DEV Community

egecan nefis
egecan nefis

Posted on

How to Create a Personal AI Assistant Using Ollama and Open WebUI

Running a Personal AI Assistant Using Ollama and Open WebUI sounds simple until something refuses to load, the model won't respond, or Open WebUI sits there with an empty chat window. I hit several of those problems while setting up my own local assistant, and the fixes turned out to be less obvious than most guides suggest.

Instead of showing only the happy path, this guide covers the setup, the common failures, and the fixes that actually worked.

Quick Answer

If your Personal AI Assistant Using Ollama and Open WebUI isn't working correctly:

Install Ollama before Open WebUI.
Confirm Ollama is running before opening the web interface.
Pull at least one model with ollama pull.
Check whether Open WebUI can reach the Ollama API.
Don't overlook firewall rules or Docker networking.
Why Your Local AI Assistant Doesn't Work

Most setup guides stop after the installation commands. That's where many people get stuck.

From what I've seen, these are the biggest causes.

  1. Ollama Isn't Actually Running

Installing Ollama doesn't always mean the background service is active.

Try:

ollama serve

Then open another terminal.

ollama list

If you receive a model list, the service is alive.

If you see connection errors, Open WebUI won't have anything to connect to.

  1. No AI Model Has Been Downloaded

I've actually forgotten this myself.

Open WebUI may launch perfectly while having absolutely nothing available to chat with.

Download one.

ollama pull llama3

Or

ollama pull mistral

After that:

ollama list

should display your installed models.

  1. Open WebUI Can't Reach Ollama

This happens surprisingly often.

If Open WebUI expects Ollama on:

http://localhost:11434

but your Docker container only sees itself, you'll just get loading errors.

But that's not entirely accurate, let me explain.

When Open WebUI runs directly on Windows or Linux, localhost is usually fine.

When Open WebUI runs inside Docker, localhost means the container itself—not your computer.

That's a networking issue, not an AI issue.

Installation Overview

Here's the order I'd follow if starting from scratch again.

Install Ollama.
Start the Ollama service.
Download a model.
Install Open WebUI.
Connect Open WebUI to Ollama.
Test with a simple prompt.

Skipping step three causes more confusion than you'd expect.

Common Setup Scenarios
Native Windows Installation

Usually the easiest option.

Possible problems include:

Windows Defender Firewall
Another application already using port 11434
Antivirus blocking local services
Linux Installation

Linux generally behaves well.

The issues I ran into were mostly permission-related and services not starting automatically after reboot.

A quick check:

systemctl status ollama

can save a lot of guessing.

Docker Installation

Docker adds another layer.

Typical problems include:

Wrong network mode
Missing environment variables
Container unable to reach Ollama
Port mapping mistakes

Honestly, Docker networking still manages to waste my time more often than I'd like.

Comparison of Common Problems
Problem Typical Symptom Most Likely Cause Usually Fixes It
Blank chat Endless loading Ollama offline Start Ollama service
No models Empty model list Nothing downloaded Pull a model
API error Connection refused Wrong endpoint Correct API address
Slow responses Long wait Small RAM or CPU Smaller model
GPU unused High CPU load GPU detection failed Install GPU drivers
Step 1: Install Ollama

Download the installer for your operating system.

Verify installation.

ollama --version

Then launch the service.

ollama serve

Leave that running.

Step 2: Download an AI Model

For example:

ollama pull llama3

Or if your PC has limited RAM:

ollama pull phi3

Smaller models respond faster on older hardware.

Step 3: Install Open WebUI

Install Open WebUI however you prefer.

Most people either:

use Docker
install with Python
run it on Linux directly

The installation itself rarely causes trouble.

The connection between Open WebUI and Ollama is where things usually break.

Step 4: Verify the API

Open:

http://localhost:11434

If Ollama is running correctly, you'll receive a simple response instead of a browser error.

So if this page doesn't load, Open WebUI won't work either.

Step 5: Test a Prompt

Try something tiny.

Hello

or

Write a Python loop.

Large prompts make debugging harder.

Start simple.

What Actually Worked For Me

My first attempt was messy.

I restarted Docker twice, reinstalled Open WebUI, and even downloaded another model because I assumed the first one had become corrupted.

None of that changed anything.

Then I remembered an old forum comment about Docker containers treating "localhost" differently. I switched the API endpoint to the correct host address, restarted the container, and everything immediately started responding. I got a little lucky because I wasn't even looking for that problem anymore.

Later I repeated the installation on another machine, and this time the only issue was forgetting to download a model. That was much easier to fix.

Advanced Troubleshooting
Check Running Processes

Linux:

ps aux | grep ollama

Windows:

tasklist | findstr ollama

No running process means nothing is serving requests.

Test the API Directly

Instead of guessing whether Open WebUI is the problem, test Ollama itself.

curl http://localhost:11434/api/tags

If models appear in JSON output, Ollama is fine.

If this command fails, fix Ollama first.

Review Docker Logs

If using Docker:

docker logs open-webui

Look for:

connection refused
timeout
API unavailable

Those messages usually point toward networking instead of installation.

Watch System Memory

Large language models consume a surprising amount of RAM.

If your computer starts swapping memory to disk, responses become painfully slow.

I originally blamed the model. It turned out Windows was simply running out of memory.

Check GPU Detection

Run:

ollama run llama3

Watch system usage.

If CPU jumps to 100% while GPU stays idle, the GPU may not be detected correctly.

Your mileage may vary depending on your graphics drivers.

Unexpected Causes People Miss

One thing people rarely check is VPN software.

Some VPN clients interfere with localhost communication.

Another overlooked cause is corporate endpoint security software.

I've seen security tools silently block local ports without displaying any warning.

And don't forget browser extensions. One privacy extension on my test machine blocked local API requests until I disabled it.

Fixes That Usually Work

Across several installations, these solved the issue most often:

Starting Ollama before launching Open WebUI.
Downloading at least one model.
Correcting the API endpoint.
Restarting Docker containers after changing environment variables.
Checking firewall permissions.
Fixes That Rarely Solve Anything

People often recommend:

rebooting Windows
reinstalling Open WebUI immediately
downloading multiple models
clearing browser cache

Sometimes those help.

Most of the time they don't address the actual problem.

Prevention Tips
Keep Ollama updated.
Test the API before changing Open WebUI settings.
Use smaller models on lower-memory systems.
Leave enough free disk space for model downloads.
Avoid changing Docker networking unless you know why you're changing it.
Save your working configuration once everything functions. You'll thank yourself later.
FAQ
Why does Open WebUI show no models?

Usually because no models have been downloaded yet or Open WebUI can't contact Ollama.

Why is Ollama responding so slowly?

Large models can overwhelm older hardware. Try a smaller model first.

Can I run everything without Docker?

Yes. Plenty of people do, and for a first setup it's often less confusing.

Why does localhost work in my browser but not inside Docker?

Because the container has its own localhost. That's a common networking misunderstanding.

My GPU isn't being used. Is something broken?

Not necessarily. Check your graphics drivers, supported hardware, and whether Ollama recognizes the GPU. From what I've seen, driver problems show up more often than hardware failures.

Does Open WebUI need an internet connection after installation?

Not for local models. If everything is installed correctly, you can chat entirely offline.

Editor's Opinion

I like this setup because once it's working, it mostly stays out of the way. The annoying part is getting there. I spent way longer chasing Docker networking than I'd admit, and the actual fix was tiny. I'd probably start with a native install next time unless I really needed containers. https://www.howtotechdaily.com/

Top comments (0)