Let me start this off by saying: I’m no stranger to chasing down tech rabbit holes. You know how it is. You see a tool, you tinker, you break something (inevitably), and then you spend the next few hours figuring out why it’s broken. But when it all clicks, there’s this moment of pure, nerdy satisfaction. That’s exactly what happened when I got Ollama, Gemma3, and OpenWebUI running on my setup.

Spoiler alert: It’s not the Cadillac of AI chatbots, but it’s not half bad either. Let me walk you through how I pulled it off, just in case you feel like setting up your own little AI playground. Trust me—it’s fun, educational, and only mildly frustrating.

The Starting Point: Why I Wanted This Setup

It all started with me wanting to mess around with some AI tools that weren’t locked behind paywalls, APIs, or overly complicated setups.

I wanted something local, something I could tinker with. Enter Ollama and OpenWebUI.

• Ollama: A sleek little tool for running models locally. I went with Gemma3, which is surprisingly decent for its size.
• OpenWebUI: A Docker-based interface for running AI models locally. Think of it as a self-hosted version of those fancy AI tools you see everywhere, minus the subscription.

I figured if I could make these two play nicely together, I’d have a pretty solid setup for running local AI experiments. Plus, I found a plugin-like tool that can scrape web pages, which just adds more chaos (the fun kind) to the mix.

Step 1: Fixing My Broken WSL/Docker Setup

First things first: My Windows Subsystem for Linux (WSL) setup was a disaster. I couldn’t even open Docker Desktop without an error. Every time I tried to run WSL, I was hit with this lovely message:

WSL2 is not supported with your current machine configuration. Please enable the Virtual Machine Platform optional component and ensure virtualization is enabled in the BIOS.

Cool. No problem. Just a quick visit to BIOS, right? (Spoiler: It’s never just that.)

What I Did:

1. Enable Virtualization in BIOS:

   • Restarted my PC.
   • Slammed F2 like my life depended on it (YMMV—yours might be Del or Esc).
   • Found the virtualization settings under CPU options and turned it on.

2. Enable WSL2 and Virtual Machine Platform:
   In an elevated PowerShell (aka “Run as Administrator”), I ran:
   
   dism.exe /online /enable-feature /featurename:Microsoft-Windows-Subsystem-Linux /all /norestart
dism.exe /online /enable-feature /featurename:VirtualMachinePlatform /all /norestart

3.  

   Then I restarted my machine.

4. Verify WSL:
   After rebooting, I ran:

   wsl --status
   

   This time, no errors. I was in business.

Step 2: Installing Ollama

With WSL behaving itself, I moved on to Ollama. Installation was a breeze, which is rare in my world of tinkering.

Steps to Install:

1. Downloaded the installer directly from the Ollama website.

2. Let the installer do its thing. (Pro tip: Don’t overthink this part. Just click through.)

3. Once installed, I ran:

   ollama pull gemma3
   

   This grabbed the Gemma3 model. It’s lightweight, fast, and good enough for casual chatbot experiments.

4. Verified it was working:

   ollama chat
   

   Boom. A working chatbot right from my terminal.

Step 3: Setting Up OpenWebUI in Docker

Next up was OpenWebUI, which is where things got a little more interesting. This tool runs inside a Docker container, so I had to make sure Docker Desktop was happy now that WSL was fixed.

Steps to Install:

1. Install Docker Desktop:

   • Downloaded it from Docker’s website.
   • Enabled WSL 2 integration in the Docker settings.

2. Run OpenWebUI:
   With Docker up and running, I pulled the OpenWebUI container:

   docker run -d -p 7860:7860 --name=openwebui openwebui/openwebui:latest
   

   This exposed the app on port 7860. All I had to do was open my browser and go to http://localhost:7860.

3. Play Around with Tools:
   OpenWebUI has this cool plugin system. I tested one that scrapes web pages, which was surprisingly decent. It’s not ChatGPT-level scraping, but it gets the job done for most basic tasks.

Step 4: Making Them Work Together

Here’s where things got fun: Ollama and OpenWebUI don’t natively integrate, but you can run them side by side. My workflow looked like this:

1. For casual chatting: I’d use ollama chat in the terminal.
2. For more advanced stuff: I’d use OpenWebUI in the browser. This setup is great for running more resource-heavy models or testing plugins.

Lessons Learned

1. WSL is great—when it works.
   If you’re on Windows and want to tinker with Linux-based tools, WSL2 is a lifesaver. Just make sure virtualization is enabled, and don’t forget to enable the right Windows features.

2. Docker Desktop + WSL2 = Local AI Playground.
   OpenWebUI running in Docker is a game-changer. It’s not perfect, but for local AI experiments, it’s hard to beat.

3. Ollama is surprisingly good.
   Gemma3 isn’t the most powerful model out there, but it’s fast, lightweight, and doesn’t require a fancy GPU to run.

Final Thoughts

If you’re looking to build your own local AI setup, give this a shot. Between Ollama, OpenWebUI, and a little Docker magic, you’ve got everything you need to tinker, experiment, and maybe even build something cool. Plus, there’s a certain satisfaction in knowing your AI isn’t phoning home to some server farm every five seconds.

So, what’s next for me? Maybe I’ll try running a larger model—or maybe I’ll just keep playing around with plugins until I break something else.

Anyway, let me know if you give this setup a shot. And if you run into trouble, don’t worry—half the fun is figuring it out.

Until next time, happy tinkering!