How to Create an Assistant

Assistants in SecureAI are custom AI configurations that combine a base model with specific instructions, knowledge, and behavior settings. Think of them as specialized versions of the AI tailored for particular tasks — a parts lookup specialist, a warranty claim helper, or a technical service bulletin interpreter.

This guide walks through creating, configuring, and sharing assistants in your SecureAI workspace.

What Is an Assistant?

An assistant wraps a base AI model with:

• System prompt — persistent instructions that shape how the model responds
• Knowledge base — documents the assistant can reference for grounded answers
• Model parameters — temperature, context window, and other generation settings
• Name and description — so your team knows what the assistant does

When a user starts a conversation with an assistant, these settings are applied automatically. The user doesn't need to write setup prompts or attach documents each time.

Creating a New Assistant

Step 1: Open the Assistant Builder

1. Click the Workspace icon in the left sidebar
2. Select the Assistants tab (you may see it labeled "Modelfiles" in some versions)
3. Click Create New Assistant (or the + button)

Step 2: Set the Basics

Choose a name that tells users exactly what the assistant is for. Avoid generic names like "Helper" or "Assistant 1."

Step 3: Write the System Prompt

The system prompt is the most important part of your assistant. It defines the assistant's role, behavior, and constraints.

A good system prompt includes:

• Role definition — who the assistant is and what it does
• Response format — how answers should be structured
• Boundaries — what the assistant should and shouldn't do
• Domain context — industry-specific terms and expectations

Example system prompt for a parts lookup assistant:

You are a brake parts specialist for the automotive aftermarket industry.

When a user asks about a part:
1. Identify the specific component (rotor, pad, caliper, hose, etc.)
2. Check the knowledge base for matching part numbers
3. Provide the part number, manufacturer, and compatible vehicles
4. If cross-references exist, list alternative manufacturer part numbers

Response format:
- Lead with the most likely match
- Include vehicle fitment (year, make, model, engine)
- Note any superseded or discontinued part numbers
- Flag if the part requires additional components for installation

Do not guess part numbers. If the knowledge base does not contain a match, say so clearly and suggest the user verify with the manufacturer catalog.

System prompt tips:

• Be specific rather than general. "Answer questions about brakes" is too vague. Define exactly how the assistant should handle different query types.
• Include output format instructions. If you want tables, bullet lists, or specific data fields in every response, say so explicitly.
• Set boundaries. Tell the assistant what NOT to do — this prevents unhelpful tangents.
• Keep it under 1,000 words. Extremely long system prompts can reduce the effective context window available for conversation.

Step 4: Attach a Knowledge Base

If your assistant should reference specific documents:

1. In the assistant configuration, find the Knowledge section
2. Click Add Knowledge Base or Select Collection
3. Choose from your existing knowledge bases, or create a new one

The assistant will use RAG (retrieval-augmented generation) to search these documents when answering questions. For best results, follow the Knowledge Base Design Best Practices guide.

When to attach a knowledge base:

Step 5: Configure Model Settings

Adjust these parameters based on your assistant's purpose:

For parts lookup and technical assistants, use low temperature (0.1-0.3). You want consistent, accurate answers, not creative variation.

For writing or brainstorming assistants, use moderate temperature (0.5-0.7). This allows more varied and interesting outputs.

Step 6: Save and Test

1. Click Save to create the assistant
2. Start a new conversation and select your assistant from the model dropdown
3. Test with realistic queries your team would actually ask
4. Refine the system prompt based on the results

Testing checklist:

• Does the assistant stay in character with the system prompt?
• Does it correctly reference knowledge base documents?
• Does it handle edge cases (unknown parts, ambiguous queries)?
• Are response formats consistent?
• Does it respect the boundaries you set?

Sharing Assistants with Your Team

Once you've built and tested an assistant:

1. Go to the assistant's settings
2. Set Visibility to "Public" to make it available to all users in your workspace, or "Private" to keep it personal
3. Optionally, add the assistant to a Workspace group so specific teams can find it

Public assistants appear in the model dropdown for all workspace users. Consider creating a naming convention so team members can quickly identify relevant assistants (e.g., prefix with the department or function: "Parts - Brake Lookup", "Service - TSB Search").

Managing and Updating Assistants

When to Update

• Knowledge base changes — new catalogs, updated part numbers, revised procedures
• Workflow changes — new response formats, additional fields, updated policies
• User feedback — if users report incorrect behavior, adjust the system prompt
• Model updates — when new base models become available, test and potentially switch

Editing an Existing Assistant

1. Navigate to Workspace > Assistants
2. Click on the assistant you want to modify
3. Make your changes to the system prompt, knowledge base, or settings
4. Click Save

Changes take effect for new conversations immediately. Existing conversations continue using the settings that were active when the conversation started.

Cloning an Assistant

To create a variant of an existing assistant:

1. Open the assistant you want to clone
2. Click Clone (or manually create a new assistant and copy the system prompt)
3. Modify the clone for its specific purpose
4. Save with a new name

This is useful when you need similar assistants for different product lines or departments.

Example Assistants for Automotive Aftermarket

Here are starter configurations you can adapt:

Parts Cross-Reference Assistant

• System prompt focus: Cross-reference OEM and aftermarket part numbers across manufacturers
• Knowledge base: Manufacturer cross-reference tables, interchange guides
• Temperature: 0.1
• Use case: Parts counter staff looking up equivalent parts

Warranty Claim Helper

• System prompt focus: Guide users through warranty claim documentation requirements
• Knowledge base: Warranty policies, claim forms, coverage terms by manufacturer
• Temperature: 0.2
• Use case: Service advisors preparing warranty submissions

Technical Bulletin Search

• System prompt focus: Find and summarize relevant TSBs and recalls for a given vehicle
• Knowledge base: TSB database, recall notices, service campaigns
• Temperature: 0.1
• Use case: Technicians diagnosing recurring issues

Customer Communication Drafter

• System prompt focus: Draft professional customer communications for estimates, approvals, follow-ups
• Knowledge base: Communication templates, company style guide
• Temperature: 0.6
• Use case: Service advisors writing customer-facing messages

Troubleshooting

Assistant doesn't use the knowledge base

• Verify the knowledge base is attached in the assistant settings
• Check that the knowledge base contains relevant documents (not empty)
• Try rephrasing your query to use terms that appear in the documents
• See Knowledge Base Design Best Practices for document preparation tips

Responses are inconsistent

• Lower the temperature setting (try 0.1 for factual tasks)
• Add more specific instructions to the system prompt about expected output format
• Include examples of good responses in the system prompt

Assistant ignores system prompt instructions

• Check that the system prompt is saved (reopen the assistant settings to verify)
• Simplify the system prompt — conflicting instructions can cause the model to ignore some of them
• Move the most critical instructions to the beginning of the system prompt
• Ensure instructions are stated positively ("always include part numbers") rather than only negatively ("don't forget part numbers")

Assistant is too slow

• Consider switching to a faster base model (see How to Choose the Right AI Model)
• Reduce the attached knowledge base size — smaller, focused collections retrieve faster
• Shorten the system prompt if it's very long

Next Steps

• Knowledge Base Design Best Practices — optimize the documents your assistants reference
• How to Choose the Right AI Model — pick the right base model for your assistant
• Team Collaboration with SecureAI — share assistants and workflows across your organization