Article compares three RAG architectures: standard RAG for single-hop lookups, Graph RAG for multi-hop relationship queries, and Agentic RAG for dynamic multi-source tasks.

Standard RAG embeds documents into vectors and retrieves similar chunks via similarity search, but fails when a query requires connecting facts spread across multiple documents.

Graph RAG adds a knowledge graph layer where an LLM extracts entities and relationships during indexing, then traverses these connections during retrieval instead of relying on embedding similarity alone. This enables multi-hop queries—for example, finding that a checkout service will be affected by maintenance on a cluster it depends on, even when the intermediate connection (the payments API) doesn't appear in the original query.

Agentic RAG uses an LLM agent that decides at query time which tools to invoke, which sources to query, and in what order, rather than following a fixed retrieval pipeline.

These three architectures solve different query types: single-hop factual lookups use standard RAG, multi-hop relationship queries use Graph RAG, and dynamic multi-source tasks with tool use use Agentic RAG.