Glossary

Retrieval-augmented generation (RAG)

An architecture that retrieves relevant documents at query time and feeds them to a language model so its answers are grounded in sources.

Also called: RAG

Retrieval-augmented generation combines a search step with a generation step: given a query, a retriever fetches the most relevant passages from an external corpus, and a large language model then generates its answer conditioned on both the query and the retrieved text, rather than on its training data alone.

The retrieval step usually relies on a vector database that stores document chunks as embeddings and finds the closest matches to the query's embedding, a form of semantic search; some implementations instead or additionally query a knowledge graph or a traditional keyword index. This differs from fine-tuning, which bakes new knowledge into the model's weights; RAG keeps knowledge external and swappable, and the retrieved passages let a system cite its sources.

RAG matters because it lets an LLM answer questions about current, private, or domain-specific material without retraining, and it reduces, but does not eliminate, hallucination by grounding answers in retrieved text. Common pitfalls include retrieving irrelevant or outdated passages, chunking documents in ways that break context, and a model that ignores or contradicts the retrieved evidence anyway.

Last reviewed September 22, 2026

In the index now

Related terms

Large language model (LLM) A neural network trained on vast amounts of text to predict and generate language, the technology behind most modern AI assistants. Vector database A database optimized to store numeric embeddings and quickly find the ones most similar to a given query vector. Embeddings Numeric vector representations of text, images, or other data, positioned so similar items sit close together in the vector space. Semantic search Search that matches results to a query's meaning rather than its exact keywords, using vector similarity between embeddings. Hallucination When a language model generates confident, fluent output that is factually wrong or unsupported by its source material. Knowledge graph A network of entities and the labeled relationships between them, used to represent and query structured knowledge.

Related tools

Chroma Open-source embedding database designed for simple local-first development, with a managed serverless cloud tier. Pinecone Fully managed, cloud-only vector database for similarity search and retrieval-augmented generation at scale. Ragas Open-source Python framework of reference-free metrics purpose-built for evaluating retrieval-augmented generation (RAG) pipelines. Vanna AI Open-source RAG framework that trains an LLM on a database's schema and past queries to turn questions into SQL.

Related guides

How to choose a vector database Pick a vector database by deployment model, scale and search type, not by benchmark charts — most teams' real constraint is who runs the cluster. How to choose an AI analytics assistant AI analytics assistants are only as accurate as the context they're given — choose by what they query and how that context is governed, not the demo. How to choose an LLM observability tool Pick an LLM tool by whether you need tracing, scoring or a gateway first — most teams eventually need all three, rarely from one vendor.