Open WebUI Logs In, But Chats Fail

This is a different problem than "no models show up"

If the login screen loads, you sign in, and models appear in the dropdown, the connection between Open WebUI and Ollama is already working - that rules out the whole class of problem covered in Ollama Is Running, But Open WebUI Can't Connect. Something is failing later, specifically when a message is actually sent.

Check the logs at the moment it fails

Send a message, let it fail, then immediately check what the container actually logged:

docker logs --tail 50 open-webui

The real error is almost always sitting right there - the difference between "the request timed out," "the model crashed," and "the response was rejected" all point to different fixes below.

The most common real causes

• The model was evicted from memory between loading and generating. If the machine is under memory pressure, Ollama can drop a loaded model to free RAM/VRAM, and the next request has to reload it - or fails outright if there isn't room. See Ollama Says "Out of Memory" if this is what the logs show.
• A stale browser session after an update. If Open WebUI was updated recently, cached JavaScript from the old version can still be loaded in your browser tab. A hard refresh (not just a normal reload) clears this - Cmd+Shift+R on Mac, Ctrl+Shift+R on Windows/Linux.
• A proxy or tunnel timing out on a slow first response. If you're reaching this through Tailscale or a reverse proxy, and the model is large enough that the first token takes a while to generate, some default timeout configurations can cut the connection before Ollama actually responds. This shows up as the request failing specifically on larger or slower models, working fine on small ones.