Some pro tips if you run Open WebUI locally (in a docker container):

Useful Shell Aliases

# Alias to easily start and stop the container
# "open" and "close" start with different letters and
# are easier to remember, so uio and uic

alias uio="docker start open-webui"
alias uic="docker stop open-webui"

# Open WebUI has frequent releases, so run this at least
# every week to be on the latest version
# Update command from https://docs.openwebui.com/getting-started/quick-start/#updating

alias uiu="docker rm -f open-webui && docker pull ghcr.io/open-webui/open-webui:main && docker run -d -p 3000:8080 -e WEBUI_AUTH=False -e BYPASS_MODEL_ACCESS_CONTROL=True -v open-webui:/app/backend/data --name open-webui ghcr.io/open-webui/open-webui:main"

Using Claude Extended Thinking

To use extended thinking with Claude models, you need to set temperature to 1 in Open WebUI. Changing temperature every time you want thinking is annoying. If you are using multiple models together, it makes things difficult as well.

My solution is to create a custom model for thinking mode. Changing models is way easier, and having a custom model is way more useful (as you’ll see later).

Go to Workspace, then Models, and then create a new model using the base model for which you want extended thinking. Name the model appropriately. For example, for the thinking version of anthropic/claude-sonnet-4-20250514, I name it anthropic-custom/claude-sonnet-4-thinking-20250514. Make sure the model ID (just below the name) is set to an appropriate string (this will come in handy later). Set visibility to public, and in Advanced Params, set temperature to 1 and Reasoning Effort to high (or medium or low). Save it, and now you can use this model whenever you want extended thinking without having to change temperature.

When you use this custom model, you should be able to see the model thoughts in a dropdown.

The image below shows this setup. You can create a similar setup for 3.7 Sonnet and 4 Opus.

Fix for localhost:3000

I run lots of things locally. If I start typing localhost in the URL bar, the autocomplete gives the wrong suggestion, and I have to fix the port number all the time. Instead of using localhost:3000, I use ui.localhost:3000 and always get correct autocomplete in the URL bar. Use this technique for everything running locally, and life will be much easier. For example, for Karakeep I use karakeep.localhost:4000, and for Python HTTP server, I use python.localhost:8000.

Direct Links for Models

I bookmark direct links to temporary chat and the models I use most frequently in a single folder. This makes accessing specific models much quicker. Remember how we created a custom model for extended thinking? You can use the model ID to bookmark it, and now you can directly access the thinking model in a couple of clicks. When creating custom models, you can also add system prompts. I create one custom model for translation and bookmark the link to the model, so translation is always a couple of clicks away.

If you like to run multiple models in parallel, you can bookmark direct links for them as well! Direct link that runs Opus 4, O3, and Gemini 2.5 Pro in parallel:

http://ui.localhost:3000/?models=anthropic/claude-opus-4-20250514,openai/o3-2025-04-16,vertex_ai/gemini-2.5-pro-preview-05-06