Open WebUI after a month as my only chat front end

Sample post. This review was written to show how a post reads on the new site. Versions, measurements and the table below are illustrative, not test results.

Why I tried it

I had four browser tabs open for four chat products, two of which I was paying for, and a GPU box in the cupboard running Ollama that I only touched from the terminal. I wanted one place to talk to all of them, with my history on my own disk and accounts for the two other people in the house who kept borrowing my laptop.

Open WebUI is the obvious candidate: it started as a front end for Ollama and has grown into a general chat interface that speaks to any OpenAI-compatible endpoint. The question wasn't whether it works. It was whether it stays working when you live in it.

Setup

The install is one container. I ran it behind Caddy for TLS, with the data directory on a volume that my nightly backup already covers.

# compose.yaml (trimmed)
services:
  open-webui:
    image: ghcr.io/open-webui/open-webui:v0.6   # pinned: see "Upgrades"
    environment:
      OLLAMA_BASE_URL: http://gpu-box.lan:11434
      WEBUI_AUTH: "true"
    volumes:
      - open-webui:/app/backend/data
    restart: unless-stopped

First login creates the admin account. From there, connections to hosted APIs live in the admin settings, and each model can be shown or hidden per user group. It took me about twenty minutes from an empty VM to three working accounts.

Daily use

The chat itself is quick and stays out of the way. Switching models mid-conversation works, and comparing two models side by side on the same prompt turned out to be the feature I used most. It is a better way to decide whether a cheaper model is good enough than any benchmark I have read.

Document chat was fine for finding things and weak for reasoning across documents. On my folder of equipment manuals it answered "what's the torque setting on the mower blade" correctly every time, and gave confident nonsense when asked to compare warranty terms across three brands. That's a retrieval problem, not a front end problem, but the default settings don't warn you.

• Per-user history and model access worked without surprises.
• The mobile web view is usable, not pleasant. I added it to my home screen and stopped noticing.
• Tools and functions are powerful and easy to misconfigure. Anything that can run code should be admin-only.

Upgrades

This is where the month got expensive in time. Open WebUI ships often, and one minor release in my test window changed a settings default that silently hid two of my hosted models from non-admin users. Nothing was lost; it took an evening to notice and ten minutes to fix.

Pin the image tag. Read the release notes. Back up the data volume before you pull. All three, every time.

Compared with the alternatives

I ran LibreChat and AnythingLLM on the same VM for a week each, pointed at the same models, and did the same tasks in each.

What I checked	Open WebUI	LibreChat	AnythingLLM
Licence	Open WebUI License (BSD-3 plus branding clause)	MIT	MIT
Install	One container	Compose stack with MongoDB and search	One container, or a desktop app
Multi-user and groups	Yes, with per-group model access	Yes	Partly; roles, fewer controls
Local models via Ollama	Built in	Yes, as a custom endpoint	Built in
Document chat	Good for lookup, weak across files	Good, needs the RAG service	The strongest of the three
Side-by-side models	Yes	Partly	No
Idle memory	about 600 MB	about 1.1 GB with MongoDB	about 450 MB
Upgrade pain in my test	One silent settings change	None, one manual migration step	None
Who it suits	A household or small team that wants one place for every model	Teams who want an MIT licence and don't mind more moving parts	Mostly document work, one or two users

The table scrolls sideways; the first column stays put.

What it cost

The software is free. The VM sits on a box that was already running, so the marginal power cost is too small for me to measure honestly. The real costs were API usage, which is the same whichever front end you use, and roughly three hours of my time over the month: setup, the upgrade evening, and tuning document settings.

Verdict

Keep it. The best self-hosted chat front end I have used, as long as you pin the version and read release notes before upgrading.

If document chat is the main job, try AnythingLLM first. If the licence matters to your client, LibreChat is the safer pick.