Basic RAG and Hybrid Search templates

Templates create a starting graph that can be inspected and configured visually. The templates are intended to teach the structure of a RAG pipeline and reduce empty-canvas setup.

Basic RAG #

The Basic RAG template contains seven nodes in this order:

Document Loader → Chunker → Embedder → Vector Store → Retriever → LLM → Answer / Output

Use this pattern when you want to understand the standard conceptual route from a collection of documents to retrieved context and an answer.

Hybrid Search RAG #

The Hybrid Search RAG template contains eight nodes:

Document Loader → Chunker → Embedder → Vector Store → Hybrid Retriever → Reranker → LLM → Answer / Output

The Hybrid Retriever represents dense semantic retrieval combined with BM25 keyword retrieval through reciprocal rank fusion (RRF). The Reranker represents a separate scoring stage after the initial retrieval results.

How to choose #

Start with Basic RAG if you are learning the building blocks. Explore Hybrid Search when you want to reason about a combined semantic and keyword retrieval strategy and the effect of reranking.

A template is not a deployment #

Selecting a template creates a design-time graph. It does not install a vector database, ingest files, query a model, or deploy an endpoint. Those require backend capabilities that are not currently connected.