Michael Limberger

The window at localhost:3000

Open http://localhost:3000. If you have used ChatGPT, most of this feels familiar. The center of the screen is the conversation. Type at the bottom, press Enter, and the model responds above. Streaming is on by default, so you see the answer word by word. The model selector sits at the top. You can switch models mid-conversation if you want.

The left sidebar is history. Every chat saves. Rename a chat by clicking its title. You can make folders, pin chats to the top, search, and delete. New Chat starts a fresh thread. Each chat keeps its own context.

Select a model and try something simple: "What can you tell me about the city I live in?" It will probably ask which city. That is fine. It proves the conversation path works. Watch the response stream in. There is a Stop button if you want to cut it short. Markdown renders for headers, code, lists, and bold text. Code blocks get a copy button.

Hover a message and you get Copy, Edit, Regenerate, and Good/Bad ratings for your own reference. Edit lets you rephrase your last user message and regenerate from that point. You do not have to start the whole chat over.

Dropdown, paperclip, @, keys

The model selector lists everything you have pulled. A checkmark marks the current one. Type to search. Custom Workspace models appear in the same list. You can also select multiple models and compare responses in parallel columns: quality, style, and speed side by side. That is a good way to pick a favorite per task.

The attachment icon accepts PDF, Word, text, images (if a vision model is selected), and any text-based code file. For documents, OpenWebUI extracts text into context. For images, use something like Qwen3-VL or Mistral Small 3.2. We will demo vision later.

Type @ to reference a model, a knowledge base, or an uploaded file without leaving the chat. Keyboard shortcuts worth knowing: Enter sends, Shift+Enter makes a new line, Ctrl+Shift+O starts a new chat, Ctrl+Shift+S toggles the sidebar, and Up Arrow edits your last message when the input box is empty.

Profile is at the bottom left. General covers dark mode and language. Interface covers display and highlighting. Audio covers text-to-speech. Connections is where the Ollama URL lives. Models lets you pull from the UI: Admin Panel, Settings, Models, then type a name. Next: custom models in Workspace.