Module: batteries/docs

This module provides the DocsQA component for document question and answering using an LLM. See: https://docs.ai-jsx.com/guides/brand-new#accessing-knowledge-docs-qa

Enumerations

• CorpusLoadingState

Classes

• FixieCorpus
• LangChainEmbeddingWrapper
• LangchainCorpus
• LoadableLangchainCorpus
• LocalCorpus

Interfaces

• Chunk
• Corpus
• CorpusLoadRequest
• CorpusLoadResponse
• CorpusPage
• CorpusPartition
• CorpusStats
• DocsQAProps
• DocsQAWithCitationsProps
• Document
• EmbeddedChunk
• Embedding
• LoadableCorpus
• RawDocument
• ScoredChunk

Type Aliases

Chunker

Ƭ Chunker<DocumentMetadata, ChunkMetadata>: (document: Document<DocumentMetadata>) => Promise<Chunk<ChunkMetadata>[]>

Type parameters

Name	Type
DocumentMetadata	extends Jsonifiable = Jsonifiable
ChunkMetadata	extends Jsonifiable = Jsonifiable

Type declaration

▸ (document): Promise<Chunk<ChunkMetadata>[]>

A function that splits a document into multiple chunks, for example to ensure it fits in appropriate context windows.

Parameters

Name	Type
document	Document<DocumentMetadata>

Returns

Promise<Chunk<ChunkMetadata>[]>

Defined in

packages/ai-jsx/src/batteries/docs.tsx:259

---

Loader

Ƭ Loader<DocumentMetadata>: (request: CorpusLoadRequest) => Promise<CorpusLoadResponse<Document<DocumentMetadata>>>

Type parameters

Name	Type
DocumentMetadata	extends Jsonifiable = Jsonifiable

Type declaration

▸ (request): Promise<CorpusLoadResponse<Document<DocumentMetadata>>>

A function responsible for loading and parsing a corpus.

Parameters

Name	Type
request	CorpusLoadRequest

Returns

Promise<CorpusLoadResponse<Document<DocumentMetadata>>>

Defined in

packages/ai-jsx/src/batteries/docs.tsx:183

---

Parser

Ƭ Parser<DocumentMetadata>: (raw: RawDocument) => Promise<Document<DocumentMetadata>>

Type parameters

Name	Type
DocumentMetadata	extends Jsonifiable = Jsonifiable

Type declaration

▸ (raw): Promise<Document<DocumentMetadata>>

A function that parses raw documents to produce text documents that can be used by an LLM.

Parameters

Name	Type
raw	RawDocument

Returns

Promise<Document<DocumentMetadata>>

Defined in

packages/ai-jsx/src/batteries/docs.tsx:65

---

QAWithCitationsResult

Ƭ QAWithCitationsResult: Partial<z.infer<typeof ResultSchema>>

Defined in

packages/ai-jsx/src/batteries/docs.tsx:820

---

RawLoader

Ƭ RawLoader: (request: CorpusLoadRequest) => Promise<CorpusLoadResponse<RawDocument>>

Type declaration

▸ (request): Promise<CorpusLoadResponse<RawDocument>>

A function responsible for loading a corpus when parsing is handled separately.

Parameters

Name	Type
request	CorpusLoadRequest

Returns

Promise<CorpusLoadResponse<RawDocument>>

Defined in

packages/ai-jsx/src/batteries/docs.tsx:188

---

Vectorizer

Ƭ Vectorizer<DocumentMetadata, ChunkMetadata>: (documents: Document<DocumentMetadata>[]) => Promise<EmbeddedChunk<ChunkMetadata>[]>

Type parameters

Name	Type
DocumentMetadata	extends Jsonifiable = Jsonifiable
ChunkMetadata	extends Jsonifiable = Jsonifiable

Type declaration

▸ (documents): Promise<EmbeddedChunk<ChunkMetadata>[]>

A function that converts documents into EmbeddedChunks.

See

asVectorizer

Parameters

Name	Type
documents	Document<DocumentMetadata>[]

Returns

Promise<EmbeddedChunk<ChunkMetadata>[]>

Defined in

packages/ai-jsx/src/batteries/docs.tsx:344

Functions

DefaultFormatter

▸ DefaultFormatter(«destructured»): Element

A default component for formatting document chunks.

Parameters

Name	Type
«destructured»	Object
› doc	ScoredChunk<Jsonifiable>

Returns

Element

Defined in

packages/ai-jsx/src/batteries/docs.tsx:727

---

DocsQA

▸ DocsQA<ChunkMetadata>(props): Promise<Element>

A component that can be used to answer questions about documents. This is a very common usecase for LLMs.

Example

  <DocsQA question="What was Hurricane Katrina?" corpus={corpus} chunkLimit={5} chunkFormatter={GetChunk} />

Type parameters

Name	Type
ChunkMetadata	extends Jsonifiable = Jsonifiable

Parameters

Name	Type
props	DocsQAProps<ChunkMetadata>

Returns

Promise<Element>

Defined in

packages/ai-jsx/src/batteries/docs.tsx:795

---

DocsQAWithCitations

▸ DocsQAWithCitations<ChunkMetadata>(props, «destructured»): AsyncGenerator<Node, Node, unknown>

Similar to DocsQA, but encourages the LLM to return citations for its answer.

Type parameters

Name	Type
ChunkMetadata	extends Jsonifiable = Jsonifiable

Parameters

Name	Type
props	DocsQAWithCitationsProps<ChunkMetadata>
«destructured»	ComponentContext

Returns

AsyncGenerator<Node, Node, unknown>

Defined in

packages/ai-jsx/src/batteries/docs.tsx:833

---

asVectorizer

▸ asVectorizer<DocumentMetadata, ChunkMetadata>(chunker, embedding): (docs: Document<DocumentMetadata>[]) => Promise<EmbeddedChunk<ChunkMetadata>[]>

Converts a Chunker + Embedding into a unified Vectorizer function.

Type parameters

Name	Type
DocumentMetadata	extends Jsonifiable = Jsonifiable
ChunkMetadata	extends Jsonifiable = Jsonifiable

Parameters

Name	Type
chunker	Chunker<DocumentMetadata, ChunkMetadata>
embedding	Embedding

Returns

fn

▸ (docs): Promise<EmbeddedChunk<ChunkMetadata>[]>

Parameters

Name	Type
docs	Document<DocumentMetadata>[]

Returns

Promise<EmbeddedChunk<ChunkMetadata>[]>

Defined in

packages/ai-jsx/src/batteries/docs.tsx:350

---

defaultChunker

▸ defaultChunker(document): Promise<Chunk<Jsonifiable>[]>

A simple token size based text chunker. This is a good starting point for text documents.

Parameters

Name	Type
document	Document<Jsonifiable>

Returns

Promise<Chunk<Jsonifiable>[]>

Defined in

packages/ai-jsx/src/batteries/docs.tsx:262

---

makeChunker

▸ makeChunker<Metadata>(chunkSize, chunkOverlap): Chunker<Metadata, Metadata>

Creates a Chunker with the given parameters.

Type parameters

Name	Type
Metadata	extends Jsonifiable = Jsonifiable

Parameters

Name	Type
chunkSize	number
chunkOverlap	number

Returns

Chunker<Metadata, Metadata>

Defined in

packages/ai-jsx/src/batteries/docs.tsx:265

---

staticLoader

▸ staticLoader<DocumentMetadata>(documents, pageSize?): Loader<DocumentMetadata>

A Loader that provides a static set of in-memory documents, optionally with pagination.

Type parameters

Name	Type
DocumentMetadata	extends Jsonifiable = Jsonifiable

Parameters

Name	Type
documents	Document<DocumentMetadata>[]
pageSize?	number

Returns

Loader<DocumentMetadata>

Defined in

packages/ai-jsx/src/batteries/docs.tsx:247

---

toLoader

▸ toLoader<DocumentMetadata>(rawLoader, parser?): Loader<DocumentMetadata>

Combines a RawLoader with a Parser to produce a Loader.

Type parameters

Name	Type
DocumentMetadata	extends Jsonifiable = Jsonifiable

Parameters

Name	Type	Default value
rawLoader	RawLoader	undefined
parser	Parser<DocumentMetadata>	defaultParser

Returns

Loader<DocumentMetadata>

Defined in

packages/ai-jsx/src/batteries/docs.tsx:191