Qdrant

Qdrant vector search engine integration

Overview

Qdrant is a high-performance vector search engine with advanced filtering capabilities and production-ready features.

Installation

npm install @llamaindex/qdrant @qdrant/js-client-rest

Basic Usage

import { QdrantVectorStore } from "@llamaindex/qdrant";
import { VectorStoreIndex, Document } from "llamaindex";

const vectorStore = new QdrantVectorStore({
  url: "http://localhost:6333",
  collectionName: "my-collection"
});

const documents = [
  new Document({ text: "LlamaIndex is a data framework." }),
  new Document({ text: "Qdrant is a vector search engine." })
];

const index = await VectorStoreIndex.fromDocuments(documents, {
  storageContext: { vectorStore }
});

const queryEngine = index.asQueryEngine();
const response = await queryEngine.query({
  query: "What is Qdrant?"
});

Constructor Options

url · string · default: "http://localhost:6333"

Qdrant server URL

collectionName · string · required

Name of the Qdrant collection

apiKey · string

API key for Qdrant Cloud

batchSize · number · default: 100

Batch size for operations

Running Qdrant

Docker

docker pull qdrant/qdrant
docker run -p 6333:6333 -p 6334:6334 qdrant/qdrant

Qdrant Cloud

Sign up at Qdrant Cloud and use the provided URL and API key:

const vectorStore = new QdrantVectorStore({
  url: "https://xyz.cloud.qdrant.io",
  apiKey: process.env.QDRANT_API_KEY,
  collectionName: "my-collection"
});

Collection Configuration

import { QdrantClient } from "@qdrant/js-client-rest";

const client = new QdrantClient({ url: "http://localhost:6333" });

await client.createCollection("my-collection", {
  vectors: {
    size: 1536,  // Embedding dimension
    distance: "Cosine"  // or "Euclid", "Dot"
  },
  optimizers_config: {
    default_segment_number: 2
  }
});

Querying

Basic Query

const index = await VectorStoreIndex.fromVectorStore(vectorStore);

const retriever = index.asRetriever({
  similarityTopK: 5
});

const nodes = await retriever.retrieve("query text");

Metadata Filtering

const documents = [
  new Document({
    text: "Doc 1",
    metadata: { category: "tech", year: 2023 }
  }),
  new Document({
    text: "Doc 2",
    metadata: { category: "science", year: 2024 }
  })
];

const index = await VectorStoreIndex.fromDocuments(documents, {
  storageContext: { vectorStore }
});

const retriever = index.asRetriever({
  filters: {
    must: [
      { key: "category", match: { value: "tech" } },
      { key: "year", range: { gte: 2023 } }
    ]
  }
});

Advanced Filtering

Qdrant supports complex filters:

const retriever = index.asRetriever({
  filters: {
    must: [
      { key: "category", match: { value: "tech" } }
    ],
    should: [
      { key: "tags", match: { any: ["ai", "ml"] } }
    ],
    must_not: [
      { key: "status", match: { value: "archived" } }
    ]
  }
});

Managing Collections

List Collections

const client = new QdrantClient({ url: "http://localhost:6333" });
const collections = await client.getCollections();
console.log(collections);

Delete Collection

await client.deleteCollection("my-collection");

Collection Info

const info = await client.getCollection("my-collection");
console.log("Vectors count:", info.vectors_count);
console.log("Indexed vectors:", info.indexed_vectors_count);

Distance Metrics

await client.createCollection("my-collection", {
  vectors: {
    size: 1536,
    distance: "Cosine"  // "Euclid", "Dot", or "Manhattan"
  }
});

Payloads (Metadata)

Qdrant stores metadata as payloads:

const doc = new Document({
  text: "Document text",
  metadata: {
    title: "My Document",
    author: "John Doe",
    tags: ["ai", "ml"],
    published: new Date("2024-01-01")
  }
});

Snapshots

Create and restore snapshots:

// Create snapshot
const snapshot = await client.createSnapshot("my-collection");
console.log("Snapshot:", snapshot.name);

// Restore snapshot
await client.recoverFromSnapshot("my-collection", {
  location: snapshot.name
});

Complete Example

import { QdrantVectorStore } from "@llamaindex/qdrant";
import { VectorStoreIndex, Document, Settings } from "llamaindex";
import { OpenAI, OpenAIEmbedding } from "@llamaindex/openai";
import { QdrantClient } from "@qdrant/js-client-rest";

// Configure settings
Settings.llm = new OpenAI({ model: "gpt-4" });
Settings.embedModel = new OpenAIEmbedding();

// Create collection
const client = new QdrantClient({ url: "http://localhost:6333" });

await client.createCollection("docs", {
  vectors: {
    size: 1536,  // OpenAI embedding dimension
    distance: "Cosine"
  }
});

// Create vector store
const vectorStore = new QdrantVectorStore({
  url: "http://localhost:6333",
  collectionName: "docs"
});

// Load documents
const documents = [
  new Document({
    text: "LlamaIndex documentation...",
    metadata: { source: "docs", page: 1 }
  })
];

// Build index
const index = await VectorStoreIndex.fromDocuments(documents, {
  storageContext: { vectorStore }
});

// Query
const queryEngine = index.asQueryEngine();
const response = await queryEngine.query({
  query: "What is LlamaIndex?"
});

console.log(response.response);

Best Practices

1. Choose appropriate distance metric: Cosine for text similarity
2. Use Qdrant Cloud for production: Managed, scalable solution
3. Leverage advanced filtering: Combine vector and metadata search
4. Create snapshots: Regular backups for data safety
5. Monitor performance: Use Qdrant’s dashboard
6. Optimize segment count: Adjust based on data size

See Also

• Qdrant Documentation
• Qdrant Cloud
• Vector Store Index