Getting Started with C.H.U.C.K. and Open WebUI

Welcome to C.H.U.C.K: Your In-House AI Assistant

This manual will guide you through using C.H.U.C.K. through a chat interface called Open WebUI, a powerful tool that allows you to interact with various Artificial Intelligence (AI) models directly within our organization. Our in-house deployment of these models ensures that all your interactions and the data you use remain secure and confidential, protecting our valuable intellectual property.

This training aims to help you get comfortable with C.H.U.C.K. and Open WebUI and understand how to leverage its capabilities to enhance your productivity and creativity.

Agenda:

Accessing the Interface
Selecting a Model
Conversation with a Model
Understanding Citations
Using Prompts:
- Generating an Out of Office Message
- Summarize Teams Meeting Transcript
- Performance Review Wording Improvements
Adding Context to a Conversation:
- Access the existing knowledge base (RAG)
- Having a model search the web
- Uploading files
- Adding a URL to the conversation
Why Ask CHUCK (and our other in-house models) and not other AI tools?

1. Accessing the Interface

Here's how to access it:

Open your web browser (e.g., Chrome, Edge, Firefox).
Go to the designated internal URL: chatbot.services.conexusnuclear.org. You might want to bookmark this page for easy access.
Log in with your Conexus Nuclear credentials if prompted. This step ensures that all interactions are secure and tied to your user profile. If you are already signed in, the application will forward you to the main chat window

Once logged in, you'll be greeted by the Ask C.H.U.C.K. chat window, ready for your first prompt. It's designed to be intuitive, so you can start interacting with the AI model right away.

2. Selecting a Model

Open WebUI provides access to several AI models, each with unique strengths. You can switch between models depending on your specific needs.

CHUCK: Based on Llama3.2, Chuck is our default general-purpose model. It also has vision capabilities, meaning it can understand and discuss images you upload. Chuck is a great starting point for a wide range of tasks.
MITCH: Based on Deepseek 32b, Mitch is another robust general-use model. Its key feature is its "thinking capability," which can be helpful for more complex reasoning or problem-solving tasks.
GEMMA: Based on Google Gemma3:32b, Gemma is currently our most powerful general-use model. It also features vision capabilities. For tasks requiring deep understanding, nuanced generation, or image analysis, Gemma is an excellent choice.

How to Select a Model:

Usually, you'll find a dropdown menu or a selection panel within the Open WebUI interface (often located near the top left side of the chat window) where you can see the list of available models. Simply click on the name of the model you wish to use for your current conversation.

While CHUCK is the default, we encourage you to experiment with GEMMA, especially for tasks demanding higher performance.

3. Conversation with a Model

Interacting with our AI models is like having a text-based conversation.

Starting a Conversation: Once you've selected a model, you'll typically see a text input box at the bottom of the screen. Type your question or instruction (this is called a "prompt") into this box and press Enter or click the send button.
Continuing the Conversation: The model will process your prompt and generate a response. You can continue the conversation by typing further questions, asking for clarifications, or providing more information. The model will remember the context of the current conversation.
Starting a New Chat: If you want to start a new topic or ensure the model doesn't refer to previous parts of your discussion, look for an option like "New Chat". This will clear the context and begin a fresh interaction.

4. Understanding Citations

When an AI model provides information, especially if it's drawing from a specific document or internal knowledge base, it may provide citations or references to the source of that information.

What are Citations? Citations indicate where the model found the information it's presenting. This is crucial for verifying accuracy and understanding the context of the response.
How they Appear: Citations might appear as footnotes, links, or direct references to document names or page numbers, depending on how the model and Open WebUI are configured.

Why they are Important: Always review citations if provided, especially for critical information. They help you trust the output and allow for deeper exploration of the source material.

(Note: The exact appearance and functionality of citations can vary. We will explore specific examples during the hands-on portion of the training.)

5. Using Prompts

A "prompt" is the instruction or question you give to the AI model. The quality of your prompt significantly influences the quality of the AI's response. Clear, specific, and well-contextualized prompts yield better results.

Here are some examples:

A. Generating an Out of Office Message

Instead of just saying "Write an out of office message," try being more specific:

Basic Prompt: "Write an out of office message."
Better Prompt: "Write a professional out of office message. I will be on vacation from [Start Date] to [End Date], returning on [Return Date]. For urgent matters, colleagues can contact [Colleague's Name] at [Colleague's Email or Phone Number]."
Even Better Prompt (with tone): "Draft a friendly and professional out of office email. I'll be on annual leave from Monday, June 9th, until Friday, June 13th, with limited access to email. Please state that for urgent operational queries, John Smith can be reached at j.smith@example.com, and for urgent sales inquiries, Jane Doe is available at j.doe@example.com."

B. Summarize Teams Meeting Transcript

Provide the transcript and ask for what you need:

Prompt: "Summarize the following Teams meeting transcript. Focus on key decisions made and action items assigned: [Paste Transcript Here]"
For specific insights: "Analyze this Teams meeting transcript and identify the main arguments presented for and against Project X. Also, list any unresolved questions: [Paste Transcript Here]"

C. Performance Review Wording Improvements

AI can help you phrase feedback constructively.

Prompt: "I need to give feedback to an employee who is excellent at technical tasks but needs to improve their communication with non-technical team members. Suggest some phrases I can use in their performance review that are constructive and encouraging."
With specific examples: "Help me rephrase this performance review comment to be more positive and constructive: 'John consistently fails to explain technical concepts clearly to the marketing team.' I want to highlight the need for improvement without being overly negative. Perhaps suggest he 'could further develop his ability to tailor technical explanations for diverse audiences'."

Experiment with different phrasing and levels of detail to see how the models respond!

6. Adding Context to a Conversation

You can enhance the model's understanding and the relevance of its responses by providing additional context.

A. Access the existing knowledge base (RAG - Retrieval Augmented Generation)

Our Open WebUI setup may be connected to an internal knowledge base (e.g., company policies, project documentation, past reports). When you ask a question, the model can search this knowledge base to find relevant information and use it to formulate its answer. This is often referred to as Retrieval Augmented Generation (RAG).

How it Works (General Idea): You don't always need to do anything special. If RAG is enabled, simply asking a question related to internal information might trigger the model to search the connected knowledge base.
Being Specific: If you suspect information exists in the knowledge base, you can prompt the model accordingly: "According to our internal product development guidelines, what are the steps for submitting a new feature proposal?"

(Note: The specific way RAG is implemented and how you interact with it will be demonstrated. There might be an indicator or option to ensure the knowledge base is being searched.)

B. Having a model search the web

Some models or configurations within Open WebUI might allow the AI to access and retrieve information from the public internet.

Availability: This feature depends on the specific model and our internal configuration. Check the interface or ask if a particular model has web Browse capabilities.
How to Use: If enabled, you can ask questions about current events, recent research, or general public knowledge. For example: "What are the latest trends in renewable energy?"
Caution: Always critically evaluate information retrieved from the web, even if provided by an AI.

C. Uploading files

Open WebUI allows you to upload documents directly into the chat interface to provide context for your conversation. This is particularly useful for tasks like summarizing documents, answering questions about a specific text, or analyzing data within a file. The vision-enabled models (Chuck and Gemma) can also analyze images you upload.

How to Upload: Look for an attachment icon (often a paperclip) or an "Upload File" button near the text input area.
Supported File Types: Check for information on supported file types (e.g., .txt, .pdf, .md, .csv, .png, .jpg).
Example Prompt after Uploading a PDF report: "Summarize the key findings in the uploaded report."
Example Prompt after Uploading an image (with Chuck or Gemma): "Describe the main elements in this image." or "What type of architecture is shown in this picture?"

D. Adding a URL to the conversation

You might be able to provide a web link (URL) to the model and ask it to process the content of that page.

How to Use: Simply paste the URL into the chat box along with your instruction.
Example Prompt: "Summarize the main points of this article: [Paste URL here]" or "What are the key arguments presented on this webpage: [Paste URL here]?"
Limitations: The model's ability to access and process URLs can depend on website structures, paywalls, and security restrictions.

7. Why Ask CHUCK (and our other in-house models) and not other AI tools?

While many external AI tools are available, using our internal Open WebUI platform with models like Chuck, Mitch, and Gemma offers significant advantages for our organization:

Data Security and Confidentiality: This is the most critical reason. When you use our in-house models, all your prompts, the data you upload, and the AI's responses remain within our secure organizational network. This protects our sensitive information, client data, and intellectual property from being exposed to third-party AI providers.
Customization and Relevance: Our models can potentially be fine-tuned or connected to our internal knowledge bases (as discussed with RAG). This means they can provide answers that are more relevant to our specific context, projects, and processes.
Control and Compliance: By hosting these tools internally, we have greater control over their use and can ensure compliance with data governance and industry regulations.
Cost-Effectiveness (Potentially): Depending on usage, hosting our own models can be more cost-effective than paying for commercial AI services for a large number of users.
Reduced Risk of External Policy Changes: We are not subject to the changing terms of service, data usage policies, or access limitations that can occur with external AI tools.
Focused Development: As we learn how these tools are used, we can focus on developing or deploying models that best suit our organization's evolving needs.

By using Chuck, Mitch, and Gemma through Open WebUI, you are leveraging powerful AI capabilities in a secure and controlled environment, tailored to support our organization's success.