Error messages

Open WebUI: Server Connection Error

Open WebUI: Server Connection Error

This one shows up as a red toast in the chat or the admin connections page:

Open WebUI: Server Connection Error

We read where it comes from in Open WebUI 0.11.4. It's a fixed string in the backend's list of error messages, and the Python server returns it when its own request to a model provider fails, say because the connection was refused or the host name didn't resolve. (Each spot that raises it carries a comment saying Open WebUI's license covers this "error identifier" and forks may not remove it, which we hadn't seen on an error message before.) It's common. As of September 25, 2026 the project has 87 issues and 57 discussions quoting it.

Who makes the request that fails

When you chat with an Ollama model, your browser talks to Open WebUI, and Open WebUI's backend talks to Ollama. So the address in the Ollama connection has to work from wherever the backend runs. Open WebUI's connection troubleshooting page puts it in a table: for backend connections, which include Ollama, localhost means "the machine/container running Open WebUI."

In the Docker install, that's the container. Ollama on the host at http://localhost:11434 answers fine in your browser tab and refuses the container. A maintainer said as much in issue #209, the longest thread on this error: "It seems like Ollama instance is not reachable from the docker container." A user in that thread ran curl inside the container and got Failed to connect to localhost port 11434: Connection refused while the host said "Ollama is running."

How to fix the address

Run the test the maintainer asked for. Open a shell in the Open WebUI container (docker exec -it open-webui bash, if you kept the default name) and curl the exact URL from your connection settings. If that fails, Open WebUI never had a chance.

The docs list what to use instead of localhost: host.docker.internal to reach the host, or a Compose service name like http://ollama:11434 when Ollama runs in its own container. The docs also say Ollama often listens only on 127.0.0.1, which the container can't reach even by the right name; set OLLAMA_HOST=0.0.0.0 and restart Ollama. Their other option is running the container with --network=host, after which Open WebUI moves to port 8080.

If you've added several connections and one of them is dead, the docs say the model list waits 10 seconds on it by default, and the logs show Connection error: Cannot connect to host.... Remove the dead one or lower AIOHTTP_CLIENT_TIMEOUT_MODEL_LIST.

When it starts with "Ollama: 500" instead

A different family of messages looks like this, and gets about as many reports (89 issues, 55 discussions):

Ollama: 500, message='Internal Server Error', url='http://localhost:11434/api/chat'

Here the connection worked. Ollama answered, and its answer was an HTTP 500, meaning Ollama itself failed on the request. Older releases (we checked 0.5.0) wrapped whatever went wrong in Ollama: plus the Python HTTP library's description, which is where the odd message= and url= come from. In 0.11.4 the backend passes on Ollama's own error text when it can read it, and falls back to Open WebUI: Server Connection Error when it can't. Either way, the useful log is Ollama's, and trying the same model with ollama run tells you whether Open WebUI is involved at all. In one thread a maintainer confirmed Open WebUI sent the request in the right format and suggested raising it with Ollama instead.