Vector search in the SQL Server EF Core Provider

Vector search

The SQL Server vector data type allows storing embeddings, which are representation of meaning that can be efficiently searched over for similarity, powering AI workloads such as semantic search and retrieval-augmented generation (RAG).

To use the vector data type, simply add a .NET property of type SqlVector<float> to your entity type, specifying the dimensions as follows:

public class Blog
{
    // ...

    [Column(TypeName = "vector(1536)")]
    public SqlVector<float> Embedding { get; set; }
}

public class Blog
{
    // ...

    public SqlVector<float> Embedding { get; set; }
}

protected override void OnModelCreating(ModelBuilder modelBuilder)
{
    modelBuilder.Entity<Blog>()
        .Property(b => b.Embedding)
        .HasColumnType("vector(1536)");
}

Once your property is added and the corresponding column created in the database, you can start inserting embeddings. Embedding generation is done outside of the database, usually via a service, and the details for doing this are out of scope for this documentation. However, the .NET Microsoft.Extensions.AI libraries contains IEmbeddingGenerator, which is an abstraction over embedding generators that has implementations for the major providers.

Once you've chosen your embedding generator and set it up, use it to generate embeddings and insert them as follows

IEmbeddingGenerator<string, Embedding<float>> embeddingGenerator = /* Set up your preferred embedding generator */;

var embedding = await embeddingGenerator.GenerateVectorAsync("Some text to be vectorized");
context.Blogs.Add(new Blog
{
    Name = "Some blog",
    Embedding = new SqlVector<float>(embedding)
});
await context.SaveChangesAsync();

Finally, use the EF.Functions.VectorDistance() function to perform similarity search for a given user query:

var sqlVector = new SqlVector<float>(await embeddingGenerator.GenerateVectorAsync("Some user query to be vectorized"));
var topSimilarBlogs = context.Blogs
    .OrderBy(b => EF.Functions.VectorDistance("cosine", b.Embedding, sqlVector))
    .Take(3)
    .ToListAsync();