@langchain/core
Version:
Core LangChain.js abstractions and schemas
429 lines (428 loc) • 19.7 kB
JavaScript
;
Object.defineProperty(exports, "__esModule", { value: true });
exports.SaveableVectorStore = exports.VectorStore = exports.VectorStoreRetriever = void 0;
const index_js_1 = require("./retrievers/index.cjs");
const serializable_js_1 = require("./load/serializable.cjs");
/**
* Class for retrieving documents from a `VectorStore` based on vector similarity
* or maximal marginal relevance (MMR).
*
* `VectorStoreRetriever` extends `BaseRetriever`, implementing methods for
* adding documents to the underlying vector store and performing document
* retrieval with optional configurations.
*
* @class VectorStoreRetriever
* @extends BaseRetriever
* @implements VectorStoreRetrieverInterface
* @template V - Type of vector store implementing `VectorStoreInterface`.
*/
class VectorStoreRetriever extends index_js_1.BaseRetriever {
static lc_name() {
return "VectorStoreRetriever";
}
get lc_namespace() {
return ["langchain_core", "vectorstores"];
}
/**
* Returns the type of vector store, as defined by the `vectorStore` instance.
*
* @returns {string} The vector store type.
*/
_vectorstoreType() {
return this.vectorStore._vectorstoreType();
}
/**
* Initializes a new instance of `VectorStoreRetriever` with the specified configuration.
*
* This constructor configures the retriever to interact with a given `VectorStore`
* and supports different retrieval strategies, including similarity search and maximal
* marginal relevance (MMR) search. Various options allow customization of the number
* of documents retrieved per query, filtering based on conditions, and fine-tuning
* MMR-specific parameters.
*
* @param fields - Configuration options for setting up the retriever:
*
* - `vectorStore` (required): The `VectorStore` instance implementing `VectorStoreInterface`
* that will be used to store and retrieve document embeddings. This is the core component
* of the retriever, enabling vector-based similarity and MMR searches.
*
* - `k` (optional): Specifies the number of documents to retrieve per search query. If not
* provided, defaults to 4. This count determines the number of most relevant documents returned
* for each search operation, balancing performance with comprehensiveness.
*
* - `searchType` (optional): Defines the search approach used by the retriever, allowing for
* flexibility between two methods:
* - `"similarity"` (default): A similarity-based search, retrieving documents with high vector
* similarity to the query. This type prioritizes relevance and is often used when diversity
* among results is less critical.
* - `"mmr"`: Maximal Marginal Relevance search, which combines relevance with diversity. MMR
* is useful for scenarios where varied content is essential, as it selects results that
* both match the query and introduce content diversity.
*
* - `filter` (optional): A filter of type `FilterType`, defined by the vector store, that allows
* for refined and targeted search results. This filter applies specified conditions to limit
* which documents are eligible for retrieval, offering control over the scope of results.
*
* - `searchKwargs` (optional, applicable only if `searchType` is `"mmr"`): Additional settings
* for configuring MMR-specific behavior. These parameters allow further tuning of the MMR
* search process:
* - `fetchK`: The initial number of documents fetched from the vector store before the MMR
* algorithm is applied. Fetching a larger set enables the algorithm to select a more
* diverse subset of documents.
* - `lambda`: A parameter controlling the relevance-diversity balance, where 0 emphasizes
* diversity and 1 prioritizes relevance. Intermediate values provide a blend of the two,
* allowing customization based on the importance of content variety relative to query relevance.
*/
constructor(fields) {
super(fields);
/**
* The instance of `VectorStore` used for storing and retrieving document embeddings.
* This vector store must implement the `VectorStoreInterface` to be compatible
* with the retriever’s operations.
*/
Object.defineProperty(this, "vectorStore", {
enumerable: true,
configurable: true,
writable: true,
value: void 0
});
/**
* Specifies the number of documents to retrieve for each search query.
* Defaults to 4 if not specified, providing a basic result count for similarity or MMR searches.
*/
Object.defineProperty(this, "k", {
enumerable: true,
configurable: true,
writable: true,
value: 4
});
/**
* Determines the type of search operation to perform on the vector store.
*
* - `"similarity"` (default): Conducts a similarity search based purely on vector similarity
* to the query.
* - `"mmr"`: Executes a maximal marginal relevance (MMR) search, balancing relevance and
* diversity in the retrieved results.
*/
Object.defineProperty(this, "searchType", {
enumerable: true,
configurable: true,
writable: true,
value: "similarity"
});
/**
* Additional options specific to maximal marginal relevance (MMR) search, applicable
* only if `searchType` is set to `"mmr"`.
*
* Includes:
* - `fetchK`: The initial number of documents fetched before applying the MMR algorithm,
* allowing for a larger selection from which to choose the most diverse results.
* - `lambda`: A parameter between 0 and 1 to adjust the relevance-diversity balance,
* where 0 prioritizes diversity and 1 prioritizes relevance.
*/
Object.defineProperty(this, "searchKwargs", {
enumerable: true,
configurable: true,
writable: true,
value: void 0
});
/**
* Optional filter applied to search results, defined by the `FilterType` of the vector store.
* Allows for refined, targeted results by restricting the returned documents based
* on specified filter criteria.
*/
Object.defineProperty(this, "filter", {
enumerable: true,
configurable: true,
writable: true,
value: void 0
});
this.vectorStore = fields.vectorStore;
this.k = fields.k ?? this.k;
this.searchType = fields.searchType ?? this.searchType;
this.filter = fields.filter;
if (fields.searchType === "mmr") {
this.searchKwargs = fields.searchKwargs;
}
}
/**
* Retrieves relevant documents based on the specified query, using either
* similarity or maximal marginal relevance (MMR) search.
*
* If `searchType` is set to `"mmr"`, performs an MMR search to balance
* similarity and diversity among results. If `searchType` is `"similarity"`,
* retrieves results purely based on similarity to the query.
*
* @param query - The query string used to find relevant documents.
* @param runManager - Optional callback manager for tracking retrieval progress.
* @returns A promise that resolves to an array of `DocumentInterface` instances
* representing the most relevant documents to the query.
* @throws {Error} Throws an error if MMR search is requested but not supported
* by the vector store.
* @protected
*/
async _getRelevantDocuments(query, runManager) {
if (this.searchType === "mmr") {
if (typeof this.vectorStore.maxMarginalRelevanceSearch !== "function") {
throw new Error(`The vector store backing this retriever, ${this._vectorstoreType()} does not support max marginal relevance search.`);
}
return this.vectorStore.maxMarginalRelevanceSearch(query, {
k: this.k,
filter: this.filter,
...this.searchKwargs,
}, runManager?.getChild("vectorstore"));
}
return this.vectorStore.similaritySearch(query, this.k, this.filter, runManager?.getChild("vectorstore"));
}
/**
* Adds an array of documents to the vector store, embedding them as part of
* the storage process.
*
* This method delegates document embedding and storage to the `addDocuments`
* method of the underlying vector store.
*
* @param documents - An array of documents to embed and add to the vector store.
* @param options - Optional settings to customize document addition.
* @returns A promise that resolves to an array of document IDs or `void`,
* depending on the vector store's implementation.
*/
async addDocuments(documents, options) {
return this.vectorStore.addDocuments(documents, options);
}
}
exports.VectorStoreRetriever = VectorStoreRetriever;
/**
* Abstract class representing a vector storage system for performing
* similarity searches on embedded documents.
*
* `VectorStore` provides methods for adding precomputed vectors or documents,
* removing documents based on criteria, and performing similarity searches
* with optional scoring. Subclasses are responsible for implementing specific
* storage mechanisms and the exact behavior of certain abstract methods.
*
* @abstract
* @extends Serializable
* @implements VectorStoreInterface
*/
class VectorStore extends serializable_js_1.Serializable {
/**
* Initializes a new vector store with embeddings and database configuration.
*
* @param embeddings - Instance of `EmbeddingsInterface` used to embed queries.
* @param dbConfig - Configuration settings for the database or storage system.
*/
// eslint-disable-next-line @typescript-eslint/no-explicit-any
constructor(embeddings, dbConfig) {
super(dbConfig);
/**
* Namespace within LangChain to uniquely identify this vector store's
* location, based on the vector store type.
*
* @internal
*/
// Only ever instantiated in main LangChain
Object.defineProperty(this, "lc_namespace", {
enumerable: true,
configurable: true,
writable: true,
value: ["langchain", "vectorstores", this._vectorstoreType()]
});
/**
* Embeddings interface for generating vector embeddings from text queries,
* enabling vector-based similarity searches.
*/
Object.defineProperty(this, "embeddings", {
enumerable: true,
configurable: true,
writable: true,
value: void 0
});
this.embeddings = embeddings;
}
/**
* Deletes documents from the vector store based on the specified parameters.
*
* @param _params - Flexible key-value pairs defining conditions for document deletion.
* @returns A promise that resolves once the deletion is complete.
*/
// eslint-disable-next-line @typescript-eslint/no-explicit-any
async delete(_params) {
throw new Error("Not implemented.");
}
/**
* Searches for documents similar to a text query by embedding the query and
* performing a similarity search on the resulting vector.
*
* @param query - Text query for finding similar documents.
* @param k - Number of similar results to return. Defaults to 4.
* @param filter - Optional filter based on `FilterType`.
* @param _callbacks - Optional callbacks for monitoring search progress
* @returns A promise resolving to an array of `DocumentInterface` instances representing similar documents.
*/
async similaritySearch(query, k = 4, filter = undefined, _callbacks = undefined // implement passing to embedQuery later
) {
const results = await this.similaritySearchVectorWithScore(await this.embeddings.embedQuery(query), k, filter);
return results.map((result) => result[0]);
}
/**
* Searches for documents similar to a text query by embedding the query,
* and returns results with similarity scores.
*
* @param query - Text query for finding similar documents.
* @param k - Number of similar results to return. Defaults to 4.
* @param filter - Optional filter based on `FilterType`.
* @param _callbacks - Optional callbacks for monitoring search progress
* @returns A promise resolving to an array of tuples, each containing a
* document and its similarity score.
*/
async similaritySearchWithScore(query, k = 4, filter = undefined, _callbacks = undefined // implement passing to embedQuery later
) {
return this.similaritySearchVectorWithScore(await this.embeddings.embedQuery(query), k, filter);
}
/**
* Creates a `VectorStore` instance from an array of text strings and optional
* metadata, using the specified embeddings and database configuration.
*
* Subclasses must implement this method to define how text and metadata
* are embedded and stored in the vector store. Throws an error if not overridden.
*
* @param _texts - Array of strings representing the text documents to be stored.
* @param _metadatas - Metadata for the texts, either as an array (one for each text)
* or a single object (applied to all texts).
* @param _embeddings - Instance of `EmbeddingsInterface` to embed the texts.
* @param _dbConfig - Database configuration settings.
* @returns A promise that resolves to a new `VectorStore` instance.
* @throws {Error} Throws an error if this method is not overridden by a subclass.
*/
static fromTexts(_texts, _metadatas, _embeddings,
// eslint-disable-next-line @typescript-eslint/no-explicit-any
_dbConfig) {
throw new Error("the Langchain vectorstore implementation you are using forgot to override this, please report a bug");
}
/**
* Creates a `VectorStore` instance from an array of documents, using the specified
* embeddings and database configuration.
*
* Subclasses must implement this method to define how documents are embedded
* and stored. Throws an error if not overridden.
*
* @param _docs - Array of `DocumentInterface` instances representing the documents to be stored.
* @param _embeddings - Instance of `EmbeddingsInterface` to embed the documents.
* @param _dbConfig - Database configuration settings.
* @returns A promise that resolves to a new `VectorStore` instance.
* @throws {Error} Throws an error if this method is not overridden by a subclass.
*/
static fromDocuments(_docs, _embeddings,
// eslint-disable-next-line @typescript-eslint/no-explicit-any
_dbConfig) {
throw new Error("the Langchain vectorstore implementation you are using forgot to override this, please report a bug");
}
/**
* Creates a `VectorStoreRetriever` instance with flexible configuration options.
*
* @param kOrFields
* - If a number is provided, it sets the `k` parameter (number of items to retrieve).
* - If an object is provided, it should contain various configuration options.
* @param filter
* - Optional filter criteria to limit the items retrieved based on the specified filter type.
* @param callbacks
* - Optional callbacks that may be triggered at specific stages of the retrieval process.
* @param tags
* - Tags to categorize or label the `VectorStoreRetriever`. Defaults to an empty array if not provided.
* @param metadata
* - Additional metadata as key-value pairs to add contextual information for the retrieval process.
* @param verbose
* - If `true`, enables detailed logging for the retrieval process. Defaults to `false`.
*
* @returns
* - A configured `VectorStoreRetriever` instance based on the provided parameters.
*
* @example
* Basic usage with a `k` value:
* ```typescript
* const retriever = myVectorStore.asRetriever(5);
* ```
*
* Usage with a configuration object:
* ```typescript
* const retriever = myVectorStore.asRetriever({
* k: 10,
* filter: myFilter,
* tags: ['example', 'test'],
* verbose: true,
* searchType: 'mmr',
* searchKwargs: { alpha: 0.5 },
* });
* ```
*/
asRetriever(kOrFields, filter, callbacks, tags, metadata, verbose) {
if (typeof kOrFields === "number") {
return new VectorStoreRetriever({
vectorStore: this,
k: kOrFields,
filter,
tags: [...(tags ?? []), this._vectorstoreType()],
metadata,
verbose,
callbacks,
});
}
else {
const params = {
vectorStore: this,
k: kOrFields?.k,
filter: kOrFields?.filter,
tags: [...(kOrFields?.tags ?? []), this._vectorstoreType()],
metadata: kOrFields?.metadata,
verbose: kOrFields?.verbose,
callbacks: kOrFields?.callbacks,
searchType: kOrFields?.searchType,
};
if (kOrFields?.searchType === "mmr") {
return new VectorStoreRetriever({
...params,
searchKwargs: kOrFields.searchKwargs,
});
}
return new VectorStoreRetriever({ ...params });
}
}
}
exports.VectorStore = VectorStore;
/**
* Abstract class extending `VectorStore` that defines a contract for saving
* and loading vector store instances.
*
* The `SaveableVectorStore` class allows vector store implementations to
* persist their data and retrieve it when needed.The format for saving and
* loading data is left to the implementing subclass.
*
* Subclasses must implement the `save` method to handle their custom
* serialization logic, while the `load` method enables reconstruction of a
* vector store from saved data, requiring compatible embeddings through the
* `EmbeddingsInterface`.
*
* @abstract
* @extends VectorStore
*/
class SaveableVectorStore extends VectorStore {
/**
* Loads a vector store instance from the specified directory, using the
* provided embeddings to ensure compatibility.
*
* This static method reconstructs a `SaveableVectorStore` from previously
* saved data. Implementations should interpret the saved data format to
* recreate the vector store instance.
*
* @param _directory - The directory path from which the vector store
* data will be loaded.
* @param _embeddings - An instance of `EmbeddingsInterface` to align
* the embeddings with the loaded vector data.
* @returns A promise that resolves to a `SaveableVectorStore` instance
* constructed from the saved data.
*/
static load(_directory, _embeddings) {
throw new Error("Not implemented");
}
}
exports.SaveableVectorStore = SaveableVectorStore;