so you can restart without re-downloading.

Method 3: Docker (For Advanced Users)

Use this if you need production-grade isolation, version pinning, or you're already running Docker Desktop.

Prerequisites

Install Docker Desktop for Windows from https://www.docker.com/products/docker-desktop. Run the installer, restart your computer (Windows needs a reboot for Docker to set up its VM layer), then verify Docker works:

docker --version

You should see "Docker version 25.x" or similar. If Docker commands fail, restart your computer and try again.

Run the Container

In PowerShell, run:

docker run -d -p 8080:8080 -v open-webui:/app/backend/data ghcr.io/open-webui/open-webui:latest

This pulls the Open WebUI image from GitHub Container Registry and starts it in the background. The -v open-webui:/app/backend/data line creates a named volume on your Windows machine to persist models and settings across container restarts.

Open your browser to localhost:8080 to see the login screen.

Docker on Windows Specifics

Docker Desktop for Windows uses Hyper-V virtualization. It reserves 2GB RAM and 2 CPU cores by default. If your machine has less than 8GB total RAM, Docker will slow down your entire system. Check Docker Desktop settings (right-click the Docker icon in the system tray, select Settings) and tune CPU/RAM allocation if needed.

Volumes on Windows work but are slower than on Linux because they're shared via SMB file sharing. If model loading is sluggish, this is why. There's no workaround; Docker on Windows is inherently slower than native for file-heavy workloads.

For production setups or team usage, consider managed hosting on a Linux server instead. That's where Docker shines. For your Windows laptop, the Desktop app (Method 1) or PowerShell (Method 2) is faster and simpler.

GPU Setup on Windows: NVIDIA Optimization

If you have an NVIDIA RTX card and want to offload model computation to GPU (much faster), follow these steps.

Step 1: Install NVIDIA CUDA Toolkit

Visit https://developer.nvidia.com/cuda-downloads to get started. Select "Windows," your architecture (x86_64 if unsure), and your Windows version. Download the installer (the "local" option is about 3GB; network option is smaller but slower).

Run the installer. It will probe your system for NVIDIA drivers. If drivers are missing, the installer offers to install them. Let it. CUDA toolkit installation takes 5-10 minutes.

Verify success by opening PowerShell and running:

nvcc --version

You should see "Cuda compilation tools, release 12.x" or similar. If not, CUDA is not in your PATH. Add C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.4\bin to your Windows environment PATH, then restart PowerShell.

Step 2: Install cuDNN (CUDA Deep Neural Network library)

Visit https://developer.nvidia.com/cudnn for the download. You'll need an NVIDIA developer account (free). Download the Windows version of cuDNN for your CUDA version (if you installed CUDA 12.4, download cuDNN 12.x).

Extract the ZIP file. Copy the bin, include, and lib folders into your CUDA toolkit installation directory: C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.4\. Windows will ask to merge folders; confirm.

Step 3: Verify GPU is Detected

In PowerShell, run the Python command to check PyTorch GPU support:

python -c "import torch; print(torch.cuda.is_available()); print(torch.cuda.get_device_name())"

If you get True and your GPU model (e.g., "NVIDIA GeForce RTX 4070"), GPU is ready.

If you get False, either CUDA is not installed, or PyTorch doesn't recognize your GPU. Common fix: use pip to reinstall PyTorch with CUDA 12.1 support, then try again.

Step 4: Tell Open WebUI to Use GPU

If you're using the Desktop app: after starting the app, open the settings (gear icon, top right). Look for "GPU" or "CUDA" settings. Toggle on GPU acceleration. Restart the app.

If you're using PowerShell (Method 2), set the GPU_DEVICE_ID environment variable before running the server:

$env:GPU_DEVICE_ID="0"; $env:DATA_DIR="C:\open-webui\data"; uvx --python 3.11 open-webui@latest serve

If you're using Docker (Method 3), add the --gpus all flag:

docker run -d --gpus all -p 8080:8080 -v open-webui:/app/backend/data ghcr.io/open-webui/open-webui:latest

Note: Docker GPU support on Windows is limited. It only works with WSL 2 backend and requires additional setup. Native methods (Desktop or PowerShell) are much simpler for Windows.

Verifying GPU is Actually Being Used

After starting Open WebUI, open the settings and look for a GPU memory gauge or model load time. Models should load in seconds with GPU, minutes without. If load times are the same as before, GPU is not being used.

Alternative check: open Task Manager (Ctrl+Shift+Esc), go to the Performance tab, and look at GPU usage while a model is running. If GPU utilization is >0%, your card is working.

Troubleshooting Windows-Specific Issues

"Windows protected your PC" -- SmartScreen Warning

Desktop app installer shows this because it's not code-signed. Click "More info" and "Run anyway." It's safe. Microsoft just requires a certificate (which costs money annually). Open WebUI chose not to pay.

"Port 8080 already in use"

Open PowerShell as Administrator and run:

netstat -ano | findstr :8080

You'll see a PID (process ID) of whatever is using port 8080. Find that PID in Task Manager and kill it. Or, start Open WebUI on a different port by adding --port 9090 in PowerShell:

uvx --python 3.11 open-webui@latest serve --port 9090

Then visit your browser at localhost:9090 to access it.

"Visual C++ Redistributable not found"

Some Windows machines lack the MSVC runtime libraries. Get the latest version from Microsoft at https://learn.microsoft.com/en-us/cpp/windows/latest-supported-vc-redist?view=msvc-170. Install the latest "Visual C++ Redistributable for Visual Studio" for x64. Then restart Open WebUI.

GPU not Detected

Run nvidia-smi in PowerShell. If it returns device info (GPU, memory, driver version), NVIDIA drivers are installed. If it says "command not found," drivers are missing. Visit the NVIDIA driver download page, select your GPU model and Windows version, then install and restart your system.

If nvidia-smi works but Open WebUI still doesn't see GPU, CUDA toolkit is not installed. Follow Step 1 in the GPU Setup section above.

Antivirus Quarantines the App

Open Windows Defender (search "Virus & threat protection" in Start menu). Go to "Virus & threat protection settings." Scroll to "Exclusions" and add the Open WebUI installation folder. For Desktop app, this is usually C:\Users\[YourUsername]\AppData\Local\Programs\Open WebUI. For PowerShell installations, it's C:\open-webui.

If you use third-party antivirus, consult its documentation. Most have a similar "Add exception" or "Whitelist" feature.

"Module not found: Ollama" (PowerShell method)

Ollama is a dependency for running local models. The pip install should grab it automatically, but if it's missing, install it manually:

pip install ollama

Then restart Open WebUI.

Slow Model Responses

If your machine has enough RAM (16GB+) but inference is still sluggish, check CPU usage in Task Manager. If CPU is 100% and GPU is 0%, your GPU is not being used. Follow the GPU Setup section above. If you have no GPU, move to a machine with one or scale up to managed hosting.

Frequently Asked Questions

Can you run Open WebUI on Windows? Yes. Open WebUI has a native Desktop app for Windows, plus PowerShell and Docker options. Windows 10 and Windows 11 are both supported.

Is there an Open WebUI desktop app available? Yes. The Open WebUI Desktop app is available on GitHub (2.7K stars). It's standalone, requires no Docker, and installs like any Windows application.

Is OpenWebUI free to use? Yes. Open WebUI is open-source and free. You pay only for any managed hosting (Opsily offers managed Open WebUI if you scale beyond your local machine).

What is the best app for Open WebUI? For Windows, the Desktop app is the simplest. PowerShell/pip works if Desktop fails. Docker is for advanced users or production setups.

Who is behind Open WebUI? Open WebUI is open-source with community contributors. The main repository is on GitHub under the open-webui organization. No single company owns it.

What can you do with Open WebUI? Chat with local language models (Mistral, Llama, Neural Chat, etc.). Upload documents for analysis. Create prompts and share them. Run models entirely on your machine with no cloud vendor lock-in.

The Bottom Line

Open WebUI runs natively on Windows and does not require Docker or Linux. The Desktop app is the easiest path for most users. If you have an NVIDIA card, GPU acceleration is worth the setup time: models run 5-10x faster.

The Desktop app is early-stage (v0.0.20), so expect occasional quirks. For team usage or production uptime, consider moving to managed hosting. For a single Windows machine, local installation is free and private.

Ready to evaluate Open WebUI beyond your Windows laptop? Opsily manages Open WebUI hosting for teams that need backups, uptime guarantees, and shared access.