Retrieval-augmented generation

Retrieval-augmented generation, or RAG, retrieves relevant material and places it in context for a generative model to use.

Plan for this page

Understand one concept in plain language and connect it to nearby ideas.

You will leave with: A RAG pipeline commonly indexes sources, represents or searches a query, ranks passages, and supplies selected content before generation. Access rules, document freshness, chunking, ranking, context limits, and citation behavior affect the result.
Time: 3–8 min
Before you begin: No prerequisite. Keep one real example in mind.
Do this now: Read the explanation, then compare the example with the common misunderstanding.

Prepared next stepInspect retrieval and tools

Retrieval-augmented generationRetrieval-augmented generation, or RAG, retrieves relevant material and places it in context for a generative model to use.

Embedding
Context window
Grounding
Tool use

Plain explanation

A RAG pipeline commonly indexes sources, represents or searches a query, ranks passages, and supplies selected content before generation. Access rules, document freshness, chunking, ranking, context limits, and citation behavior affect the result.

Example

A support assistant retrieves passages from current manuals before drafting an answer and linking to those passages.

Why it matters

RAG can connect a model to current or private knowledge without retraining it, but reliability depends on both retrieval and generation. It is an architecture pattern, not a factuality guarantee.

Common misunderstanding

Misconception: RAG means the model searches the whole internet perfectly. Correction: it searches a configured collection through a fallible retrieval pipeline.

Related concepts

Follow relationships, not only the alphabet

EmbeddingAn embedding is a numerical vector that represents an item so useful relationships can be measured or learned.Context windowA context window is the limited amount of tokenized information a model can consider together during one generation or inference.GroundingGrounding connects a model’s output to selected external information, observations, rules, or evidence relevant to the task.Tool useTool use is a system’s ability to invoke an external function, service, interface, or device and incorporate the result.

Inspect retrieval and tools

Sources and scope

What supports this object

Machine Learning GlossaryGoogle for Developers

Defines retrieval-augmented generation and related retrieval concepts; production pipelines vary in indexing, ranking, access control and citation.

Open source

Loading page content...

…