README

Introduction

This package provides an elegant, fluent interface for interacting with the Qdrant Vector Database in Laravel. Qdrant is a vector similarity search engine that makes it easy to store and search for embeddings, making it ideal for AI-powered applications.

Key features:

• Simple collection management
• Fluent search API with filtering and grouping
• Efficient point operations (insert, upsert, delete)
• Vector operations (update, delete)
• Laravel Facade support
• Convenient payload handling

Installation

1. Install via Composer

composer require mcpuishor/qdrant-laravel

2. Publish the Configuration File

php artisan vendor:publish --tag=qdrant-laravel-config

This will create a config/qdrant-laravel.php file where you can set your Qdrant connections and defaults.

3. Set Up Your .env File

Update your .env file with your Qdrant host details:

QDRANT_DEFAULT=main
QDRANT_HOST=http://localhost:6333
QDRANT_COLLECTION=collection_name
QDRANT_VECTOR_SIZE=1536
QDRANT_DEFAULT_DISTANCE_METRIC=Cosine

Configuration

The config/qdrant-laravel.php file allows multiple connections:

return [
    'default' => env('QDRANT_DEFAULT', 'main'),

    'connections' => [
        'main' => [
            'host' => env('QDRANT_HOST', 'http://localhost:6333'),
            'api_key' => env('QDRANT_API_KEY', null),
            'collection' => env('QDRANT_COLLECTION', 'default_collection'),
            'vector_size' => env('QDRANT_VECTOR_SIZE', 128),
        ],
    ],

    'default_distance_metric' => env('QDRANT_DEFAULT_DISTANCE_METRIC', 'Cosine'),
];

Schema Management (Migrations)

Creating a new collection

A collection must contain at least one vector. An optional parameter options can contain additional parameters described as an associative array. See the Qdrant documentation for details. The options can be specified using arrays or DataObjects defined in the package.

The response is a boolean value, unless an exception is thrown.

use \Mcpuishor\QdrantLaravel\Facades\Schema;
use \Mcpuishor\QdrantLaravel\Enums\DistanceMetric;
use \Mcpuishor\QdrantLaravel\DTOs\Vector;

$vector = Vector::fromArray([
            'size' => 128,
            'distance' => DistanceMetric::COSINE
       ]);

$response = Schema::create(
                   name: "new_collection",
                   vector: $vector,
                   options: []
                );

if ($response) {
    echo "Schema created successfully";
}

Creating a new collection on a different connection

You can switch the connection at runtime. The connection must be defined in the config\qdrant-laravel.php file.

use \Mcpuishor\QdrantLaravel\Schema;
use \Mcpuishor\QdrantLaravel\Enums\DistanceMetric;
use \Mcpuishor\QdrantLaravel\DTOs\Vector;

$vector = Vector::fromArray([
            'size' => 128,
            'distance' => DistanceMetric::COSINE
       ]);

$response = Schema::connection('backup')
                ->create(
                   name: "new_collection",
                   vector: $vector,
                );

if ($response) {
    echo "Schema created successfully";
}

Creating a collection with multiple vectors

A collection can contain multiple vectors per point. They need to be passed on to the Schema::create as an array containing the definitions of each vector. The vectors can have different definitions. The optional parameters can be specified using Data Objects defined in the package.

use \Mcpuishor\QdrantLaravel\Schema;
use \Mcpuishor\QdrantLaravel\QdrantTransport;
use \Mcpuishor\QdrantLaravel\Enums\DistanceMetric;
use \Mcpuishor\QdrantLaravel\DTOs\Vector;
use \Mcpuishor\QdrantLaravel\DTOs\HnswConfig;

$vector1 = Vector::fromArray([
            'size' => 128,
            'distance' => DistanceMetric::COSINE
            //optional parameters
            'on_disk' => true,
            ]);

$vector2 = Vector::fromArray([
            'size' => 1024,
            'distance' => DistanceMetric::COSINE,
            //optional parameters
            'hsnw_config' => Hnswconfig::fromArray([
                    'm' => 10,
                    'ef_construct' => 4,
                    'on_disk' => true,
                ]),
            ]);

$response = Schema::create(
               name: "new_collection",
               vector: array($vector1, $vector2),
            );

if ($response) {
    echo "Schema created successfully";
}

Deleting a collection

To delete a collection, you can call the delete method on the Schema facade. It returns a Mcpuishor\QdrantLaravel\DTOs\Response object.

use \Mcpuishor\QdrantLaravel\Facades\Schema;

$result = Schema::delete('collection_name');

if ($result) {
    echo "Collection has been successfully deleted.";
}

Collection existence

To check if the collection defined in the config on the current connection exists:

use \Mcpuishor\QdrantLaravel\Facades\Schema;

if ( Schema::exists() ) {
    echo "Collection exists.";
}

At the same time, you can check the existence of a different collection on the same connection:

use \Mcpuishor\QdrantLaravel\Facades\Schema;

if ( Schema::exists( 'another_collection' ) ) {
    echo "Collection 'another_collection' exists.";
}

Updating a collection

Updating parameters on an existing collection can be done in a similar fashion to creating one. The parameters updated can be specified using arrays or Data Objects defined in the package.

Updating the collection defined in the config\qdrant-laravel.php:

use \Mcpuishor\QdrantLaravel\Facades\Schema;
use \Mcpuishor\QdrantLaravel\DTOs\HnswConfig;
use \Mcpuishor\QdrantLaravel\DTOs\Collection\Params;

Schema::update(
    vectors: [

    ], 
    options: [
       'hnsw_config' => HnswConfig::fromArray([
                'm' => 100,
                'ef_construct' => 5,
            ]),
       'params' => Params::fromArray([
                'replication_factor' => 4,
                'on_disk_payload' => true,
            ]),
    ]
);

Indexing a collection

Indexes in a Qdrant vector collection are created on the payload for each vector. For more details see the Qdrant documentation.

Creating an index

To create a payload index over a field:

use \Mcpuishor\QdrantLaravel\Facades\Qdrant;
use \Mcpuishor\QdrantLaravel\Enums\FieldType;

$result = Qdrant::indexes()->add('field_name', FieldType::KEYWORD);

It returns true if the operation was successful, or false otherwise.

You can use dot notation to create indexes over nested fields.

By default, indexes are stored in memory. If you have large indexes, and they need to be stored on the disk, you can use the ->onDisk() method before creating the index. Choose carefully when to store an index on the disk, as this will introduce some latency in your future queries.

Parameterized integer indexes

Qdrant v1.8.0 has introduced a parameterized variant of the integer index. To turn the parameterized index on you can call the ->parameterized() method before creating an integer index. This setting is used only for integer fields in the payload.

Values of the lookup and range can be toggled in the config\qdrant-laravel.php file. For more information on parameterized integer indexes and how they affect performance check the Qdrant documentation

    $result = Qdrant::indexes()->parameterized()->add('field_name', FieldType::INTEGER);

It returns true if the operation was successful, or false otherwise.

Full-text indexes

Qdrant supports full-text search for string payload. Full-text index allows you to filter points by the presence of a word or a phrase in the payload field.

    use \Mcpuishor\QdrantLaravel\Enums\TokenizerType;
    use \Mcpuishor\QdrantLaravel\Facades\Qdrant;

    $result = Qdrant::indexes()->fulltext('text_field_name', TokenizerType::WORD);

It returns true if the operation was successful, or false otherwise.

Deleting an index

    use \Mcpuishor\QdrantLaravel\Facades\Qdrant;

    $result = Qdrant::indexes()->delete('payload_field');

It returns true if the operation was successful, or false otherwise.

Searching

The package provides a fluent interface for searching vectors in your Qdrant collection.

Basic Vector Search

To perform a simple search with a vector:

use Mcpuishor\QdrantLaravel\Facades\Qdrant;

// Search using a vector
$results = Qdrant::search()
    ->vector([0.2, 0.3, 0.4, ...]) // Your vector data
    ->limit(10)
    ->get();

Search by Point ID

You can also search for similar points to an existing point by its ID:

use Mcpuishor\QdrantLaravel\Facades\Qdrant;
use Mcpuishor\QdrantLaravel\DTOs\Point;

$point = new Point(id: 123);
$results = Qdrant::search()
    ->point($point)
    ->limit(5)
    ->get();

Including Payload and Vectors

Control what data is returned with your search results:

// Include all payload data
$results = Qdrant::search()
    ->vector($vector)
    ->withPayload()
    ->get();

// Include only specific payload fields
$results = Qdrant::search()
    ->vector($vector)
    ->include(['name', 'description'])
    ->get();

// Exclude specific payload fields
$results = Qdrant::search()
    ->vector($vector)
    ->exclude(['internal_id'])
    ->get();

// Include vector data in results
$results = Qdrant::search()
    ->vector($vector)
    ->withVectors()
    ->get();

Pagination

Control the number of results and implement pagination:

// Limit results
$results = Qdrant::search()
    ->vector($vector)
    ->limit(20)
    ->get();

// Pagination with offset
$results = Qdrant::search()
    ->vector($vector)
    ->limit(10)
    ->offset(20) // Skip first 20 results
    ->get();

Filtering Results

Apply filters to search results (and to count, scroll, facet, discover and matrix, since they all share the same HasFilters trait) using must / mustNot / should / minShould, each taking a payload key, a FilterConditions case, and the operand for that condition:

use Mcpuishor\QdrantLaravel\Enums\FilterConditions;

// Match filter (equality) — the value is wrapped under `match.value`
$results = Qdrant::search()
    ->vector($vector)
    ->must('category', FilterConditions::MATCH, 'electronics')
    ->get();

// Range filter — the operand is passed through as-is (gte/lte/gt/lt)
$results = Qdrant::search()
    ->vector($vector)
    ->must('price', FilterConditions::RANGE, ['gte' => 100, 'lte' => 500])
    ->get();

// Combine MUST and MUST NOT
$results = Qdrant::search()
    ->vector($vector)
    ->must('category', FilterConditions::MATCH, 'electronics')
    ->mustNot('discontinued', FilterConditions::MATCH, true)
    ->get();

// SHOULD (at least one of these should match) and MIN SHOULD (at least N of these)
$results = Qdrant::search()
    ->vector($vector)
    ->should('category', FilterConditions::MATCH, 'electronics')
    ->should('category', FilterConditions::MATCH, 'gadgets')
    ->minShould('tag', FilterConditions::MATCH, 'featured', min_count: 1)
    ->get();

// is_empty / is_null only need the key, no value
$results = Qdrant::search()
    ->vector($vector)
    ->must('description', FilterConditions::IS_EMPTY)
    ->get();

Note (issue #3, fixed in 0.2.0): filter conditions used to serialize incorrectly for several FilterConditions cases. As of 0.2.0, match correctly wraps its operand under {"match": {"value": ...}}, range/geo_bounding_box/geo_polygon/geo_radius/values_count pass their operand through unwrapped (e.g. {"range": {"gte": 100}}), and is_empty/is_null emit {"is_empty": {"key": "..."}} with no value at all. If you built filters against a version prior to 0.2.0, re-check any range, is_empty, or is_null filters — their JSON shape has changed.

Grouping Results

Group search results by a payload field:

// Group results by category
$results = Qdrant::search()
    ->vector($vector)
    ->groupBy('category', 5) // 5 results per group
    ->get();

Batch Searching

Perform multiple searches in a single request:

$search1 = Qdrant::search()->vector($vector1)->limit(5);
$search2 = Qdrant::search()->vector($vector2)->limit(5);

$batchResults = Qdrant::search()->batch([$search1, $search2]);

Random Sampling

Get random points from the collection:

$randomPoints = Qdrant::search()->random();

Using Named Vectors

If your collection has multiple named vectors, specify which one to use:

$results = Qdrant::search()
    ->vector($vector)
    ->using('image_embedding') // Use the named vector
    ->get();

Recommendations

The package provides a recommendation system based on positive and negative examples.

Basic Recommendations

Get recommendations based on positive examples:

use Mcpuishor\QdrantLaravel\Facades\Qdrant;

// Recommend based on point IDs
$recommendations = Qdrant::recommend()
    ->positive([123, 456]) // Points you like
    ->limit(10)
    ->get();

Positive and Negative Examples

Refine recommendations with both positive and negative examples:

$recommendations = Qdrant::recommend()
    ->positive([123, 456]) // Points you like
    ->negative([789, 101]) // Points you don't like
    ->limit(10)
    ->get();

Recommendation Strategy

Control how vectors are combined for recommendations:

use Mcpuishor\QdrantLaravel\Enums\AverageVectorStrategy;

$recommendations = Qdrant::recommend()
    ->positive([123, 456])
    ->strategy(AverageVectorStrategy::WEIGHTED) // Use weighted average
    ->limit(10)
    ->get();

Available strategies include:

• AverageVectorStrategy::MEAN - Simple average of vectors
• AverageVectorStrategy::WEIGHTED - Weighted average based on similarity

Point Operations

The package provides methods for managing points in your Qdrant collection.

Retrieving Points

Get points by their IDs:

use Mcpuishor\QdrantLaravel\Facades\Qdrant;

// Get multiple points
$points = Qdrant::points()->get([123, 456, 789]);

// Find a single point
$point = Qdrant::points()->find(123);

Controlling Returned Data

Control what data is returned with the points:

// With payload (default)
$points = Qdrant::points()->withPayload()->get([123, 456]);

// Without payload
$points = Qdrant::points()->withoutPayload()->get([123, 456]);

// With vector data
$points = Qdrant::points()->withVector()->get([123, 456]);

// Without vector data (default)
$points = Qdrant::points()->withoutVector()->get([123, 456]);

Inserting Points

Insert a new point into the collection:

use Mcpuishor\QdrantLaravel\DTOs\Point;

// Create a point
$point = new Point(
    id: 123,
    vector: [0.2, 0.3, 0.4, ...],
    payload: ['name' => 'Example', 'category' => 'test']
);

// Insert the point
$success = Qdrant::points()->insert($point);

Upserting Points

Insert or update multiple points:

use Mcpuishor\QdrantLaravel\PointsCollection;
use Mcpuishor\QdrantLaravel\DTOs\Point;

// Create points collection
$points = new PointsCollection([
    new Point(id: 123, vector: [0.2, 0.3, 0.4, ...], payload: ['name' => 'First']),
    new Point(id: 456, vector: [0.5, 0.6, 0.7, ...], payload: ['name' => 'Second'])
]);

// Upsert the points
$success = Qdrant::points()->upsert($points);

Deleting Points

Delete points by their IDs:

// Delete specific points
$success = Qdrant::points()->delete([123, 456]);

// Delete points matching a filter
$success = Qdrant::points()
    ->where('category', '=', 'test')
    ->delete([]);

Autochunking

Efficiently handle large numbers of points with automatic chunking:

// Create an autochunker with chunk size of 100
$chunker = Qdrant::points()->autochunk(100);

// Add points - they'll be automatically upserted when the chunk size is reached
foreach ($largeDataset as $data) {
    $point = new Point(
        id: $data['id'],
        vector: $data['embedding'],
        payload: $data['metadata']
    );
    $chunker->add($point);
}

// Manually flush any remaining points
$chunker->flush();

Vector Operations

The package provides methods for managing vectors in your Qdrant collection.

Updating Vectors

Update vectors for existing points:

use Mcpuishor\QdrantLaravel\Facades\Qdrant;
use Mcpuishor\QdrantLaravel\PointsCollection;
use Mcpuishor\QdrantLaravel\DTOs\Point;

// Create a collection of points with updated vectors
$points = new PointsCollection([
    new Point(id: 123, vector: [0.2, 0.3, 0.4, ...]),
    new Point(id: 456, vector: [0.5, 0.6, 0.7, ...])
]);

// Update the vectors
$success = Qdrant::vectors()->update($points);

Deleting Vectors

Delete vectors for specific points:

use Mcpuishor\QdrantLaravel\Facades\Qdrant;

// Delete vectors for specific points
$success = Qdrant::vectors()->delete([123, 456]);

Collection Aliases

Aliases let you point a stable name at a collection and swap the underlying collection atomically — useful for zero-downtime reindexing (build plants_v2, then repoint the plants alias to it).

Alias mutations are queued fluently and committed with a single apply() call:

use Mcpuishor\QdrantLaravel\Facades\Qdrant;

// Create an alias, or several, then commit them in one atomic request
Qdrant::collection('plants')->aliases()
    ->add('plants', 'plants_v2')   // alias name, target collection
    ->delete('plants_old')
    ->apply(); // bool

// Atomically move an alias from its current target to a new collection
Qdrant::collection('plants')->aliases()
    ->switch('plants', 'plants_v2') // delete + re-add in one request
    ->apply();

apply() throws a CommandException if no actions have been queued.

List aliases — scoped to the current collection, or all aliases across the server:

// Aliases pointing at the current collection
$aliases = Qdrant::collection('plants')->aliases()->get(); // Illuminate\Support\Collection

// All aliases on the server (no collection scope)
$all = Qdrant::aliases()->get();

New in 0.2.0

Version 0.2.0 adds full coverage of the Qdrant 1.18.x REST API — counting, scrolling, batch updates, named-vector management, facets, discovery, a distance matrix, service/health/telemetry, snapshots (collection/storage/shard), cluster management, shard keys, and beta issues — plus the issue #3 filter fix described above.

Counting Points

Count points matching (optional) filters without fetching them:

use Mcpuishor\QdrantLaravel\Facades\Qdrant;
use Mcpuishor\QdrantLaravel\Enums\FilterConditions;

$total = Qdrant::collection('plants')
    ->count()
    ->exact()
    ->must('category', FilterConditions::MATCH, 'tropical')
    ->get(); // int

Scrolling Through Points

Page through a collection's points without vector search, optionally ordered by a payload key:

$scroll = Qdrant::collection('plants')
    ->scroll()
    ->limit(50)
    ->orderBy('created_at', 'desc')
    ->withPayload()
    ->withVector();

$page = $scroll->get();               // PointsCollection
$nextOffset = $scroll->nextPageOffset(); // pass this to ->offset() on the next call

Batch Updates

Combine multiple point/payload/vector operations into a single request:

use Mcpuishor\QdrantLaravel\PointsCollection;

$success = Qdrant::collection('plants')
    ->batch()
    ->upsert($pointsCollection) // a PointsCollection
    ->deletePoints([1, 2, 3])
    ->setPayload(['watered' => true], points: [4, 5])
    ->clearPayload(points: [6])
    ->updateVectors($otherPointsCollection)
    ->deleteVectors(ids: [7, 8], vectorNames: ['image_embedding'])
    ->execute(); // bool

Named Vectors

Add, remove, or inspect named vector configurations on an existing collection:

use Mcpuishor\QdrantLaravel\DTOs\Vector;
use Mcpuishor\QdrantLaravel\Enums\DistanceMetric;

Qdrant::collection('plants')->namedVectors()->create(
    'image_embedding',
    Vector::fromArray(['size' => 512, 'distance' => DistanceMetric::COSINE])
);

Qdrant::collection('plants')->namedVectors()->delete('image_embedding');

$status = Qdrant::collection('plants')->namedVectors()->optimizations(); // array

Facets

Get distinct payload values (and their counts) for a key, similar to a search facet/aggregation:

$facets = Qdrant::collection('plants')
    ->facet('category')
    ->limit(20)
    ->exact()
    ->get(); // FacetResponse

foreach ($facets->hits() as $hit) {
    // ['value' => ..., 'count' => ...]
}

Discovery

Find points using positive/negative context pairs plus an optional target — Qdrant's discovery search:

$results = Qdrant::collection('plants')
    ->discover()
    ->target(123)
    ->context([['positive' => 456, 'negative' => 789]])
    ->using('image_embedding')
    ->limit(10)
    ->get(); // PointsCollection

// Batch discovery
$batchResults = Qdrant::collection('plants')->discover()->batch([$discover1, $discover2]);

Recommendations on the Query API

Qdrant::recommend() is rebuilt on top of the Query API (POST /points/query) internally, so the public interface is unchanged — see the Recommendations section above.

Distance Matrix

Compute pairwise distances between a sample of points:

$offsets = Qdrant::collection('plants')
    ->matrix()
    ->sample(50)
    ->limit(10)
    ->using('image_embedding')
    ->offsets(); // array

$pairs = Qdrant::collection('plants')->matrix()->sample(50)->pairs(); // array

Service, Health, and Telemetry

Inspect the Qdrant server itself, independent of any collection:

Qdrant::service()->root();          // array — server identity/version
Qdrant::service()->healthz();       // bool
Qdrant::service()->livez();         // bool
Qdrant::service()->readyz();        // bool
Qdrant::service()->telemetry();     // array
Qdrant::service()->metrics();       // string — Prometheus text format

Collection Snapshots

Create, list, delete, and download snapshots of a collection:

$snapshot = Qdrant::collection('plants')->snapshots()->create(); // SnapshotDescription

Qdrant::collection('plants')->snapshots()->list();               // Collection<SnapshotDescription>
Qdrant::collection('plants')->snapshots()->delete($snapshot->name);
Qdrant::collection('plants')->snapshots()->download($snapshot->name); // Illuminate\Http\Client\Response

// Recover from (register) a snapshot that already exists at a server-visible location
Qdrant::collection('plants')->snapshots()->recover('file:///qdrant/snapshots/plants/plants.snapshot');

Limitation: snapshots()->upload($path) currently throws a SnapshotException — multipart upload of local file bytes from your application to the server is not yet implemented. Use recover($location) with a path the Qdrant server can already see instead.

Storage Snapshots

Snapshot the entire storage (not tied to a single collection):

$snapshot = Qdrant::storageSnapshots()->create(); // SnapshotDescription
Qdrant::storageSnapshots()->list();
Qdrant::storageSnapshots()->delete($snapshot->name);
Qdrant::storageSnapshots()->download($snapshot->name);

Shard Snapshots

Snapshot an individual shard in a distributed deployment:

$snapshot = Qdrant::collection('plants')->shardSnapshots(shardId: 0)->create(); // SnapshotDescription
Qdrant::collection('plants')->shardSnapshots(0)->list();
Qdrant::collection('plants')->shardSnapshots(0)->delete($snapshot->name);
Qdrant::collection('plants')->shardSnapshots(0)->download($snapshot->name);
Qdrant::collection('plants')->shardSnapshots(0)->recover('file:///qdrant/snapshots/plants/0/shard.snapshot');

Cluster Management

Inspect and manage a distributed Qdrant cluster:

$status = Qdrant::collection('plants')->cluster()->status(); // ClusterStatus
Qdrant::collection('plants')->cluster()->telemetry();         // array
Qdrant::collection('plants')->cluster()->recover();           // bool
Qdrant::collection('plants')->cluster()->removePeer(peerId: 4, force: false);
Qdrant::collection('plants')->cluster()->collection();        // array — this collection's cluster info
Qdrant::collection('plants')->cluster()->moveShard(shardId: 0, fromPeer: 1, toPeer: 2);
Qdrant::collection('plants')->cluster()->replicateShard(shardId: 0, fromPeer: 1, toPeer: 2);

Shard Keys

Manage custom sharding for a collection:

Qdrant::collection('plants')->shards()->keys();          // array
Qdrant::collection('plants')->shards()->create('region-eu');
Qdrant::collection('plants')->shards()->delete('region-eu');

Issues (Beta)

Read and clear the server's self-diagnosed issues (a beta Qdrant API):

Qdrant::issues()->get();   // array
Qdrant::issues()->clear(); // bool

Artisan Commands

Creating a Collection with indexes

php artisan qdrant:migrate --collection=plants --vector-size=256 --distance-metric=euclidean --indexes='{"species":"text","age":"integer"}'

Rolling Back a Migration (Dropping Collection & Indexes)

php artisan qdrant:migrate --rollback --collection=plants

Extending with Macros

The query builder and client are Macroable, allowing custom methods:

use Mcpuishor\QdrantLaravel\QdrantClient;

QdrantClient::macro('byClimate', function ($climate) {
    return $this->where('climate', '=', $climate);
});

$results = Qdrant::collection('plants')->byClimate('tropical')->get();

Conclusion

This package simplifies working with Qdrant in Laravel, making it easy to integrate vector search and AI-powered applications. Contributions are welcome!

License

This package is open-source and available under the MIT License.