RAG

Want a working RAG agent rather than wiring the pipeline yourself? The Registry ships document-Q&A starters you can install and point at your data.

#Core pipeline

• createRAG({ embed, store }) — one-liner ingest + retrieve + search. Defaults: chunkSize 512, chunkOverlap 50.
• chunkText(text, { chunkSize, chunkOverlap, split? }) — standalone splitter. split is an optional (text: string) => string[].

#Reranking + hybrid

• createRerankedRetriever(base, { candidatePool?, topK?, rerank? }) — plug in voyageReranker, jinaReranker, or the built-in bm25Rerank.
• createHybridRetriever — vector + BM25 blend with weighted normalization.
• Recipe: RAG reranking.

#Document loaders

• loadUrl (raw response text), loadGitHubFile, loadGitHubTree, loadNotionPage, loadConfluencePage, loadGoogleDriveFile, loadPdf (BYO parser). All return InputDocument[].
• Recipe: Document loaders.

Loader and reranker guides are available through the document loaders recipe and RAG reranking recipe.

#Related

• Concepts: Retriever
• Package: @agentskit/rag
• For agents: rag

Explore nearby

• Peer
  createRAG

  One-liner RAG pipeline — chunk, embed, store, retrieve, search.

• Peer
  Chunking

  Split docs before embedding. Sensible defaults; override per doc type.

• Peer
  Rerank

  Post-retrieval reranker. Voyage / Jina / custom functions / built-in BM25.