This is a condensed guide covering the three things you need to know to use SecureAI effectively: setting up knowledge bases, writing good prompts, and understanding who can do what. For the full details on any topic, follow the links to the detailed articles.
Part 1: Knowledge Bases
A knowledge base is a collection of documents (PDFs, Word docs, text files) that you upload so the AI can search them when answering questions. Instead of relying only on general knowledge, the AI pulls from your actual catalogs, procedures, and reference materials.
Creating a Knowledge Base
- Click Knowledge in the left sidebar.
- Click + Create a Knowledge Base.
- Give it a clear, descriptive name (e.g., "Brake Components 2024-2025" or "Shop Warranty Procedures" -- not "KB1" or "Test").
- Click Create, then Upload Files to add your documents.
Supported formats: PDF, DOCX, TXT, Markdown, CSV. Scanned PDFs work if they have an OCR text layer.
Using a Knowledge Base in a Conversation
- Start a new conversation.
- Click the + icon in the message input area.
- Select one or more knowledge bases from the list.
- Ask your question. The AI searches the attached knowledge bases for relevant content.
Check the AI's response for source citations -- it shows which document and section it used.
Organizing Tips
| Do | Don't |
|---|---|
| One topic per knowledge base (e.g., "Parts Catalogs" separate from "Warranty Policies") | Dump everything into a single knowledge base |
| Use descriptive file names ("2024-camry-front-brake-service.pdf") | Upload files named "doc_final_v3_REVISED(2).pdf" |
| Replace outdated documents -- delete old, upload new | Keep both old and new versions (causes conflicting answers) |
| Remove cover pages, table of contents, and boilerplate before uploading | Upload raw scanned PDFs with no OCR text layer |
Who Can Do What with Knowledge Bases
| Action | User | Admin |
|---|---|---|
| Search the knowledge base | Yes | Yes |
| Upload to personal workspace | Yes | Yes |
| Upload to shared knowledge base | Configurable | Yes |
| Delete own documents | Yes | Yes |
| Delete anyone's documents | No | Yes |
| Manage knowledge base collections | No | Yes |
Full guide: How to Create and Manage Knowledge Bases and Knowledge Base Design Best Practices
Part 2: Prompting
How you phrase your questions directly affects the quality of the answers. The single most impactful thing you can do is be specific.
The Prompt Formula
For most queries, follow this pattern:
[Context: Your role and what you're trying to accomplish]
[Details: Specific requirements, constraints, preferences]
[Format: How you want the answer structured]
Examples
Vague (weak):
What brake pads fit a Camry?
Specific (strong):
What ceramic brake pads fit a 2023 Toyota Camry LE 2.5L?
Include front and rear part numbers.
Structured request:
I need help with a brake job quote for a customer.
Vehicle: 2022 Honda Civic EX 1.5T
Work requested: Front brake pads and rotors
Please provide:
1. Recommended part numbers (ceramic pads, OEM-equivalent rotors)
2. Estimated labor time
3. Any related items to inspect (calipers, hardware, brake fluid)
Working with Knowledge Bases in Prompts
When you have a knowledge base attached, you can reference it:
Using our Dorman chassis catalog, what is the control arm part number
for a 2021 Toyota RAV4 LE front lower? Include any related hardware kits.
Mentioning the knowledge base by name isn't required (the UI selection is what matters), but it can help the AI focus when multiple knowledge bases are active.
Follow-Up Questions
You don't need to get everything in one prompt. The AI remembers the conversation:
- First: "What are the common causes of a P0300 random misfire on a 2019 Subaru Outback 2.5L?"
- Follow-up: "The customer says it's worse on cold starts. Does that narrow it down?"
- Follow-up: "What's the part number for the ignition coil if we replace all four?"
Start a new conversation when you switch to a different vehicle or topic.
Common Mistakes
| Mistake | Why it hurts |
|---|---|
| Being too vague ("What do I need for a brake job?") | AI doesn't know the vehicle, scope, or brand preferences |
| Multiple unrelated questions in one message | AI answers part of the question and misses the rest |
| Assuming the AI remembers yesterday's conversation | Each new conversation starts fresh -- restate context |
| Not verifying part numbers | Always cross-check against your catalog before quoting a customer |
Full guide: Prompting Best Practices
Part 3: Assistants, Models, and User Restrictions
How the Pieces Fit Together
Workspace
|-- Models (AI engines available to your team)
|-- Knowledge Bases (your uploaded documents)
|-- Assistants (pre-configured combos of model + knowledge + instructions)
|-- Model (one AI engine)
|-- System Prompt (instructions that shape behavior)
|-- Knowledge Bases (one or more document collections)
|-- Tools (optional external API integrations)
Workspace = shared environment for a team. Groups models, knowledge bases, and settings together.
Model = the AI engine that generates responses. Different models have different speed/capability/cost trade-offs. Your admin controls which models are available.
Assistant = a pre-packaged AI setup. Combines a model with specific instructions, knowledge bases, and tools so users don't have to configure anything each time. Select an assistant from the model dropdown to use it.
Creating an Assistant (Admin or Permitted Users)
- Go to Workspace > Assistants and click Create New Assistant.
- Set a clear name and description.
- Write a system prompt that defines the assistant's role, response format, and boundaries:
You are a brake parts specialist for the automotive aftermarket industry.
When a user asks about a part:
1. Identify the specific component
2. Check the knowledge base for matching part numbers
3. Provide the part number, manufacturer, and compatible vehicles
4. List alternative manufacturer part numbers if cross-references exist
Do not guess part numbers. If the knowledge base does not contain
a match, say so clearly.
- Attach relevant knowledge bases.
- Set temperature (0.1-0.3 for factual lookups, 0.5-0.7 for writing tasks).
- Save and test with realistic queries before sharing.
User Roles
SecureAI has three roles. Every user gets exactly one.
| Role | Who it's for | What they can do |
|---|---|---|
| User | Everyone day-to-day | Chat, upload docs to personal workspace, use assistants and tools, share conversations |
| Admin | IT staff, platform managers (2-5 per org) | Everything a User can do, plus: manage users, configure models, set permissions, view analytics, export audit logs |
| Pending | New self-registered users | Log in and update profile only -- cannot chat or access knowledge bases until an admin approves them |
What Admins Can Restrict
Beyond the three roles, admins can fine-tune access through feature-level controls:
| Control | Where to configure | What it does |
|---|---|---|
| Model visibility | Admin Panel > Settings > Models | Hide expensive or experimental models from standard users |
| Chat features | Admin Panel > Settings > Interface | Toggle web search, image generation, code execution, file uploads |
| Shared KB uploads | Admin Panel > Settings > Knowledge Base | Limit shared uploads to admins only |
| Assistant publishing | Admin Panel > Settings > Assistants | Control who can publish assistants to the org directory |
| Registration | Admin Panel > Settings > Authentication | Open (with approval), invite-only, or SSO-only |
Disabled features are hidden entirely -- users don't see grayed-out buttons, the feature just doesn't appear.
Quick Permission Reference
| Action | User | Admin |
|---|---|---|
| Start conversations | Yes | Yes |
| Use all available models | Yes | Yes |
| View other users' conversations | No | No |
| Upload to personal workspace | Yes | Yes |
| Upload to shared knowledge base | Configurable | Yes |
| Create personal assistants | Yes | Yes |
| Publish assistants to org | Configurable | Yes |
| Access Admin Panel | No | Yes |
| Manage users and roles | No | Yes |
| Configure models and providers | No | Yes |
| View analytics and audit logs | No | Yes |
Full guides: Managing User Roles and Permissions, How to Create an Assistant, and How to Share Assistants with Your Team
Quick Reference Card
Getting Started Checklist
- Log in at your organization's SecureAI URL
- Start a new conversation and try a test question
- Check which knowledge bases are available (ask your admin or the AI)
- Attach a relevant knowledge base to your conversation
- Use specific prompts: include year, make, model, engine, and what you need
- Verify part numbers against your catalog before quoting customers
Admin Setup Checklist
- Configure model providers and set visibility per model
- Create knowledge bases and upload core documents
- Build assistants for your team's common workflows
- Set user roles and feature restrictions
- Choose registration mode (open, invite-only, or SSO)
- Review the permission matrix and restrict as needed
Where to Go Next
| Topic | Article |
|---|---|
| Full interface walkthrough | SecureAI Interface Tour |
| Model selection guidance | How to Choose the Right AI Model |
| Document upload troubleshooting | Troubleshooting Document Indexing |
| Team collaboration | Team Collaboration with SecureAI |
| SSO configuration | Configuring SAML SSO or Configure OIDC SSO |
| Content filtering | Content Filtering and Safety Settings |
| Audit and compliance | How to Audit User Activity |