19. Technical specifications for RAG
RAG allows your agent to respond using real information (documents) instead of relying only on general knowledge. This helps limit responsibilities and prevents the agent from hallucinating answers or falling short when it lacks sufficient information.
19.1. Supported file types
Currently, you can use:
- Word documents (.docx)
- PDFs
- Images
19.2. Important considerations
A RAG is the set of additional information that the agent uses as context to generate more accurate and realistic responses. It works with text-based content, has no strict page limit, and allows you to include all relevant information needed, preventing the agent from giving incomplete or insufficient answers.
The model performs best with:
- Clear text
- Well-structured paragraphs
It may struggle with:
- Complex tables
- Scanned PDFs
- Highly visual layouts
19.3. Document quality = response quality
Good RAG:
- Clear information
- Well-structured
- Noise-free
Poor RAG:
- Disorganized text
- Duplicate information
- Very long documents without structure
19.4. Be specific with your content
Better:
- FAQs
- Scripts
- Clear policies
- Concrete information
Worse:
- Generic documents
- Ambiguous information