Retrieval-Augmented Generation, commonly called RAG, adds a search step before an AI system produces an answer. Instead of relying only on general model knowledge, the application retrieves relevant information from sources you approve.
The core flow
Content is prepared and indexed. A user asks a question. The system searches for relevant passages, sends selected context to the model, and generates an answer that can include source references.
What RAG can search
Common sources include service pages, product catalogs, policies, PDFs, knowledge-base articles, support tickets, SQL database records, project descriptions, and internal documentation.
What RAG does not solve automatically
It does not guarantee truth, security, or good user experience. Access controls, source quality, retrieval evaluation, prompt design, monitoring, and human review still matter.
When to use it
RAG is useful when answers must reflect changing or private business knowledge and when the organization wants more control than a general chatbot provides.