Manage data

Once you've defined your data model and obtained a VectorStoreCollection<TKey,TRecord>, you can manage records in your vector store collection. This article covers the core data management operations: upserting, deleting, and retrieving records. Searching your data using vector similarity is covered in Vector search.

All the examples below use the following data model, which uses automatic embedding generation:

public class Hotel
{
    [VectorStoreKey]
    public ulong HotelId { get; set; }

    [VectorStoreData(IsIndexed = true)]
    public required string HotelName { get; set; }

    [VectorStoreData(IsFullTextIndexed = true)]
    public required string Description { get; set; }

    [VectorStoreVector(Dimensions: 1536, DistanceFunction = DistanceFunction.CosineSimilarity)]
    public required string DescriptionEmbedding { get; set; }

    [VectorStoreData(IsIndexed = true)]
    public required string[] Tags { get; set; }
}

Note that the DescriptionEmbedding property is a string rather than a vector type such as ReadOnlyMemory<float>. This means MEVD automatically generates embeddings when upserting records, using the IEmbeddingGenerator<TInput,TEmbedding> configured on the vector store.

The following setup code gets a collection and ensures it exists:

var vectorStore = new QdrantVectorStore(
    new QdrantClient("localhost"),
    ownsClient: true,
    new QdrantVectorStoreOptions
    {
        EmbeddingGenerator = embeddingGenerator
    });

var collection = vectorStore.GetCollection<ulong, Hotel>("hotels");
await collection.EnsureCollectionExistsAsync();

Upsert records

Use UpsertAsync to insert a new record or update an existing one. If a record with the same key already exists, it's replaced; otherwise, a new record is created:

var hotel = new Hotel
{
    HotelId = 1,
    HotelName = "Hotel Happy",
    Description = "A happy hotel with good vibes.",
    DescriptionEmbedding = "A happy hotel with good vibes.",
    Tags = ["happy", "vibes"]
};

await collection.UpsertAsync(hotel);

To upsert multiple records at once, pass a collection:

var hotels = new[]
{
    new Hotel
    {
        HotelId = 1,
        HotelName = "Hotel Happy",
        Description = "A happy hotel with good vibes.",
        DescriptionEmbedding = "A happy hotel with good vibes.",
        Tags = ["happy", "vibes"]
    },
    new Hotel
    {
        HotelId = 2,
        HotelName = "Hotel Luxury",
        Description = "A luxurious hotel with top-notch amenities.",
        DescriptionEmbedding = "A luxurious hotel with top-notch amenities.",
        Tags = ["luxury", "amenities"]
    }
};

await collection.UpsertAsync(hotels);

Tip

Prefer the batch overload when upserting multiple records, as this reduces database roundtrips. Also, with automatic embedding generation, the batch overload usually generates embeddings for all records in a single call to the embedding model, which is more efficient than generating them one at a time.

Delete records

Use DeleteAsync to remove records from the collection by key:

await collection.DeleteAsync(key: 1);

To delete multiple records at once, pass a collection:

await collection.DeleteAsync(keys: [1, 2, 3]);

Note

If a record with the specified key doesn't exist, the delete operation is silently ignored.

Retrieve records by key

Use GetAsync to retrieve records by their key:

Hotel? hotel = await collection.GetAsync(key: 1);

GetAsync returns the record if found, or null if no record with the given key exists.

To retrieve multiple records at once, pass a collection of keys. This overload returns an IAsyncEnumerable<TRecord> containing only the records that were found:

IAsyncEnumerable<Hotel> hotels = collection.GetAsync(keys: [1, 2, 3]);

await foreach (Hotel hotel in hotels)
{
    Console.WriteLine(hotel.HotelName);
}

Note

Only found records are returned, so the result set might be smaller than the set of requested keys. There is no guarantee on the ordering of the returned records.

Include vectors in results

By default, vector properties aren't included in the retrieved records, which reduces data transfer. To include them, pass a RecordRetrievalOptions with IncludeVectors set to true:

var hotel = await collection.GetAsync(
    key: 1,
    new() { IncludeVectors = true });

Retrieve records by filter

Use the filter overload of GetAsync to retrieve records that match a given condition, rather than by key or vector similarity:

IAsyncEnumerable<Hotel> hotels = collection.GetAsync(
    filter: h => h.HotelName == "Hotel Happy",
    top: 10);

await foreach (Hotel hotel in hotels)
{
    Console.WriteLine(hotel.HotelName);
}

Filters are expressed as LINQ expressions. The supported expressions vary depending on the database provider, but all providers support common comparisons such as equality, inequality, and logical && and ||.

Important

For properties to be usable in filters, many databases require them to be indexed. Even where indexes aren't required, they can greatly increase the performance of filters. Set IsIndexed = true on the VectorStoreDataAttribute when defining your data model. For more information, see Data property.

Filter options

Use FilteredRecordRetrievalOptions<TRecord> to control paging, ordering, and whether vectors are included in the results:

IAsyncEnumerable<Hotel> hotels = collection.GetAsync(
    filter: h => h.Tags.Contains("luxury"),
    top: 10,
    new()
    {
        Skip = 20,
        IncludeVectors = false,
        OrderBy = order => order.Ascending(h => h.HotelName)
    });

The following table describes the available options:

Option	Description
Skip	Number of matching records to skip before returning results. Useful for paging. Default is 0.
OrderBy	Specifies the property and direction to order results by. Use .Ascending() or .Descending() to define the sort. If not provided, the order is non-deterministic.
IncludeVectors	Whether to include vector data in the returned records. Default is false.