Skip to content

LangChain.Databases

Database abstractions and vector store backends for the LangChain.NET ecosystem.

Philosophy

This project aligns with the Microsoft.Extensions ecosystem. For vector search, we use Microsoft.Extensions.VectorData.Abstractions (MEVA) — the industry-standard .NET vector data abstraction. For backends that have official Semantic Kernel connectors, consumers should reference those packages directly rather than using wrappers.

Packages

Vector stores (MEVA implementations)

These packages implement VectorStore / VectorStoreCollection<TKey, TRecord> from Microsoft.Extensions.VectorData:

Package Backend NuGet
LangChain.Databases.Postgres PostgreSQL + pgvector NuGet
LangChain.Databases.OpenSearch OpenSearch k-NN NuGet

Chat message history

Package Backend NuGet
LangChain.Databases.Abstractions In-memory, file-based NuGet
LangChain.Databases.Redis Redis NuGet
LangChain.Databases.Mongo MongoDB NuGet

Use SK connectors for other backends

For backends with official Semantic Kernel connectors, use those packages directly:

Backend SK Connector Package
In-Memory Microsoft.SemanticKernel.Connectors.InMemory
SQLite Microsoft.SemanticKernel.Connectors.Sqlite
Chroma Microsoft.SemanticKernel.Connectors.Chroma
Qdrant Microsoft.SemanticKernel.Connectors.Qdrant
Pinecone Microsoft.SemanticKernel.Connectors.Pinecone
Weaviate Microsoft.SemanticKernel.Connectors.Weaviate
Milvus Microsoft.SemanticKernel.Connectors.Milvus
DuckDB Microsoft.SemanticKernel.Connectors.DuckDB
Azure AI Search Microsoft.SemanticKernel.Connectors.AzureAISearch
Redis (vector) Microsoft.SemanticKernel.Connectors.Redis
MongoDB (vector) Microsoft.SemanticKernel.Connectors.MongoDB

Quick start

1. Define a record type

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
using Microsoft.Extensions.VectorData;

public class DocumentRecord
{
    [VectorStoreKey]
    public string Id { get; set; } = Guid.NewGuid().ToString();

    [VectorStoreData]
    public string Text { get; set; } = string.Empty;

    [VectorStoreVector(1536)]
    public ReadOnlyMemory<float> Embedding { get; set; }
}

2. Create a store and collection

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
// PostgreSQL + pgvector
var store = new PostgresVectorStore(connectionString);

// Or OpenSearch
var store = new OpenSearchVectorStore(new OpenSearchVectorDatabaseOptions
{
    ConnectionUri = new Uri("http://localhost:9200"),
});

// Or SK InMemory
var store = new InMemoryVectorStore();

var collection = store.GetCollection<string, DocumentRecord>("my_collection");
await collection.EnsureCollectionExistsAsync();

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
await collection.UpsertAsync(new DocumentRecord
{
    Text = "hello world",
    Embedding = embedding,
});

await foreach (var result in collection.SearchAsync(
    new ReadOnlyMemory<float>(queryEmbedding), top: 5))
{
    Console.WriteLine($"{result.Record.Text} (score: {result.Score})");
}

Migration from IVectorDatabase

The legacy IVectorDatabase / IVectorCollection interfaces have been removed in favor of MEVA. See the migration guide for detailed before/after examples.

License

This project is licensed under the MIT License.