From Naive RAG to Agentic RAG: The Evolution

A language model only knows what it learned during training. That knowledge is frozen at a point in time and can be incomplete or outdated. RAG, Retrieval-Augmented Generation, fixes this by retrieving relevant external information at the moment of the query and feeding it into the model before it answers. The result is an answer that is current and grounded in real sources rather than invented from memory.

The survey describes RAG as having three core components, and it is worth fixing these in mind because everything later builds on them:

• Retrieval: querying external sources (knowledge bases, APIs, vector databases) to find relevant information.
• Augmentation: processing and summarizing the retrieved data to fit the query.
• Generation: combining the retrieved information with the model's own knowledge to produce the answer.