Advanced and agentic RAG

Techniques beyond single-shot retrieval: query transformation, graph-based retrieval and agents that search iteratively.

Changes fast: check the latest updates too. Reviewed 17 Oct 2026.

Key ideas

• Query rewriting / HyDE / multi-query: turn vague questions into better search queries.
• Agentic RAG: give the agent search tools and let it run several searches, read documents and decide when it has enough.
• GraphRAG: extract entities and relationships into a graph for questions that span many documents ("what are the main themes").
• Text-to-SQL for structured data: retrieve table schemas and examples, generate SQL, run read-only, validate.
• Long-context models can replace RAG for small, static corpora; compare cost and accuracy.

Common mistakes

• Adding GraphRAG before simple hybrid search plus reranking has been measured.

Go deeper

• Microsoft GraphRAG