Skip to main content

​
Knowledge Bases & RAG System

SecureAI uses Retrieval-Augmented Generation (RAG) to connect AI models with your private data and documents, ensuring secure, accurate, and context-aware responses.

​
What is RAG?

Retrieval-Augmented Generation (RAG) is an AI technique that combines language models with external knowledge sources. Instead of relying solely on the model’s pre-trained data:
  1. Query Processing: The system analyzes the user prompt to identify relevant knowledge requirements.
  2. Knowledge Retrieval: Semantic search retrieves relevant document chunks from your index.
  3. Context Augmentation: Retrieved context is injected into the model prompt.
  4. Response Generation: The AI generates a factual, grounded response citing your documents.

​
Index Types in SecureAI

Indexes are structured knowledge bases categorized by accessibility:

​
1. Personal Indexes

  • Zero-Knowledge Privacy: Your files and conversations are encrypted with tenant/user keys.
  • Creation:
    • Users: Open the chat interface, click the index selector next to the model picker, switch to the Personal tab, and click +.
    • Admins: Go to Admin ? Index Management ? Create Index and assign it to a specific user.

​
2. Group Indexes

  • Collaboration: Multiple team members can query and contribute documents to the same knowledge base.
  • Creation:
    • Go to Admin ? Group Management ? Create Group, assign members, and link a shared Group Index.

​
3. Global Indexes

  • Company-wide Standards: Centralized repositories for enterprise compliance, legal terms, and operational guidelines.
  • Creation: Go to Admin ? Index Management ? Create Index and set scope to Global.

​
How Indexing Works

When you upload files (PDFs, Word docs, CSV, JSON, Markdown, or plain text): Raw Document --? Text Extraction --? Semantic Chunking --? Vector Embeddings --? Vector Storage (Index)
  1. Document Processing: Text and metadata (timestamps, file names) are extracted and cleaned.
  2. Chunking: Content is split into optimized semantic chunks with overlap to maintain contextual continuity.
  3. Embedding Generation: Chunks are converted into high-dimensional vector embeddings.
  4. Vector Storage: Embeddings are stored in high-performance vector databases for instant similarity retrieval.

​
Best Practices

  • Naming: Use clear, descriptive names for indexes and uploaded files (e.g., HR-Policy-2026 instead of doc1).
  • File Preparation: Ensure uploaded documents are clear text; clean up low-quality scans before uploading.
  • File Limits: Standard uploads support files up to 10MB per batch.
  • Scope Appropriately: Keep sensitive personal notes in Personal Indexes; move team resources into Group Indexes.