Keeping Everything Updated

Why bother updating

Once this is running well, it's tempting to never touch it again - and for the model itself, that's mostly fine, nothing breaks by leaving a version alone. But this is also a piece of software running continuously on your network, and like anything in that position, security fixes land in updates. Combine that with genuine capability improvements in newer model versions, and a little periodic maintenance is worth the ten minutes it takes.

This isn't a "check daily" situation. It's closer to updating a router's firmware - infrequent, low-effort, and worth doing when you think of it.

Before you update anything

None of this takes long, and all of it makes "something feels off after updating" much faster to diagnose or undo:

• ☐ Back up Open WebUI's data - see Backing Up Your Local AI Setup, especially if it's been a while since the last one
• ☐ Record current versions (ollama --version, docker --version, the Open WebUI version shown in its own settings) - useful to know exactly what you're rolling back from if something breaks
• ☐ Save your current model list: ollama list > ollama-models.txt
• ☐ Confirm your Docker volume names with docker volume ls before touching anything
• ☐ If it's a major version jump for Ollama or Open WebUI, skim the release notes for breaking changes rather than assuming it's a routine update

Updating Ollama and models

On Linux, re-running the same install command updates an existing install in place:

curl -fsSL https://ollama.com/install.sh | sh

On Mac, the Ollama app checks for updates automatically and prompts you when one's available. On Windows, re-download and run the installer the same way you did the first time.

Models update the same simple way - pull the tag again:

ollama pull llama3.2

If nothing changed since you last pulled it, this finishes almost instantly since there's nothing new to download. If a newer version exists under that tag, only the changed layers are downloaded, not the whole model from scratch.

Updating Open WebUI

Pull the newer image, then recreate the container:

docker pull ghcr.io/open-webui/open-webui:main
docker stop open-webui
docker rm open-webui

Then run the exact same install command from Getting a Proper Chat Interface Running again. Because your chat history lives in a separate Docker volume, not inside the container itself, removing and recreating the container this way doesn't touch your data.

If you've moved to Docker Compose, this whole sequence is just docker compose pull && docker compose up -d instead.

After you update

A quick pass through these catches most real problems before you've forgotten what you just changed:

• ☐ Open WebUI actually loads
• ☐ Your models still show up in the dropdown
• ☐ A real test chat gets a real response
• ☐ RAG still works, if you use it - see Talking to Your Own Documents with RAG
• ☐ GPU acceleration still shows up in ollama ps, if you have one - see Why Isn't Ollama Using My GPU? if it doesn't
• ☐ Remote access still works, if you use LAN access or Tailscale

If something's off, compare against the versions you recorded before updating - that's the whole reason to have written them down.

What actually goes wrong

Ollama update seems to install but nothing changes

The service usually needs an explicit restart to pick up the new version. On Linux:

sudo systemctl restart ollama

On Mac, quit and reopen the Ollama app from the menu bar.

A model is missing from ollama list after updating

See A Model Disappeared After an Update - this is almost always OLLAMA_MODELS pointing somewhere different after the update, not the update actually deleting anything.

Chat history disappeared after updating Open WebUI

This almost always means the container was removed with docker rm -v instead of a plain docker rm, deleting the data volume along with the container. If you have a separate backup of the volume (see Backing Up Your Local AI Setup), restore from that. Otherwise, unfortunately, that history is gone - which is exactly why the warning above matters.

A custom model built with a Modelfile stopped acting the way it used to

Custom models built with ollama create pin to the base model version that existed at the moment you built them - they don't automatically pick up a newer base model just because you pulled one. If you want a custom model to reflect an updated base, rebuild it: run the same ollama create command from Persistent Instructions With Ollama Modelfiles again after pulling the new base version.

Common questions