Using the Workspace in Open WebUI

In the ‘Models’ tab, you do not create a new model at the server level, but a preset built around an existing model. This way, for example, you combine a system prompt, fixed knowledge bases, tools, skills, and capabilities into a reusable assistant.

When creating a model preset, you specify the following from top to bottom:

1. Access: who has access to your preset? (default, only yourself).
2. Image: an image; this helps you quickly recognize your preset.
3. Model name: the name of your preset; this does not have to be the name of the actual model.
4. Select a base model: choose from one of the models that the admin has made available within Open WebUI.
5. Description: A short description; more useful than ‘what’ the model does is ‘what it is for’ and which tools, skills, and knowledge it has to do so.
6. System prompt: The instruction you give the model, for example ‘You’re a helpful support assistant'.
7. Advanced Params: You are free to adjust these as desired. However, the content falls outside the scope of this manual.
8. Prompts: You can pass standard prompts to the preset model, for example ‘Acting as a senior translator, translate the provided text, list spelling and grammar mistakes in the original, etc.’. See also ‘Prompts’ below.
9. Knowledge: Do you want to train the custom model based on your own data, such as user manuals or knowledge bases? Then add the ‘Knowledge’ here that you added under the ‘Knowledge’ tab; see below under ‘Knowledge’.
10. Tools: Select available tools that your custom model can use automatically; see ‘Tools’ further down in this section.
11. Skills: Select available tools that your custom model can use automatically; see ‘Skills’ further down in this section.
12. Capabilities: Which capabilities your custom model has available out-of-the-box.
13. Features: Which platform-wide features you make available to your model. 
14. Builtin tools: Open WebUI comes with various built-in tools that you can turn on and off as desired; see this page for details. These built-in tools are separate from the tools you can configure yourself. 
15. TTS Voice: Do you use text-to-speech? Then you can specify a specific voice here per custom model.
16. Finally, click on ‘Save & Create’ to create your custom model.

In the ‘Knowledge’ tab, you create knowledge bases that Open WebUI can consult later. During creation, you specify a name, description, and access rights, among other things.

Afterwards, you add documents or files so that Open WebUI can index them for RAG use. In a chat, you can then load a knowledge item with #.

Click on ‘+ New Knowledge’ at the top right and give your new knowledge a name (1), description (2) and optionally adjust who has access to the knowledge base (3) and click on ‘Create Knowledge’ (4).

You will now automatically arrive at the screen below. Click on the ‘+’ (1) and click on the resource you want to use for your knowledge base (2). You can repeat this process to add different resources. 

You store reusable prompt templates in the ‘Prompts’ tab. You use a prompt in a chat session via /, for example as a fixed summary prompt, translation prompt, SEO template, or standard instruction for emails.

Click on ‘+ New Prompt’ at the top right. Next, enter a ‘Name’ (1), a slash command to call the prompt from the chat (2), optional ‘tags’ (3), the ‘Prompt’ (4), who has access to the prompt via ‘Access’ (5), and finally click on ‘Save & Create’ (6) to create the prompt.

Skills are reusable instructions in plain text or Markdown. They teach a model how to approach a task, for example, according to a fixed writing style, review procedure, or support workflow.

In a chat, you add a skill with $. You can also link skills to a model preset (see ‘Models’ above) so that they are available by default.

Click ‘+ New Skill’ at the top right. Then in the create view, enter a 'Skill Name' (1), 'Skill ID' (2), 'Skill Description' (3) and 'Skill Instructions' (4). Customize who has access to your skill via 'Access' (5) and click 'Save & Create’ (6) to create the skill.

In the 'Tools’ tab, you manage Python scripts that allow a model to perform actions, such as API calls, calculations, scraping, or other Python logic.

Click on ‘+ New Tool’ > ‘New Tool’ at the top right (or import via a link).

Now enter a ‘Tool Name’, ‘Tool ID’ and ‘Tool Description’. Next, you place the Python code of the tool itself in the editor (4), indicate who has access to the tool via ‘Access’ (5), and save the tool via ‘Save’ (6).