Retrieval-Augmented Generation (RAG)
Definition
Retrieval-Augmented Generation (RAG) is an architecture in which a language model performs targeted research in a connected knowledge base before each answer and integrates the retrieved content as context into its response.
Meaning in a Business Context
RAG is the standard architecture for enterprise AI that needs to work with up-to-date, internal or confidential information. Classic language models only know what was publicly available up to the time of training. A RAG system is given access to its own data source — for example product manuals, contract documents, support tickets or wikis — and can produce precise, evidenced answers from it. This reduces hallucinations, enables source citations per answer and makes rights and permission management manageable, because the language model itself does not contain any data. In the mid-market, RAG is typically used for knowledge portals, customer service chatbots, technical assistance systems and internal search.
Frequently Asked Questions
When is RAG more useful than fine-tuning?
Whenever the knowledge changes frequently, the volume of data is large, or traceability via source citation is required. Fine-tuning pays off for stable style and format requirements, not for factual knowledge.
Do I need my own vector database for RAG?
For production systems in the mid-market: yes. Suitable options include Qdrant, Weaviate or Postgres with pgvector — all available as on-premises variants.