Skip to content
AgentsKit

@agentskit/rag β€” Functions

API functions for @agentskit/rag.

#Function: bm25Score()

bm25Score(query, documents, options?): RetrievedDocument[]

Defined in: packages/rag/src/rerank.ts:187

Score a set of documents against a query using classic BM25. Returns new document objects with a finite .score field, sorted descending. Input documents are never mutated.

#Parameters

#query

string

Search query.

#documents

RetrievedDocument[]

Candidate documents.

#options?

BM25Options = \{\}

Optional BM25 tuning parameters.

#Returns

RetrievedDocument[]

New documents with finite scores in descending order.


#Function: chunkText()

chunkText(text, options): string[]

Defined in: packages/rag/src/chunker.ts:24

Splits text into non-empty chunks using an optional custom splitter.

#Parameters

#text

string

#options

ChunkOptions

#Returns

string[]


#Function: createHybridRetriever()

createHybridRetriever(base, options?): Retriever

Defined in: packages/rag/src/rerank.ts:290

Combine a vector-backed base retriever with a BM25 keyword pass over the same candidate pool. Final score is a weighted sum of the two min-max-normalized scores using a finite relative weight pair that sums to 1 (both zero β†’ 0.5/0.5).

#Parameters

#base

Retriever

Retriever that supplies vector search results.

#options?

HybridRetrieverOptions = \{\}

Candidate pool and relative score weights.

#Returns

Retriever

A retriever with normalized hybrid scores.

#Throws

When candidate scores are invalid or retrieval fails.


#Function: createPartitionedRAG()

createPartitionedRAG(config): RAG

Defined in: packages/rag/src/partitioned.ts:25

Compose createRAG with metadata isolation; stores must faithfully preserve metadata.

#Parameters

#config

PartitionedRAGConfig

#Returns

RAG


#Function: createRAG()

createRAG(config): RAG

Defined in: packages/rag/src/rag.ts:62

Creates a retriever that chunks, embeds, stores, and searches documents.

#Parameters

#config

RAGConfig

Store, embedding function, and optional chunking and search settings.

#Returns

RAG

A RAG retriever with an ingest method.

#Example

const rag = createRAG({ store, embed })
await rag.ingest([{ id: 'guide', content: 'Install the package.' }])
const results = await rag.retrieve({ query: 'How do I install it?' })

#Function: createRerankedRetriever()

createRerankedRetriever(base, options?): Retriever

Defined in: packages/rag/src/rerank.ts:123

Wrap any base Retriever with a reranker. Typical flow:

  1. Vector search returns ~20 candidates (candidatePool)
  2. rerank re-scores them with a stronger signal (Cohere Rerank, BGE cross-encoder, or BM25 for keyword-aware hybrid search)
  3. Top topK are returned

#Parameters

#base

Retriever

Retriever that supplies the candidate documents.

#options?

RerankedRetrieverOptions = \{\}

Reranker and candidate limits.

#Returns

Retriever

A retriever that returns reranked documents.

#Throws

When reranking returns invalid results or fails.


#Function: jinaReranker()

jinaReranker(options): RerankFn

Defined in: packages/rag/src/rerankers/jina.ts:38

Jina AI cross-encoder reranker. Drop-in RerankFn for createRerankedRetriever.

#Parameters

#options

JinaRerankerOptions

API key, model, and optional request settings.

#Returns

RerankFn

A reranking function for retrieved documents.

#Throws

When the request fails or the response is invalid.


#Function: loadConfluencePage()

loadConfluencePage(pageId, options): Promise<InputDocument[]>

Defined in: packages/rag/src/loaders/documents.ts:245

Loads a Confluence page as a document.

#Parameters

#pageId

string

Confluence page identifier.

#options

ConfluenceLoaderOptions

Page identifier, credentials, and loader limits.

#Returns

Promise<InputDocument[]>

The page content as a document.

#Throws

When the request or response parsing fails.


#Function: loadDropbox()

loadDropbox(options): Promise<InputDocument[]>

Defined in: packages/rag/src/loaders/cloud.ts:110

Lists and loads eligible files from Dropbox.

#Parameters

#options

DropboxLoaderOptions

Access token, path, filtering, and loader limits.

#Returns

Promise<InputDocument[]>

Successfully loaded files; fails if every eligible download fails.

#Throws

When listing fails or all eligible downloads fail.


#Function: loadGcs()

loadGcs(options): Promise<InputDocument[]>

Defined in: packages/rag/src/loaders/cloud.ts:33

Lists and loads eligible objects from a Google Cloud Storage bucket.

#Parameters

#options

GcsLoaderOptions

Bucket, credentials, filtering, and loader limits.

#Returns

Promise<InputDocument[]>

Successfully loaded objects; fails if every eligible download fails.

#Throws

When listing fails or all eligible downloads fail.


#Function: loadGitHubFile()

loadGitHubFile(owner, repo, path, options?): Promise<InputDocument[]>

Defined in: packages/rag/src/loaders/documents.ts:72

Loads one GitHub file as a document.

#Parameters

#owner

string

Repository owner.

#repo

string

Repository name.

#path

string

Repository-relative file path.

#options?

GitHubLoaderOptions = \{\}

GitHub credentials and loader limits.

#Returns

Promise<InputDocument[]>

The loaded file as a document.

#Throws

When the request or response parsing fails.


#Function: loadGitHubTree()

loadGitHubTree(owner, repo, options?): Promise<InputDocument[]>

Defined in: packages/rag/src/loaders/documents.ts:110

Loads matching files from a GitHub repository tree.

#Parameters

#owner

string

Repository owner.

#repo

string

Repository name.

#options?

GitHubTreeOptions = \{\}

Repository, filtering, credentials, and loader limits.

#Returns

Promise<InputDocument[]>

Successfully loaded documents; fails if every eligible download fails.

#Throws

When listing fails or all eligible downloads fail.


#Function: loadGoogleDriveFile()

loadGoogleDriveFile(fileId, options): Promise<InputDocument[]>

Defined in: packages/rag/src/loaders/documents.ts:279

Loads an exported Google Drive file as a document.

#Parameters

#fileId

string

Google Drive file identifier.

#options

DriveLoaderOptions

File identifier, credentials, and loader limits.

#Returns

Promise<InputDocument[]>

The file content as a document.

#Throws

When the request or response parsing fails.


#Function: loadNotionPage()

loadNotionPage(pageId, options): Promise<InputDocument[]>

Defined in: packages/rag/src/loaders/documents.ts:180

Loads supported text blocks from a Notion page, following child pagination.

#Parameters

#pageId

string

Notion page identifier.

#options

NotionLoaderOptions

Page identifier, credentials, and loader limits.

#Returns

Promise<InputDocument[]>

The page content as a document.

#Throws

When loading or pagination fails.


#Function: loadOneDrive()

loadOneDrive(options): Promise<InputDocument[]>

Defined in: packages/rag/src/loaders/cloud.ts:197

Lists and loads eligible files from OneDrive, including nested folders.

#Parameters

#options

OneDriveLoaderOptions

Access token, folder, filtering, and loader limits.

#Returns

Promise<InputDocument[]>

Successfully loaded files; fails if every eligible download fails.

#Throws

When listing fails or all eligible downloads fail.


#Function: loadPdf()

loadPdf(url, options): Promise<InputDocument[]>

Defined in: packages/rag/src/loaders/documents.ts:305

Fetches a PDF and extracts its text into a document.

#Parameters

#url

string

URL of the PDF.

#options

PdfLoaderOptions

Fetch and loader limits.

#Returns

Promise<InputDocument[]>

The extracted PDF document.

#Throws

When fetching or parsing fails.


#Function: loadS3()

loadS3(options): Promise<InputDocument[]>

Defined in: packages/rag/src/loaders/s3.ts:61

Lists and loads eligible objects from S3.

#Parameters

#options

S3LoaderOptions

Bucket, client or commands, filtering, and loader limits.

#Returns

Promise<InputDocument[]>

Successfully loaded objects; fails if every eligible download fails.

#Throws

When listing fails or all eligible downloads fail.


#Function: loadUrl()

loadUrl(url, options?): Promise<InputDocument[]>

Defined in: packages/rag/src/loaders/documents.ts:44

Fetches a URL and returns its supported content as one or more documents.

#Parameters

#url

string

URL to fetch.

#options?

UrlLoaderOptions = \{\}

Origin allowlist, fetch, timeout, and content-size limits.

#Returns

Promise<InputDocument[]>

Documents extracted from the response.

#Throws

When fetching, reading, or parsing the response fails.

#Example

const [document] = await loadUrl('https://example.com/guide', {
  allowedOrigins: ['https://example.com'],
})

#Function: voyageReranker()

voyageReranker(options): RerankFn

Defined in: packages/rag/src/rerankers/voyage.ts:39

Voyage AI cross-encoder reranker. Drop-in RerankFn for createRerankedRetriever.

#Parameters

#options

VoyageRerankerOptions

API key, model, and optional request settings.

#Returns

RerankFn

A reranking function for retrieved documents.

#Throws

When the request fails or the response is invalid.

Explore nearby