Java library for the Qdrant vector search engine.

Qdrant Java Client

Java client library with handy utility methods and overloads for interfacing with Qdrant.

📥 Installation

Important

Requires Java 8 or above.

To install the library, add the following lines to your build config file.

Maven

<dependency>
  <groupId>io.qdrant</groupId>
  <artifactId>client</artifactId>
  <version>1.7.0</version>
</dependency>

Scala SBT

libraryDependencies += "io.qdrant" % "client" % "1.7.0"

Gradle

implementation 'io.qdrant:client:1.7.0'

📖 Documentation

• QdrantClient Reference

🔌 Getting started

Creating a client

A client can be instantiated with

QdrantClient client = 
  new QdrantClient(QdrantGrpcClient.newBuilder("localhost").build());

which creates a client that will connect to Qdrant on https://localhost:6334.

Internally, the high level client uses a low level gRPC client to interact with Qdrant. Additional constructor overloads provide more control over how the gRPC client is configured. The following example configures a client to use TLS, validating the certificate using the root CA to verify the server's identity instead of the system's default, and also configures API key authentication:

ManagedChannel channel = Grpc.newChannelBuilder(
  "localhost:6334",
  TlsChannelCredentials.newBuilder()
    .trustManager(new File("ssl/ca.crt"))
    .build())
.build();

QdrantClient client = new QdrantClient(
  QdrantGrpcClient.newBuilder(channel)
    .withApiKey("<apikey>")
    .build());

The client implements AutoCloseable, though a client will typically be created once and used for the lifetime of the application. When a client is constructed by passing a ManagedChannel, the client does not shut down the channel on close by default. The client can be configured to shut down the channel on close with

ManagedChannel channel = Grpc.newChannelBuilder(
  "localhost:6334", 
  TlsChannelCredentials.create())
.build();

QdrantClient client = new QdrantClient(
  QdrantGrpcClient.newBuilder(channel, true)
    .withApiKey("<apikey>")
    .build());

All client methods return ListenableFuture<T>.

Working with collections

Once a client has been created, create a new collection

client.createCollectionAsync("my_collection",
  VectorParams.newBuilder()
    .setDistance(Distance.Cosine)
    .setSize(4)
    .build())
  .get();

Insert vectors into a collection

// import static convenience methods
import static io.qdrant.client.PointIdFactory.id;
import static io.qdrant.client.ValueFactory.value;
import static io.qdrant.client.VectorsFactory.vector;

Random random = new Random();
List<PointStruct> points = IntStream.range(1, 101)
  .mapToObj(i -> PointStruct.newBuilder()
    .setId(id(i))
    .setVectors(vector(IntStream.range(1, 101)
        .mapToObj(v -> random.nextFloat())
        .collect(Collectors.toList())))
    .putAllPayload(ImmutableMap.of(
      "color", value("red"),
      "rand_number", value(i % 10))
    )
    .build()
  )
  .collect(Collectors.toList());

UpdateResult updateResult = client.upsertAsync("my_collection", points).get();

Search for similar vectors

List<Float> queryVector = IntStream.range(1, 101)
  .mapToObj(v -> random.nextFloat())
  .collect(Collectors.toList());

List<ScoredPoint> points = client.searchAsync(SearchPoints.newBuilder()
  .setCollectionName("my_collection")
  .addAllVector(queryVector)
  .setLimit(5)
  .build()
).get();

Search for similar vectors with filtering condition

// import static convenience methods
import static io.qdrant.client.ConditionFactory.range;

List<ScoredPoint> points = client.searchAsync(SearchPoints.newBuilder()
  .setCollectionName("my_collection")
  .addAllVector(queryVector)
  .setFilter(Filter.newBuilder()
    .addMust(range("rand_number", Range.newBuilder().setGte(3).build()))
    .build())
  .setLimit(5)
  .build()
).get();

⚖️ LICENSE

Apache 2.0 © 2023