← Home

assistants-knowledge-bases

Articles (4)

FAQs (7)

Can I update or version documents?

Yes. You can replace documents in a knowledge base at any time by removing the old file and uploading the new one. SecureAI does not maintain automatic version history for uploaded documents, so you manage versions by controlling which files are in the knowledge base.

How to replace a document

  1. Navigate to Knowledge in the left sidebar.
  2. Open the knowledge base containing the document.
  3. Remove the outdated document (click the delete icon next to the file).
  4. Upload the replacement document.
  5. Wait for indexing to complete.

The AI immediately starts using the new document for future queries. No re-attachment to conversations is required -- if the knowledge base is already attached, the updated content is used automatically.

Why you should replace rather than keep both versions

If you upload a new version alongside the old one, the AI may retrieve conflicting information. For example, a 2024 parts catalog and a 2025 catalog for the same product line could return different part numbers for the same vehicle. Always remove the outdated version before uploading the replacement.

Tips for managing document versions

  • Name files with version or date indicators: Use names like Brake-Components-Catalog-2025.pdf rather than catalog.pdf. This makes it clear which version is currently loaded.
  • Test after replacing: Ask a question you know the answer to from the new document. Verify the AI cites the updated source.
  • Keep originals outside SecureAI: Store prior versions on your local drive or shared network folder for reference. SecureAI does not archive removed documents.

Related articles

Can assistants call APIs and use tools?

Yes. SecureAI assistants can call external APIs and use custom tools during conversations, giving them access to live data like parts inventory, pricing, shop management systems, and more.

How it works

An administrator creates a custom tool -- a Python function that calls an external API or service. The tool is then assigned to one or more assistants. When a user asks a question that needs external data, the assistant automatically decides whether to call the tool, runs it, and incorporates the results into its response.

For example, a "Parts Inventory Lookup" tool could query your parts database and return current stock levels and pricing when a user asks "Do we have part number 12345 in stock?"

What assistants can do with tools

Capability Example
Query external APIs Check live parts inventory or pricing
Look up internal databases Search shop management system records
Call third-party services Pull vehicle data from a VIN decoder API
Perform calculations Convert measurements or compute estimates
Retrieve real-time data Check order status or shipping updates

What you need

  • Admin access is required to create and manage tools. Regular users cannot create tools but can use assistants that have tools assigned to them.
  • Tools are written in Python and configured through the SecureAI admin panel under Workspace > Tools.
  • The external API or service must be network-accessible from your SecureAI deployment.

Do tools work automatically?

Once an admin assigns a tool to an assistant, users don't need to do anything special. The assistant reads the tool's description and decides when to use it based on the user's question. Users interact with the assistant the same way they always do -- the tool calls happen behind the scenes.

If an assistant has multiple tools, it can call more than one in a single response when needed.

Limitations

  • Tools require an admin to set up. Users cannot add their own tools.
  • Tool execution depends on the external API being available. If the API is down, the tool call will fail and the assistant will let the user know.
  • Each tool call adds processing time to the response. Assistants with many complex tools may respond more slowly.
  • Tool access follows your organization's security policies. Admins control which assistants have which tools.

Getting started

  • Admins: See Building Custom Tool Integrations for step-by-step setup instructions.
  • Users: Ask your administrator which assistants have tools enabled, or look for assistants with descriptions mentioning live data, lookups, or integrations.
  • Creating assistants: See How to Create an Assistant for the full assistant setup guide.
Can assistants use multiple knowledge bases?

Yes. You can attach more than one knowledge base to an assistant or to a single conversation. When multiple knowledge bases are attached, SecureAI searches all of them and combines the most relevant results to generate its answer.

How to attach multiple knowledge bases

In a conversation

  1. Start a new conversation or open an existing one.
  2. Click the + icon in the message input area (or use the Knowledge selector above the chat).
  3. Select the first knowledge base you want to use.
  4. Repeat to add additional knowledge bases -- there is no limit to how many you can attach.
  5. Ask your question. SecureAI searches across all attached knowledge bases for relevant content.

On an assistant (admin)

If your organization uses custom assistants (model presets configured by an administrator):

  1. Open Workspace > Models in the admin panel.
  2. Select the assistant you want to configure (or create a new one).
  3. In the assistant's knowledge settings, add one or more knowledge bases.
  4. Save the assistant. Every conversation using that assistant will automatically search the attached knowledge bases.

How retrieval works with multiple knowledge bases

When you ask a question, SecureAI queries all attached knowledge bases simultaneously. It ranks the retrieved chunks by relevance regardless of which knowledge base they came from, then uses the top results as context for its response. Source citations in the answer indicate which knowledge base and document each piece of information came from.

This means you do not need to merge documents into a single knowledge base just to search them together -- keep them organized by topic or source and attach whichever ones are relevant.

Tips for using multiple knowledge bases effectively

  • Separate by subject area -- keep parts catalogs, service procedures, and warranty policies in distinct knowledge bases. This lets you mix and match based on the question at hand.
  • Avoid duplicate content -- if the same document appears in two attached knowledge bases, the AI may retrieve redundant or conflicting chunks. Keep each document in one knowledge base only.
  • Watch for version conflicts -- if you attach a knowledge base with a 2024 catalog and another with a 2025 catalog for the same product line, the AI may mix information from both years. Remove outdated versions or use only the current knowledge base when accuracy matters.
  • Start focused, then expand -- attach only the knowledge bases relevant to your question. Attaching every available knowledge base increases noise and may reduce answer quality.

Can knowledge bases be versioned?

SecureAI does not have built-in version control for knowledge bases. Each knowledge base reflects its current contents -- when you add, remove, or replace a document, the change takes effect immediately.

To manage versions manually:

  • Include the version or date in the knowledge base name (e.g., Brake Components 2025, Brake Components 2024-Archive).
  • Keep only current documents active -- move outdated documents to a separate archive knowledge base that you do not attach to everyday conversations.
  • Replace rather than accumulate -- when a new edition of a catalog arrives, remove the old document and upload the new one in the same knowledge base rather than keeping both.

This approach gives you a clear history of what changed and prevents the AI from mixing outdated and current information.

Related articles

Can documents be private, shared with a team, or restricted by role?

Yes. SecureAI supports all three levels of document visibility through its knowledge base and assistant sharing settings. You control who can access your uploaded documents by choosing where and how you create your knowledge bases.

Private documents

When you create a knowledge base under your own account, it defaults to private. Only you can see the knowledge base, attach it to conversations, or query its contents. Other users -- including administrators -- cannot browse or search your private knowledge base documents through the SecureAI interface.

Private knowledge bases are ideal for:

  • Personal reference materials you don't need to share
  • Draft documents you're still refining before team use
  • Sensitive information relevant only to your role (e.g., individual customer notes)

Shared documents

To share documents with your team, create a workspace-level knowledge base. Workspace knowledge bases are visible to all users in the workspace. Any team member can attach them to conversations and query the documents inside.

To share a knowledge base:

  1. Navigate to Knowledge in the left sidebar.
  2. Create a new knowledge base (or open an existing one).
  3. Set the visibility to Workspace (or ask your administrator to move it to the workspace level if you originally created it as personal).
  4. Upload the documents you want to share.

All workspace members can now attach this knowledge base to their conversations.

Good to know: Sharing a knowledge base gives other users read access to its contents through AI responses. They cannot download the original files or modify the knowledge base unless they have admin permissions.

Restricting access by role

SecureAI's role system controls what users can do with knowledge bases:

Capability User role Admin role
Create personal knowledge bases Yes Yes
Attach workspace knowledge bases to conversations Yes Yes
Create workspace-level knowledge bases Depends on admin settings Yes
Upload documents to workspace knowledge bases Depends on admin settings Yes
Delete workspace knowledge bases No Yes
Manage knowledge base visibility settings No Yes

Administrators can restrict whether regular users are allowed to create workspace knowledge bases or upload documents to shared collections. This is configured in Settings > User Permissions in the admin panel.

If you need access to a knowledge base you can't see, or want to restrict a shared knowledge base to a specific group, contact your SecureAI administrator.

Assistants and document access

When you create an assistant and attach a knowledge base to it, the assistant's visibility setting controls who can use those documents indirectly:

  • Private assistant with a private knowledge base -- only you can access the documents through that assistant.
  • Public assistant with a workspace knowledge base -- all workspace users can query the documents through the assistant.
  • Public assistant with a private knowledge base -- other users can use the assistant but the knowledge base content is accessible through the assistant's responses. Consider whether the documents should be in a workspace knowledge base instead if you intend to share the assistant widely.

Summary

Goal How
Keep documents private Create a personal knowledge base (default)
Share documents with the whole team Create a workspace-level knowledge base
Limit who can create shared knowledge bases Admin configures user permissions
Share documents through an assistant Attach knowledge base to a public assistant

For step-by-step instructions on creating and managing knowledge bases, see How to Create and Manage Knowledge Bases. For details on assistant sharing, see How to Create an Assistant.

How does RAG work in SecureAI?

RAG (Retrieval-Augmented Generation) is how SecureAI searches your uploaded documents and uses them to answer questions. Instead of relying only on what the AI model was trained on, RAG pulls in relevant information from your knowledge bases so the AI can give answers grounded in your organization's actual data.

How document search works step by step

When you ask a question with a knowledge base attached, SecureAI follows this process:

  1. Your question is converted into a search query. SecureAI transforms your message into a numerical representation (called an embedding) that captures its meaning.
  2. The system searches your knowledge base. It compares your query against all the document chunks stored in the knowledge base and finds the ones most closely related to what you asked.
  3. The most relevant chunks are retrieved. SecureAI selects the top-matching passages from your documents. These are the "retrieval" part of RAG.
  4. The AI generates a response using those chunks. The retrieved passages are included as context alongside your question, and the AI model produces an answer grounded in that content. This is the "augmented generation" part.
  5. Sources are cited. The response includes references to the specific documents and passages the AI used, so you can verify the information.

What is a knowledge base?

A knowledge base is a collection of documents you upload to SecureAI. When you upload a file, SecureAI:

  • Parses the document -- extracts text from PDFs, Word files, spreadsheets, and other supported formats.
  • Splits the text into chunks -- breaks long documents into smaller passages (typically a few hundred words each) so the search can return focused, relevant sections rather than entire documents.
  • Creates embeddings -- converts each chunk into a numerical vector that represents its meaning. These embeddings are stored in a vector database for fast similarity search.

When does RAG activate?

RAG only runs when a knowledge base is attached to your conversation or assistant. If no knowledge base is attached, the AI responds using only its built-in training data.

You can attach a knowledge base by:

  • Clicking the + icon in the message input area and selecting a knowledge base.
  • Using the Knowledge selector above the chat area.
  • Using an assistant that has knowledge bases pre-configured by an administrator.

What affects the quality of RAG results?

Factor Impact
Document quality Clean, well-structured documents with clear headings produce better chunks and more accurate retrieval. Scanned PDFs without OCR or documents with heavy formatting noise may not parse well.
Question specificity Specific questions (e.g., "What is the torque spec for a 2024 Camry front brake caliper?") retrieve more relevant results than vague ones (e.g., "Tell me about brakes").
Knowledge base scope Attaching only the knowledge bases relevant to your question reduces noise. Attaching every available knowledge base can dilute results.
Document size and chunking Very short documents may not provide enough context per chunk. Very long documents with mixed topics may produce chunks that blend unrelated content.
Recency Knowledge bases reflect their current contents. If you need up-to-date information, make sure the latest version of a document has been uploaded.

Limitations to be aware of

  • RAG does not search the internet. It only searches documents you have uploaded to your knowledge bases.
  • Context window limits apply. The AI can only use a limited number of retrieved chunks per response. If your question spans many topics across many documents, some relevant information may not be included.
  • The AI may still generate text beyond what the documents say. While RAG grounds the AI in your data, it can still combine retrieved information with its general knowledge. Check source citations to confirm which parts of an answer came from your documents.
  • Exact keyword matches are not guaranteed. RAG uses semantic (meaning-based) search, not keyword search. A question about "replacing brake pads" will match documents about "brake pad replacement" even if the exact wording differs -- but it may also match loosely related content about brakes in general.

Related articles

What file types and sizes are supported for upload?

SecureAI supports a range of document and image file types for upload to knowledge bases and conversations. This page covers supported formats, size limits, and how many documents you can upload.

Supported file types

Documents

Format Extension Notes
PDF .pdf Text-based PDFs are fully indexed. Scanned PDFs without an OCR text layer are not searchable -- run them through OCR software before uploading.
Plain text .txt Best for structured data like parts lists or logs.
Markdown .md Headings and formatting are preserved during indexing.
Microsoft Word .docx Text and basic formatting are extracted. Embedded images are not indexed.
Microsoft Excel .xlsx Tabular data is extracted. Works well for parts cross-reference tables and fitment data.
Microsoft PowerPoint .pptx Slide text is extracted. Embedded images are not indexed.
CSV .csv Good for structured datasets like parts catalogs or price lists.

Images (in conversations)

Format Extension Notes
JPEG .jpg, .jpeg Best for photos of parts, labels, or VIN plates.
PNG .png Best for screenshots and diagrams.
GIF .gif Static images only; animation frames are not analyzed.
WebP .webp Supported for general image uploads.

Images uploaded in a conversation are sent directly to the AI model for visual analysis. Images uploaded to a knowledge base are not visually analyzed -- only any embedded text metadata is indexed.

File size limits

Limit Value
Maximum file size 100 MB per file
Recommended file size Under 25 MB for fastest processing
PDF page limit No hard limit, but PDFs over 100 pages may take significantly longer to index. Consider splitting very large PDFs into smaller sections.

Large files are processed in the background after upload. You will see a progress indicator while indexing completes. Documents are not searchable until indexing finishes.

How many documents can I upload?

There is no fixed limit on the number of documents per knowledge base. However, keep these practical considerations in mind:

  • Retrieval quality decreases with unfocused collections. A knowledge base with 500 unrelated documents will return less relevant results than several focused knowledge bases with 50 documents each. See How should I structure knowledge bases? for guidance.
  • Indexing time scales with volume. Uploading hundreds of documents at once may take several minutes to fully index. Upload in batches if you need some documents available immediately.
  • Storage quotas may apply. Your administrator may configure storage limits for your organization. If you receive an error when uploading, check with your administrator about available storage.

Tips for automotive aftermarket documents

  • Parts catalog PDFs: Split large multi-section catalogs into one PDF per product line or category. This improves both indexing speed and retrieval accuracy.
  • Fitment data: CSV or Excel format works best for fitment tables. The AI can cross-reference rows more accurately than when parsing PDF tables.
  • Technical service bulletins: Upload as individual PDFs rather than combined volumes. This lets the AI retrieve the specific bulletin relevant to a question.
  • Scanned documents: If you have scanned paper manuals or spec sheets, run them through OCR software first. Without a text layer, SecureAI cannot search the content.

What happens if my file is not supported?

If you try to upload an unsupported file type, SecureAI will display an error message and the file will not be added. Convert the file to a supported format (PDF or plain text work for most cases) and try again.

Related articles

What is the difference between an assistant and a model?

A model is a raw AI engine -- it reads your messages and generates responses. An assistant is a pre-configured package that wraps a model together with a system prompt, knowledge bases, and tools so it behaves in a specific way out of the box.

Models

A model (such as GPT-4o, Claude, or Gemini) provides general-purpose intelligence. When you select a model directly in the model selector, you get its default behavior with no additional instructions, no attached documents, and no tool integrations. You are starting from a blank slate every conversation.

Models differ in capability, speed, and cost:

Trait What it means
Capability Some models are better at reasoning, writing, or code. Others are faster but less detailed.
Speed Smaller models respond quickly. Larger models take longer but produce more thorough answers.
Cost Each message uses tokens. Larger models cost more tokens per response.

Assistants

An assistant bundles a model with extra configuration so users don't have to set it up themselves:

  • System prompt -- instructions that shape the assistant's tone, focus, and behavior (e.g., "You are a parts specialist for the automotive aftermarket").
  • Knowledge bases -- documents the assistant can search automatically using RAG, such as parts catalogs, service procedures, or warranty policies.
  • Tools -- external API integrations the assistant can call, like inventory lookups or VIN decoders.

When you select an assistant instead of a plain model, all of that configuration is applied instantly.

When to use each

Scenario Choose
Quick, general-purpose question with no specialized context A model
Conversation that needs your organization's documents or live data An assistant with the right knowledge bases and tools
Repeatable workflow (e.g., parts lookup, estimate drafting) An assistant configured for that job

If your administrator has created assistants for your team, those will appear alongside regular models in the model selector. Select one to start a conversation with its full configuration applied.

A practical example

Imagine your organization has a Parts Lookup Assistant:

  • Model: A fast model optimized for quick answers.
  • System prompt: "You are a parts specialist. Always include part numbers, fitment, and pricing when available."
  • Knowledge bases: Your aftermarket parts catalogs and pricing sheets.
  • Tools: A live inventory lookup that checks current stock levels.

If you ask "Do we have brake pads for a 2022 Camry?" using this assistant, it searches your parts catalogs, calls the inventory tool, and responds with part numbers, prices, and stock status -- all without you configuring anything.

If you asked the same question using the raw model, it would only have its general training to draw on and could not access your catalogs or inventory system.

Related articles