Skip to content

Custom Assistants

Custom assistants, referred to as "Models" within the Open WebUI interface, represent a sophisticated approach to AI specialization. These assistants combine carefully crafted system prompts, knowledge bases, and behavioral configurations to create AI personas that excel at specific tasks and workflows.

Why Create Custom Assistants?

A standard assistant is a generalist. A custom assistant is a specialist. You create an assistant with a specific personality for recurring tasks to achieve consistent, high-quality results and accelerate your workflows.

Examples include assistants for: - The Project Expert: An assistant that exclusively answers questions about a specific project, based on all relevant documents you provide in a knowledge store - The Translation Professional: An assistant that always uses the same professional style for German-English translations and ensures correct technical terminology - The Meeting Scribe: An assistant trained to convert unstructured notes into clean minutes with action items and decisions

How Assistants Work: The Technology Behind Them

To understand how to build a good assistant, it's helpful to know its three core components:

🗂️ The Knowledge Store (The Expertise)

This is your assistant's data foundation. Here you upload documents. When you ask a question, the AI first searches this store for relevant information.

🧠 The Foundation Model (The Engine)

This is the underlying language model (e.g., C-KIT-Standard-Creative) that processes language and formulates responses.

📄 The System Prompt (The Control Center)

This is the most important component that you define. It's a detailed instruction to the AI. Here you define the assistant's role, personality, tasks, and boundaries.

Step-by-Step: Creating Your First Assistant

Step 1: Define Idea and Goal

Ask yourself: - What specific problem should my assistant solve? - What task should it take off my hands? - Who will use it? (Just me or my team?) - Example: I want an assistant that helps me understand internal expense guidelines.

Step 2: Prepare Knowledge Base (Optional)

If your assistant should be based on specific knowledge, create a knowledge store and upload relevant documents. - Example: I create a knowledge store named "Expense-Guidelines" and upload the official PDF policy.

Step 3: Write the System Prompt (The Control Center)

This is the crucial step. A good system prompt is precise and clear. Use these building blocks:

Role & Goal

Who is the assistant? What is its main objective? - Example: You are a helpful expert for internal expense guidelines. Your goal is to answer requests clearly and precisely based on official documents.

Behavioral Instructions

How should it behave? What tone should it adopt? - Example: Answer factually and formally. Use bullet points to improve readability. Avoid speculation.

Working with the Knowledge Base

How should it handle documents? - Example: Your answers must be based exclusively on information from the provided context. If you cannot find information, state this clearly.

Formatting Rules

Should the output follow a specific format? - Example: Use Markdown for formatting. Important deadlines or amounts must be bold.

Step 4: Configure Your Assistant in Open WebUI

  1. Click Workspace in the left sidebar
  2. Click Model and create a new assistant with the plus icon
  3. Title: Give a short name (e.g., "Expense Assistant")
  4. Prompt: Insert your carefully crafted system prompt
  5. Assign Technical Model: Choose the appropriate AI model (e.g., C-KIT-Standard-Chat)
  6. Link Knowledge Store: Select your prepared knowledge store (e.g., #Expense-Guidelines)
  7. Save the assistant

Your new assistant is immediately ready to use. Either start a new chat and select your assistant from the list, or click on your assistant under Workspace > Models.

Step 5: Test and Refine

No assistant is perfect immediately. Test thoroughly: - Ask questions whose answers you know - Try unclear or ambiguous formulations - Ask about something not in the documents to see if it follows your instructions

Adjust your system prompt as needed to correct behavior. This iterative process of testing and adjusting is key to creating a truly useful assistant.

Important Limitations

Custom assistants that rely on collecting and attaching relevant text fragments have limitations, particularly with: - Complex, ambiguous, or highly context-dependent tasks - Tasks requiring comprehensive analysis of all available data - Finding specific instances of values across large datasets (since they can only access a limited number of document fragments at once)

For such tasks, it's better to use specialized tools like spreadsheets or databases and get AI assistance with approach and formulas.

Best Practices

  • Be clear and specific with unambiguous language
  • Keep knowledge up to date with regular reviews
  • Use descriptive naming conventions
  • Test thoroughly across various scenarios
  • Understand boundaries - avoid legal advice, medical diagnoses, or financial recommendations
  • Implement quality assurance through regular reviews and feedback

Team Collaboration

Enhance assistant value through: - Sharing across teams for consistency - Collaborative development for continuous improvement - Collective expertise and diverse perspectives

Custom assistants fundamentally transform general AI capabilities into specialized organizational tools that understand your specific needs, follow your established procedures, and deliver consistent quality results for recurring tasks and workflows.