Open WebUI

Appearance

Workspace ​

The Workspace in Open WebUI provides a comprehensive environment for managing your AI interactions and configurations. It consists of several key components:

  • 🤖 Models - Create and manage custom models tailored to specific purposes
  • 📚 Knowledge - Manage your knowledge bases for retrieval augmented generation
  • 📝 Prompts - Create and organize reusable prompts
  • 🔒 Permissions - Configure access controls and feature availability

Each section of the Workspace is designed to give you fine-grained control over your Open WebUI experience, allowing for customization and optimization of your AI interactions.

bbv AI Hub ​

The modern interface reminds of ChatGPT, but here much more can be done than chatting and web search.

  • Create custom agents, e.g. a DB Agent which can directly query a DB and work with the retrieved data.
  • Retrieval Augmented Generation, create knowledge collections and upload documents to provide more context for your agent.
  • Web Search, enable searching the web to answer your questions with more relevant information.
  • Custom Prompts, define custom prompts that can be used to streamline tasks.
  • Image Generation, create images from chat messages with the press of a button.
  • Audio, instead of typing what you want to say - just say it.
  • Coding, no need to open your IDE to run code - run it in the chat.

Agents ​

The combination of knowledge collections, predefined prompts and different models allows for creating agents that can support users in a wide range of tasks.

RAG ​

Whether its collections of knowledge or single documents the hub is able to handle all of them. By referencing a piece of information in the chat using the "#" symbol. The model or agent will have access to that information when generating its response. The process of ingesting this information is highly customizable, from which embeding model to use to the prompt that is passed together with the retrieved information. Google Drive and OneDrive can also be attached for easy access to files.

The agent can now generate search queries and look for information on the web making its responses relevant and minimize the chances of hallucination.

Prompts ​

Custom prompt templates can be provided for any number of tasks which the hub should assist with. These are called by typing "/" and a selection of available prompts will appear. These can be filled with placeholder values which the user can fill in directly to suit their specific use case. For easy switching between placeholders use the tab key.

Image Generation ​

Either use an existing response from the model to generate an image or input a new prompt. Both lead to an image being generated by either DALL-E or some other model used to generate images.

Audio ​

Be it speech-to-text or text-to-speech - both are supported in the hub. Input can be given through speech and output read aloud. This provides ease of use for longer inputs or outputs.

Coding ​

Run generated code directly in the chat, by referencing the "Pydiode environment" the model or agent will know to generate code which is executed using said environment. To run existing code in the chat encase it in backticks to mark it as code. This allows for quick development without using a IDE.