Embeddings - LangChain.js

Embeddings are numerical representations of text that capture semantic meaning. LangChain.js provides integrations with various embedding model providers to help you build semantic search, recommendation systems, and RAG (Retrieval-Augmented Generation) applications.

Popular Embedding Providers

OpenAI

text-embedding-3-small and text-embedding-3-large

Cohere

Embed-v3 with multilingual support

Google

Vertex AI and Generative AI embeddings

Nomic

Open-source embedding models

OpenAI Embeddings

OpenAI provides state-of-the-art embedding models with excellent performance and cost-efficiency.

Installation

npm install @langchain/openai

Usage

import { OpenAIEmbeddings } from "@langchain/openai";

const embeddings = new OpenAIEmbeddings({
  model: "text-embedding-3-small",
  apiKey: process.env.OPENAI_API_KEY,
});

// Embed a single document
const docEmbedding = await embeddings.embedQuery(
  "What is the meaning of life?"
);

console.log(docEmbedding.length); // 1536 dimensions for text-embedding-3-small

Batch Embeddings

// Embed multiple documents
const documents = [
  "LangChain is a framework for LLM applications",
  "Vector embeddings capture semantic meaning",
  "Retrieval-Augmented Generation improves LLM accuracy",
];

const docEmbeddings = await embeddings.embedDocuments(documents);

console.log(docEmbeddings.length); // 3
console.log(docEmbeddings[0].length); // 1536

Model Options

// text-embedding-3-small: Efficient and cost-effective
const smallEmbeddings = new OpenAIEmbeddings({
  model: "text-embedding-3-small",
  dimensions: 512, // Optional: reduce dimensions for efficiency
});

// text-embedding-3-large: Higher quality for demanding tasks
const largeEmbeddings = new OpenAIEmbeddings({
  model: "text-embedding-3-large",
  dimensions: 3072, // Default dimension
});

Cohere Embeddings

Cohere’s embedding models support multiple languages and task-specific optimizations.

Installation

npm install @langchain/cohere

Usage

import { CohereEmbeddings } from "@langchain/cohere";

const embeddings = new CohereEmbeddings({
  model: "embed-english-v3.0",
  apiKey: process.env.COHERE_API_KEY,
});

const query = "What is machine learning?";
const queryEmbedding = await embeddings.embedQuery(query);

console.log(queryEmbedding.length); // 1024 dimensions

Input Types

Cohere embeddings support different input types for optimization:

import { CohereEmbeddings } from "@langchain/cohere";

// For search queries
const searchEmbeddings = new CohereEmbeddings({
  model: "embed-english-v3.0",
  inputType: "search_query",
});

// For documents being indexed
const docEmbeddings = new CohereEmbeddings({
  model: "embed-english-v3.0",
  inputType: "search_document",
});

// For classification tasks
const classificationEmbeddings = new CohereEmbeddings({
  model: "embed-english-v3.0",
  inputType: "classification",
});

Google Embeddings

Google offers embedding models through both Vertex AI and Generative AI APIs.

Installation

npm install @langchain/google-genai

Generative AI (API Key)

import { GoogleGenerativeAIEmbeddings } from "@langchain/google-genai";

const embeddings = new GoogleGenerativeAIEmbeddings({
  model: "text-embedding-004",
  apiKey: process.env.GOOGLE_API_KEY,
});

const text = "LangChain simplifies LLM application development";
const embedding = await embeddings.embedQuery(text);

console.log(embedding.length); // 768 dimensions

Vertex AI

npm install @langchain/google-vertexai

import { GoogleVertexAIEmbeddings } from "@langchain/google-vertexai";

const embeddings = new GoogleVertexAIEmbeddings({
  model: "text-embedding-004",
  // Requires Google Cloud credentials
});

const embedding = await embeddings.embedQuery("Sample text");

Nomic Embeddings

Nomic provides open-source embedding models that can run locally or via their API.

Installation

npm install @langchain/nomic

Usage

import { NomicEmbeddings } from "@langchain/nomic";

const embeddings = new NomicEmbeddings({
  model: "nomic-embed-text-v1.5",
  apiKey: process.env.NOMIC_API_KEY,
});

const query = "How do transformers work?";
const embedding = await embeddings.embedQuery(query);

Additional Providers

AWS Bedrock

@langchain/aws - Titan and other embedding models

Azure OpenAI

@langchain/openai - OpenAI embeddings via Azure

Ollama

@langchain/ollama - Local embedding models

Mixedbread AI

@langchain/mixedbread-ai - High-quality embeddings

Community Integrations

Additional embedding providers are available in @langchain/community:

npm install @langchain/community

import { HuggingFaceInferenceEmbeddings } from "@langchain/community/embeddings/hf";
import { DeepInfraEmbeddings } from "@langchain/community/embeddings/deepinfra";
import { JinaEmbeddings } from "@langchain/community/embeddings/jina";
import { FireworksEmbeddings } from "@langchain/community/embeddings/fireworks";

HuggingFace Example

import { HuggingFaceInferenceEmbeddings } from "@langchain/community/embeddings/hf";

const embeddings = new HuggingFaceInferenceEmbeddings({
  model: "sentence-transformers/all-MiniLM-L6-v2",
  apiKey: process.env.HUGGINGFACEHUB_API_KEY,
});

const embedding = await embeddings.embedQuery("Hello world");

Local Embeddings

For privacy-sensitive applications, you can run embeddings locally:

Ollama

npm install @langchain/ollama

import { OllamaEmbeddings } from "@langchain/ollama";

const embeddings = new OllamaEmbeddings({
  model: "nomic-embed-text",
  baseUrl: "http://localhost:11434", // Default Ollama URL
});

const embedding = await embeddings.embedQuery("Private document content");

Transformers.js (Browser/Node)

npm install @langchain/community @xenova/transformers

import { HuggingFaceTransformersEmbeddings } from "@langchain/community/embeddings/huggingface_transformers";

const embeddings = new HuggingFaceTransformersEmbeddings({
  model: "Xenova/all-MiniLM-L6-v2",
});

const embedding = await embeddings.embedQuery("Embedded in browser or Node.js");

Common Patterns

Similarity Search

import { OpenAIEmbeddings } from "@langchain/openai";

const embeddings = new OpenAIEmbeddings();

// Generate embeddings
const query = "machine learning applications";
const queryEmbedding = await embeddings.embedQuery(query);

const documents = [
  "Machine learning is used in recommendation systems",
  "The weather today is sunny",
  "Deep learning powers image recognition",
];

const docEmbeddings = await embeddings.embedDocuments(documents);

// Calculate cosine similarity
function cosineSimilarity(a: number[], b: number[]): number {
  const dotProduct = a.reduce((sum, val, i) => sum + val * b[i], 0);
  const magnitudeA = Math.sqrt(a.reduce((sum, val) => sum + val * val, 0));
  const magnitudeB = Math.sqrt(b.reduce((sum, val) => sum + val * val, 0));
  return dotProduct / (magnitudeA * magnitudeB);
}

const similarities = docEmbeddings.map(docEmbed => 
  cosineSimilarity(queryEmbedding, docEmbed)
);

console.log(similarities); // [0.85, 0.12, 0.82] - first and third docs are most similar

Caching for Performance

import { OpenAIEmbeddings } from "@langchain/openai";
import { CacheBackedEmbeddings } from "@langchain/core/embeddings";
import { InMemoryStore } from "@langchain/core/stores";

const underlyingEmbeddings = new OpenAIEmbeddings();
const inMemoryStore = new InMemoryStore();

const cachedEmbeddings = CacheBackedEmbeddings.fromBytesStore(
  underlyingEmbeddings,
  inMemoryStore,
  {
    namespace: "my-app-embeddings",
  }
);

// First call hits the API
const embedding1 = await cachedEmbeddings.embedQuery("sample text");

// Second call uses cache
const embedding2 = await cachedEmbeddings.embedQuery("sample text");

Best Practices

Choose the right model: Balance cost, quality, and dimension size
Batch when possible: Embed multiple documents in one call for efficiency
Cache embeddings: Store embeddings to avoid redundant API calls
Normalize inputs: Consistent text preprocessing improves quality
Monitor costs: Track embedding generation for budget management
Consider local models: For privacy or high-volume applications

Dimension Comparison

Provider	Model	Dimensions	Use Case
OpenAI	text-embedding-3-small	1536	General purpose, cost-effective
OpenAI	text-embedding-3-large	3072	High quality, detailed semantics
Cohere	embed-english-v3.0	1024	Multilingual, task-specific
Google	text-embedding-004	768	Google ecosystem integration
Nomic	nomic-embed-text-v1.5	768	Open-source, local deployment

Next Steps

Vector Stores

Store and search embeddings efficiently

Document Loaders

Load documents for embedding

Retrieval

Build RAG applications with embeddings

Chat Models

Combine embeddings with LLMs

Getting Started

Core Concepts

Guides

Integrations

Development

​Popular Embedding Providers

OpenAI

Cohere

Google

Nomic

​OpenAI Embeddings

​Installation

​Usage

​Batch Embeddings

​Model Options

​Cohere Embeddings

​Installation

​Usage

​Input Types

​Google Embeddings

​Installation

​Generative AI (API Key)

​Vertex AI

​Nomic Embeddings

​Installation

​Usage

​Additional Providers

AWS Bedrock

Azure OpenAI

Ollama

Mixedbread AI

​Community Integrations

​HuggingFace Example

​Local Embeddings

​Ollama

​Transformers.js (Browser/Node)

​Common Patterns

​Similarity Search

​Caching for Performance

​Best Practices

​Dimension Comparison

​Next Steps

Vector Stores

Document Loaders

Retrieval

Chat Models

Build docs developers (and LLMs) love

Popular Embedding Providers

OpenAI Embeddings

Installation

Usage

Batch Embeddings

Model Options

Cohere Embeddings

Installation

Usage

Input Types

Google Embeddings

Installation

Generative AI (API Key)

Vertex AI

Nomic Embeddings

Installation

Usage

Additional Providers

Community Integrations

HuggingFace Example

Local Embeddings

Ollama

Transformers.js (Browser/Node)

Common Patterns

Similarity Search

Caching for Performance

Best Practices

Dimension Comparison

Next Steps