Class RAG

Orchestrates Retrieval Augmented Generation. Coordinates a VectorStore and an LLM to ingest, retrieve, and generate.

Example

const rag = await new RAG({ vectorStore, llm }).load();
const answer = await rag.generate({ input: 'What is RAG?' });

Index

Constructors

constructor

Methods

addDocument deleteDocument generate interrupt load splitAddDocument unload updateDocument

Constructors

constructor

new RAG(params: { llm: LLM; vectorStore: VectorStore }): RAG
Creates a new RAG instance.
Parameters
- params: { llm: LLM; vectorStore: VectorStore }
  Object containing the implementations.
  - llm: LLM
    Large Language Model used for generation.
  - vectorStore: VectorStore
    Vector store used for retrieval.
Returns RAG

Methods

addDocument

addDocument(
    params: {
        document?: string;
        embedding?: number[];
        id?: string;
        metadata?: Record<string, any>;
    },
): Promise<string>
Adds a document to the vector store.
Parameters
- params: {
      document?: string;
      embedding?: number[];
      id?: string;
      metadata?: Record<string, any>;
  }
  Parameters for the operation.
  - Optionaldocument?: string
    Raw text content for the document.
  - Optionalembedding?: number[]
    Embedding for the document.
  - Optionalid?: string
    ID for the document.
  - Optionalmetadata?: Record<string, any>
    Metadata for the document.
Returns Promise<string>
Promise that resolves to the ID of the newly added document.

deleteDocument

deleteDocument(
params: { predicate: (value: GetResult) => boolean },
): Promise<void>
Deletes documents from the vector store by the provided predicate.
Parameters
- params: { predicate: (value: GetResult) => boolean }
  Parameters for deletion.
  - predicate: (value: GetResult) => boolean
    Predicate to match documents for deletion.
Returns Promise<void>
Promise that resolves once the documents are deleted.

generate

generate(
    params: {
        augmentedGeneration?: boolean;
        callback?: (token: string) => void;
        input: string | Message[];
        nResults?: number;
        predicate?: (value: QueryResult) => boolean;
        promptGenerator?: (
            messages: Message[],
            retrievedDocs: QueryResult[],
        ) => string;
        questionGenerator?: (messages: Message[]) => string;
    },
): Promise<string>
Generates a response based on the input messages and retrieved documents. If augmentedGeneration is true, it retrieves relevant documents from the vector store and includes them in the prompt for the LLM.
Parameters
- params: {
      augmentedGeneration?: boolean;
      callback?: (token: string) => void;
      input: string | Message[];
      nResults?: number;
      predicate?: (value: QueryResult) => boolean;
      promptGenerator?: (
          messages: Message[],
          retrievedDocs: QueryResult[],
      ) => string;
      questionGenerator?: (messages: Message[]) => string;
  }
  Generation parameters.
  - OptionalaugmentedGeneration?: boolean
    Whether to augment with retrieved context (default: true).
  - Optionalcallback?: (token: string) => void
    Token callback for streaming.
  - input: string | Message[]
    Input messages or a single string.
  - OptionalnResults?: number
    Number of docs to retrieve (default: 3).
  - Optionalpredicate?: (value: QueryResult) => boolean
    Filter applied to retrieved docs.
  - OptionalpromptGenerator?: (messages: Message[], retrievedDocs: QueryResult[]) => string
    Builds the context-augmented prompt from messages and retrieved docs.
  - OptionalquestionGenerator?: (messages: Message[]) => string
    Maps the message list to a search query (default: last message content).
Returns Promise<string>
Promise that resolves to the generated text.

interrupt

interrupt(): Promise<void>
Interrupts the ongoing text generation process.

Returns Promise<void>
Promise that resolves when the interruption is complete.

load

load(): Promise<RAG>
Initializes the RAG system by loading the vector store and LLM.

Returns Promise<RAG>
A promise that resolves to the same RAG instance.

splitAddDocument

splitAddDocument(
    params: {
        document: string;
        metadataGenerator?: (chunks: string[]) => Record<string, any>[];
        textSplitter?: TextSplitter;
    },
): Promise<string[]>
Splits a document into chunks and adds them to the vector store. If no textSplitter is provided, a default RecursiveCharacterTextSplitter({ chunkSize: 500, chunkOverlap: 100 }) is used.
Parameters
- params: {
      document: string;
      metadataGenerator?: (chunks: string[]) => Record<string, any>[];
      textSplitter?: TextSplitter;
  }
  Parameters for the operation.
  - document: string
    The content of the document to split and add.
  - OptionalmetadataGenerator?: (chunks: string[]) => Record<string, any>[]
    Function to generate metadata for each chunk. Must return an array which length is equal to the number of chunks.
  - OptionaltextSplitter?: TextSplitter
    Text splitter implementation.
Returns Promise<string[]>
Promise that resolves to the IDs of the newly added chunks.

unload

unload(): Promise<void>
Unloads the RAG system, releasing resources used by the vector store and LLM.

Returns Promise<void>
A promise that resolves when unloading is complete.

updateDocument

updateDocument(
    params: {
        document?: string;
        embedding?: number[];
        id: string;
        metadata?: Record<string, any>;
    },
): Promise<void>
Updates a document in the vector store by its ID.
Parameters
- params: {
      document?: string;
      embedding?: number[];
      id: string;
      metadata?: Record<string, any>;
  }
  Parameters for the update.
  - Optionaldocument?: string
    New content for the document.
  - Optionalembedding?: number[]
    New embedding for the document. If not provided, it will be generated based on the document.
  - id: string
    The ID of the document to update.
  - Optionalmetadata?: Record<string, any>
    New metadata for the document.
Returns Promise<void>
Promise that resolves once the document is updated.

Class RAG

Example

Index

Constructors

Methods

Constructors

constructor

Parameters

llm: LLM

vectorStore: VectorStore

Returns RAG

Methods

addDocument

Parameters

Optionaldocument?: string

Optionalembedding?: number[]

Optionalid?: string

Optionalmetadata?: Record<string, any>

Returns Promise<string>

deleteDocument

Parameters

predicate: (value: GetResult) => boolean

Returns Promise<void>

generate

Parameters

OptionalaugmentedGeneration?: boolean

Optionalcallback?: (token: string) => void

input: string | Message[]

OptionalnResults?: number

Optionalpredicate?: (value: QueryResult) => boolean

OptionalpromptGenerator?: (messages: Message[], retrievedDocs: QueryResult[]) => string

OptionalquestionGenerator?: (messages: Message[]) => string

Returns Promise<string>

interrupt

Returns Promise<void>

load

Returns Promise<RAG>

splitAddDocument

Parameters

document: string

OptionalmetadataGenerator?: (chunks: string[]) => Record<string, any>[]

OptionaltextSplitter?: TextSplitter

Returns Promise<string[]>

unload

Returns Promise<void>

updateDocument

Parameters

Optionaldocument?: string

Optionalembedding?: number[]

id: string

Optionalmetadata?: Record<string, any>

Returns Promise<void>

Settings

On This Page

`Optional`document?: string

`Optional`embedding?: number[]

`Optional`id?: string

`Optional`metadata?: Record<string, any>

`Optional`augmentedGeneration?: boolean

`Optional`callback?: (token: string) => void

`Optional`nResults?: number

`Optional`predicate?: (value: QueryResult) => boolean

`Optional`promptGenerator?: (messages: Message[], retrievedDocs: QueryResult[]) => string

`Optional`questionGenerator?: (messages: Message[]) => string

`Optional`metadataGenerator?: (chunks: string[]) => Record<string, any>[]

`Optional`textSplitter?: TextSplitter

`Optional`document?: string

`Optional`embedding?: number[]

`Optional`metadata?: Record<string, any>