Skip to main content

RAG

RAG stands for retrieval-augmented generation.

What Is RAG​

RAG retrieves relevant information from external sources and gives it to a language model before generating an answer.

It helps models answer with project-specific or up-to-date context.

How To Use RAG​

Split documents into chunks, create embeddings, store them, retrieve relevant chunks, and pass them into the model prompt.

Basic Example​

const docs = await retriever.search("How do agents use memory?");
const answer = await model.generate({
  question: "How do agents use memory?",
  context: docs,
});

Common Concepts​

  • Chunking controls document size.
  • Embeddings support semantic search.
  • Retrieval finds relevant context.
  • Evaluation checks whether answers are grounded.

What To Learn Next​

Learn chunking strategies, vector search, reranking, citations, retrieval evaluation, and hybrid search.