Goal: a ChatGPT-style UI in your browser, talking to models that live on your Windows PC. We’ll install Ollama (the engine) and Open WebUI (the friendly frontend). Then we’ll cover the failures that waste evenings: ports, firewalls, GPU drivers, and Defender.
Before you start
- Windows 10/11, admin rights for installers.
- Preferably 16GB+ RAM (see model guide).
- ~10–20GB free disk for a couple of models.
- If you have NVIDIA: install current Game Ready / Studio drivers before blaming Ollama for CPU-only speeds.
1) Install Ollama for Windows
- Download the official Windows installer from ollama.com/download.
- Run the installer. Accept defaults unless you have a reason not to.
- Open a new PowerShell or Terminal and run:
ollama --version - Pull a starter model:
ollama pull llama3.1:8b(or a current 7B–8B Qwen tag). - Test:
ollama run llama3.1:8b— type a short prompt, confirm a reply, then/bye.
By default Ollama serves an API at http://127.0.0.1:11434. Keep that address handy.
2) Install Open WebUI on Windows
Two common paths — pick one.
Option A — Docker Desktop (cleanest for most people)
- Install Docker Desktop for Windows and finish the reboot / WSL2 prompts it asks for.
- Confirm Docker is running (whale icon healthy).
- Run Open WebUI (example — verify the current official image/tag in Open WebUI docs):
docker run -d -p 3000:8080 --add-host=host.docker.internal:host-gateway -e OLLAMA_BASE_URL=http://host.docker.internal:11434 --name open-webui --restart always ghcr.io/open-webui/open-webui:main
- Browse to
http://localhost:3000, create the local admin account, and select your Ollama model.
Option B — Native / Python install
If you avoid Docker, follow the current Open WebUI Windows instructions (pip / uv / bundled release). Point the Ollama base URL at http://127.0.0.1:11434. The UI URL is often http://localhost:8080 — check the docs for your install method.
3) First successful chat checklist
- Ollama tray / service running.
curl http://127.0.0.1:11434or browser hit returns Ollama’s response.- Open WebUI shows your pulled models.
- One short chat completes without spinning forever.
Common failures (and fixes)
Port already in use
11434 (Ollama) or 3000/8080 (WebUI) taken by another app. In PowerShell: netstat -ano | findstr :11434 — note the PID, check Task Manager, stop the conflict, or change the WebUI publish port (-p 3001:8080).
Firewall blocking
Windows Defender Firewall may prompt on first run. Allow on private networks for local use. If WebUI in Docker can’t see Ollama, confirm host.docker.internal and that Ollama is listening on localhost (not only a weird bind address).
GPU drivers / “why is this so slow?”
- NVIDIA: install current drivers; reboot; confirm GPU appears in Task Manager → Performance.
- Laptop hybrid graphics: plug in power; set high-performance mode; ensure the discrete GPU isn’t forced off.
- No discrete GPU: CPU inference is normal — use smaller models from the 16GB guide.
Windows Defender / antivirus
Real-time scanning of multi-GB model files can make pulls and loads crawl. After you trust the official installers, consider excluding Ollama’s models directory (often under your user profile .ollama) — don’t blanket-disable Defender.
WSL2 / Docker Desktop stuck
Update WSL (wsl --update), enable virtualization in BIOS if needed, and give Docker enough RAM in settings (don’t starve Windows or Ollama).
Model pull fails mid-download
Flaky Wi‑Fi and disk-full are #1. Free space, retry ollama pull …. Prefer wired or stable Wi‑Fi for multi-GB downloads.
Hardware that makes install day less painful
Mini-PC or laptop with 16–32GB RAM
Quiet always-on box for Ollama + WebUI. Prioritize RAM over fancy CPU marketing.
https://www.amazon.com/s?k=mini+PC+32GB+RAM&tag=wethepeopl0b9-20USB SSD (1TB class)
Overflow storage for models if the internal drive is tight — USB 3.2 / USB4 enclosure preferred.
https://www.amazon.com/s?k=1TB+USB+SSD&tag=wethepeopl0b9-20Next steps
- Pick models from Best Ollama models for 16GB RAM.
- If the machine is the bottleneck, see Budget AI PC under $1,000.
- Want the beginner PDF when it launches? Contact / waitlist.