Saltar al contenido principal

RAG (Retrieval-Augmented Generation)

RAG allows the AI assistant to reference your organization's uploaded documents when answering questions. This grounds responses in real documentation rather than relying solely on the AI model's training data.

How it works​

User asks a question
       │
       ▼
┌─────────────────────┐
│ 1. Extract keywords │
│    from question +   │
│    page context      │
└──────────┬──────────┘
           │
           ▼
┌─────────────────────┐
│ 2. Search document  │
│    embeddings for   │
│    relevant chunks  │
└──────────┬──────────┘
           │
           ▼
┌─────────────────────┐
│ 3. AI generates     │
│    response using:  │
│    - Question       │
│    - Page context   │
│    - Knowledge      │
│    - RAG chunks     │
└─────────────────────┘

Supported file types​

  • PDF — Policy documents, SOPs, manuals
  • Text files — Plain text documentation
  • Additional formats may be supported — check the Backoffice for the latest list

File association​

Files can be associated with:

  • A specific domain — The file is only used for questions on that domain
  • A specific URL pattern — Even more precise scoping
  • Organization-wide — Used as general context for any question

Best practices​

  • Upload structured documents — Well-organized docs with clear headings produce better results
  • Keep files current — Outdated files lead to outdated answers
  • Scope appropriately — Associate files with specific domains rather than organization-wide when possible
  • Size matters — Smaller, focused documents work better than large monolithic files

For admins​

See RAG Files for instructions on uploading and managing files in the Backoffice.