Langchain Vector Stores

langchain-vector-stores (JavaScript/TypeScript)

Overview

Vector stores are databases optimized for storing and searching high-dimensional vectors (embeddings). They enable semantic search by finding documents similar to a query based on vector similarity rather than keyword matching. Essential for RAG (Retrieval-Augmented Generation) systems.

Key Concepts

• Vector Database: Specialized database for storing and querying embeddings
• Similarity Search: Finding similar documents using vector distance metrics (cosine, euclidean)
• Metadata Filtering: Combining vector search with metadata filters
• Persistence: Some vector stores run in-memory, others persist to disk/cloud
• Scaling: Different stores have different scalability characteristics

Vector Store Selection Decision Table

When to Choose Each Store

Choose FAISS if:

• You need high performance local vector search
• You want to avoid external dependencies
• You have many vectors (millions) to search quickly

Choose Chroma if:

• You want simple local development
• You need easy persistence without complex setup
• You're building a prototype or small application

Choose Pinecone if:

• You're building production applications
• You want zero operational overhead
• You need auto-scaling and high availability

Choose Memory Vector Store if:

• You're testing or prototyping
• Data persistence isn't needed
• You want the simplest possible setup

Choose Weaviate/Qdrant if:

• You need advanced filtering and hybrid search
• You want flexibility in deployment (cloud or self-hosted)
• You need high performance at scale

Code Examples

Memory Vector Store (Simplest)

import { MemoryVectorStore } from "langchain/vectorstores/memory";
import { OpenAIEmbeddings } from "@langchain/openai";

// In-memory vector store - great for testing
const vectorStore = new MemoryVectorStore(new OpenAIEmbeddings());

// Add documents
await vectorStore.addDocuments([
  { pageContent: "LangChain is a framework for LLM apps", metadata: { source: "docs" } },
  { pageContent: "Vector stores enable semantic search", metadata: { source: "docs" } },
  { pageContent: "Paris is the capital of France", metadata: { source: "wiki" } },
]);

// Similarity search
const results = await vectorStore.similaritySearch("What is LangChain?", 2);
console.log(results);

// Search with score
const resultsWithScore = await vectorStore.similaritySearchWithScore("LangChain", 2);
resultsWithScore.forEach(([doc, score]) => {
  console.log(`Score: ${score}, Content: ${doc.pageContent}`);
});

FAISS Vector Store

import { FaissStore } from "@langchain/community/vectorstores/faiss";
import { OpenAIEmbeddings } from "@langchain/openai";

const embeddings = new OpenAIEmbeddings();

// Create from documents
const vectorStore = await FaissStore.fromDocuments(
  [
    { pageContent: "Document 1 content", metadata: { id: 1 } },
    { pageContent: "Document 2 content", metadata: { id: 2 } },
  ],
  embeddings
);

// Search
const results = await vectorStore.similaritySearch("query", 3);

// Save to disk
await vectorStore.save("./faiss_index");

// Load from disk
const loadedStore = await FaissStore.load("./faiss_index", embeddings);

Chroma Vector Store

import { Chroma } from "@langchain/community/vectorstores/chroma";
import { OpenAIEmbeddings } from "@langchain/openai";

// Requires Chroma server running: docker run -p 8000:8000 chromadb/chroma
const vectorStore = await Chroma.fromDocuments(
  [
    { pageContent: "Text 1", metadata: { category: "A" } },
    { pageContent: "Text 2", metadata: { category: "B" } },
  ],
  new OpenAIEmbeddings(),
  {
    collectionName: "my-collection",
    url: "http://localhost:8000", // Chroma server URL
  }
);

// Search with metadata filter
const results = await vectorStore.similaritySearch("query", 3, {
  category: "A"
});

// Delete collection
await vectorStore.delete({ collectionName: "my-collection" });

Pinecone Vector Store

import { PineconeStore } from "@langchain/pinecone";
import { Pinecone } from "@pinecone-database/pinecone";
import { OpenAIEmbeddings } from "@langchain/openai";

// Initialize Pinecone client
const pinecone = new Pinecone({
  apiKey: process.env.PINECONE_API_KEY,
});

const pineconeIndex = pinecone.Index("my-index");

// Create vector store
const vectorStore = await PineconeStore.fromDocuments(
  [
    { pageContent: "Content 1", metadata: { topic: "tech" } },
    { pageContent: "Content 2", metadata: { topic: "science" } },
  ],
  new OpenAIEmbeddings(),
  {
    pineconeIndex,
    maxConcurrency: 5,
  }
);

// Search with metadata filter
const results = await vectorStore.similaritySearch("query", 3, {
  topic: "tech"
});

// Use as retriever
const retriever = vectorStore.asRetriever({
  k: 5,
  searchType: "similarity",
});
const docs = await retriever.getRelevantDocuments("query");

Adding Documents Incrementally

import { MemoryVectorStore } from "langchain/vectorstores/memory";
import { OpenAIEmbeddings } from "@langchain/openai";

const vectorStore = new MemoryVectorStore(new OpenAIEmbeddings());

// Add documents one at a time or in batches
await vectorStore.addDocuments([
  { pageContent: "Document 1", metadata: {} },
]);

// Add more later
await vectorStore.addDocuments([
  { pageContent: "Document 2", metadata: {} },
  { pageContent: "Document 3", metadata: {} },
]);

// Or from texts
await vectorStore.addTexts(
  ["Text 1", "Text 2"],
  [{ source: "A" }, { source: "B" }]
);

Using as a Retriever

import { MemoryVectorStore } from "langchain/vectorstores/memory";
import { OpenAIEmbeddings } from "@langchain/openai";

const vectorStore = await MemoryVectorStore.fromDocuments(
  documents,
  new OpenAIEmbeddings()
);

// Convert to retriever
const retriever = vectorStore.asRetriever({
  k: 4, // Return top 4 results
  searchType: "similarity", // or "mmr" for maximum marginal relevance
});

// Use in a chain
import { ChatOpenAI } from "@langchain/openai";
import { createRetrievalChain } from "langchain/chains/retrieval";
import { createStuffDocumentsChain } from "langchain/chains/combine_documents";
import { ChatPromptTemplate } from "@langchain/core/prompts";

const llm = new ChatOpenAI();
const prompt = ChatPromptTemplate.fromTemplate(`
Answer based on context:
{context}

Question: {input}
`);

const combineDocsChain = await createStuffDocumentsChain({ llm, prompt });
const chain = await createRetrievalChain({
  retriever,
  combineDocsChain,
});

const result = await chain.invoke({ input: "What is LangChain?" });

Maximum Marginal Relevance (MMR)

import { MemoryVectorStore } from "langchain/vectorstores/memory";
import { OpenAIEmbeddings } from "@langchain/openai";

const vectorStore = await MemoryVectorStore.fromTexts(
  ["text1", "text2", "text3"],
  [{}, {}, {}],
  new OpenAIEmbeddings()
);

// MMR balances relevance and diversity
const results = await vectorStore.maxMarginalRelevanceSearch("query", {
  k: 3,
  fetchK: 10, // Fetch 10 candidates, return 3 diverse results
  lambda: 0.5, // 0 = max diversity, 1 = max relevance
});

Boundaries

What Agents CAN Do

✅ Initialize vector stores

• Set up any supported vector store
• Configure with embeddings and connection details

✅ Add and query documents

• Add documents with metadata
• Perform similarity search
• Use metadata filters

✅ Persist and load

• Save vector stores to disk (FAISS, Chroma)
• Load existing vector stores
• Manage collections

✅ Use as retrievers

• Convert vector stores to retrievers
• Integrate with chains and agents
• Configure search parameters (k, filters, etc.)

✅ Choose appropriate store

• Select based on scale, performance, persistence needs
• Switch between stores with minimal code changes

What Agents CANNOT Do

❌ Mix embeddings from different models

• Cannot use different embedding models within same vector store
• Must use consistent embeddings

❌ Bypass provider limits

• Cannot exceed Pinecone index size limits
• Cannot bypass free tier restrictions

❌ Modify vector dimensions after creation

• Cannot change embedding dimensions once store is created
• Must recreate store with new embeddings

❌ Query without proper setup

• Cannot use Chroma without server running
• Cannot use Pinecone without API key and index

Gotchas

1. Embedding Model Consistency

// ❌ BAD: Different embeddings for indexing and querying
const store1 = await MemoryVectorStore.fromDocuments(
  docs,
  new OpenAIEmbeddings({ model: "text-embedding-3-small" })
);

const results = await store1.similaritySearch("query"); // Uses same embeddings ✓

// But if you recreate:
const store2 = new MemoryVectorStore(
  new OpenAIEmbeddings({ model: "text-embedding-ada-002" }) // Different!
);
// Queries won't work correctly!

// ✅ GOOD: Keep embedding instance
const embeddings = new OpenAIEmbeddings({ model: "text-embedding-3-small" });
const store = await MemoryVectorStore.fromDocuments(docs, embeddings);
// Always use same embeddings instance

Fix: Use the same embedding model instance for creation and queries.

2. Chroma Server Must Be Running

// ❌ Chroma not running
import { Chroma } from "@langchain/community/vectorstores/chroma";

const store = await Chroma.fromDocuments(docs, embeddings, {
  url: "http://localhost:8000"
});
// Error: Connection refused!

// ✅ Start Chroma first
// Terminal: docker run -p 8000:8000 chromadb/chroma
// Or: chroma run --path ./chroma_data

const store = await Chroma.fromDocuments(docs, embeddings, {
  url: "http://localhost:8000"
}); // Works!

Fix: Ensure Chroma server is running before creating the vector store.

3. Pinecone Index Must Exist

// ❌ Index doesn't exist
import { Pinecone } from "@pinecone-database/pinecone";

const pinecone = new Pinecone({ apiKey: process.env.PINECONE_API_KEY });
const index = pinecone.Index("nonexistent-index"); // Won't create index!

// ✅ Create index first
// Do this in Pinecone dashboard or via API
const pinecone = new Pinecone({ apiKey: process.env.PINECONE_API_KEY });

// Check if index exists, create if needed
const indexList = await pinecone.listIndexes();
if (!indexList.indexes?.some(idx => idx.name === "my-index")) {
  await pinecone.createIndex({
    name: "my-index",
    dimension: 1536, // Must match embedding dimensions
    metric: "cosine",
    spec: { serverless: { cloud: "aws", region: "us-east-1" } }
  });
}

const index = pinecone.Index("my-index");

Fix: Create Pinecone index before using it.

4. FAISS Save/Load Path Issues

// ❌ Relative path issues
await vectorStore.save("./index"); // May fail depending on cwd

// ✅ Use absolute paths or be explicit
import path from "path";
const indexPath = path.join(process.cwd(), "data", "faiss_index");
await vectorStore.save(indexPath);

// Load with same path
const loadedStore = await FaissStore.load(indexPath, embeddings);

Fix: Use absolute paths or be careful with working directory.

5. Memory Vector Store is Ephemeral

// ❌ Expecting persistence
const vectorStore = new MemoryVectorStore(embeddings);
await vectorStore.addDocuments(docs);
// App restarts...
// All data is lost!

// ✅ Use persistent store for production
import { FaissStore } from "@langchain/community/vectorstores/faiss";

const vectorStore = await FaissStore.fromDocuments(docs, embeddings);
await vectorStore.save("./faiss_index"); // Persists to disk

Fix: Use FAISS, Chroma, or cloud stores (Pinecone) for persistent data.

6. Metadata Filtering Syntax Varies

// Different stores have different filter syntaxes
// Pinecone
const pineconeResults = await pineconeStore.similaritySearch("query", 3, {
  category: "tech" // Simple key-value
});

// Chroma
const chromaResults = await chromaStore.similaritySearch("query", 3, {
  where: { category: "tech" } // Nested structure
});

// Check each store's documentation for filter syntax!

Fix: Read the specific vector store's documentation for filter syntax.

7. Dimension Mismatch

// ❌ Creating Pinecone index with wrong dimensions
await pinecone.createIndex({
  name: "my-index",
  dimension: 1536, // OpenAI ada-002 dimensions
});

// But using different embeddings!
const store = await PineconeStore.fromDocuments(
  docs,
  new OpenAIEmbeddings({ model: "text-embedding-3-small", dimensions: 512 }),
  { pineconeIndex }
); // Error: dimension mismatch!

// ✅ Match dimensions
const embeddings = new OpenAIEmbeddings({ model: "text-embedding-3-small" });
// Default is 1536, matches Pinecone index

Fix: Ensure vector store dimension configuration matches embedding dimensions.

Links and Resources

Official Documentation

• LangChain JS Vector Stores
• FAISS
• Chroma
• Pinecone
• Memory Vector Store

Provider Documentation

• FAISS Library
• Chroma
• Pinecone
• Weaviate
• Qdrant

Package Installation

# Community package (includes FAISS, Chroma, etc.)
npm install @langchain/community

# Pinecone
npm install @langchain/pinecone @pinecone-database/pinecone

# Weaviate
npm install @langchain/weaviate weaviate-ts-client

# Qdrant
npm install @langchain/qdrant @qdrant/js-client-rest