Configuring Open WebUI for AI Interaction|Introduction
Applies to SUSE AI 1.0

1 Introduction

 

Open WebUI is a Web-based interface that allows you to interact with AI models, such as large language models (LLM). It simplifies working with AI by providing a graphical user interface (GUI) instead of relying on command-line tools. The goal of Open WebUI is to make AI accessible to a broader audience, including those without technical expertise in AI development.

1.1 What are the key benefits of using Open WebUI?

 
Interaction with AI models

You can load a preferred AI model and interact with it using prompts in your natural language.

Custom data handling for RAG workflows

Open WebUI allows you to upload documents or specify URLs of pages that the AI can parse and reference during interaction, which is particularly useful for RAG workflows.

User-friendly interface

By providing a graphical user interface, Open WebUI makes AI tools more accessible to non-technical users. It helps them use the power of advanced AI models with no need for technical knowledge.

Data privacy

You can deploy Open WebUI locally or in a private environment. Such deployment is required for users concerned about privacy and data security.

1.2 How does an AI chatbot process user queries?

 

When you enter a user prompt, several processes happen in the background to generate the response.

  1. Input processing. The AI first processes the text of your prompt to understand its meaning. This involves identifying the subject, intent and any details or context provided. This process is called Natural Language Understanding (NLU).

  2. Contextual analysis. If you are interacting with AI in a session where you have already asked previous questions, the AI considers the context of the conversation. This results in more relevant and coherent answers.

  3. Knowledge retrieval. The AI retrieves information from its pre-trained knowledge base. This database includes facts, data and concepts that the AI has been trained on. AI models can also utilize retrieval-augmented generation (RAG) systems to get contextual information from the data provided by the organization. If the AI has access to real-time data, it can search for the latest information online to provide an up-to-date response.

  4. Response generation. Using natural language generation (NLG) techniques, the AI constructs a coherent and grammatically correct response based on the information it retrieved.

  5. Output. The AI delivers the response in a requested form, usually in a human-readable format. This can be a detailed explanation, a step-by-step guide, or a customized image, depending on the complexity of your query.

  6. Feedback Loop (Optional). In specific AI systems, your feedback or follow-up questions can help refine future responses, allowing the AI to improve its answers over time.

1.3 Known Issues

 

This section lists the known issues with the Open WebUI and their solutions.

1.3.1 User password reset

 

An Open WebUI user can only reset their own password if they are logged in to Open WebUI and have an active Web browser session. However, if the user has lost their password and cannot log in to Open WebUI, the password must be reset by an SUSE Rancher Prime: RKE2 administrator using the kubectl command.

  1. Generate a new password hash using the htpasswd command. htpasswd is part of the apache2-utils package.

    > htpasswd -bnBC 10 "NEW_PASSWORD"| tr -d ':\n'

  2. Identify the Pod where the Open WebUI application is running.

    > kubectl get pod -n SUSE_AI_NAMESPACE| grep open-webui | grep -v pipelines

  3. Update the password.

    > kubectl exec open-webui-0-n SUSE_AI_NAMESPACE – \
  python3.11 -c 'import sqlite3; con = sqlite3.connect("/usr/lib/python3.11/site-packages/data/webui.db"); \
  con.cursor().execute("UPDATE auth SET password=\"PASSWORD_HASH\" WHERE email=\"admin@example.org\""); con.commit()'

1.3.2 Audio processing stops after several minutes

 

When the voice recorder is used as described in the section Chat window, Open WebUI tries to process the audio, but it times out without any warning and no text is generated. The API endpoint /audio/api/v1/transcriptions, which receives the audio to be processed, results in a 504 Gateway Timeout.

1.4 For more information