What are the AI and Vector Data Extensions<\/h2>\n
![\"Diagram](\"https:\/\/devblogs.microsoft.com\/dotnet\/wp-content\/uploads\/sites\/10\/2025\/05\/extensions-layers.png\")
<\/p>\n
The AI and Vector Data Extensions<\/strong> are a set of .NET libraries that offer shared abstractions and utilities for working with AI models and vector stores.<\/p>\n These libraries are available as NuGet packages:<\/p>\n These packages serve as foundational building blocks for higher-level components, promoting:<\/p>\n If you’re building a library<\/strong>, it\u2019s critical to remain agnostic to specific AI or vector systems. By depending only on these shared abstractions, you avoid locking your consumers into particular providers and ensure your library can interoperate with others. This promotes flexibility and broad compatibility across the ecosystem.<\/p>\n If you\u2019re building an application<\/strong>, you have more freedom to choose concrete implementations.<\/p>\n What does this mean in practice?<\/p>\n The AI and Vector Data Extensions provide the essential building blocks that make it easier to implement advanced AI capabilities in your applications. By offering consistent abstractions for features like structured output, tool invocation, and observability, these libraries enable you to build robust, maintainable, and production-ready solutions tailored to your specific needs.<\/p>\n Below are examples of common scenarios where these building blocks come together to empower real-world AI-powered applications.<\/p>\n Whether you’re using different model providers for local development and production, or building agents that rely on various models, the AI and Vector Data extensions offer a consistent set of APIs across environments.<\/p>\n Thanks to a growing ecosystem of official and community-supported packages that implement these abstractions, it’s easy to integrate models and vector databases into your applications.<\/p>\n Below is an example of how you can switch between providers based on the environment, while keeping your code clean and consistent:<\/p>\n Using AI models is just the beginning. Building production-grade applications requires logging, caching, and observability through tools like OpenTelemetry.<\/p>\n The AI extensions support these needs out of the box. They integrate seamlessly with existing .NET primitives, allowing you to plug in your own Here\u2019s a simple example of how to enable these features:<\/p>\n Need more control? The extensions are fully extensible. You can inject custom logic\u2014like rate limiting\u2014directly into the pipeline.<\/p>\n Generative AI models can process more than just text\u2014they\u2019re also capable of handling images, audio, and other data types.<\/p>\n To support this, the AI extensions provide flexible primitives for representing diverse data formats.<\/p>\n One common challenge is that model outputs are often unstructured, making integration with your application more difficult.<\/p>\n Fortunately, many models now support structured output\u2014a feature where responses are formatted according to a predefined schema, such as JSON. This adds reliability and predictability to the model\u2019s responses, simplifying integration.<\/p>\n The AI extensions are designed to work seamlessly with structured outputs, making it easy to map model responses directly to your C# types.<\/p>\n AI models can process data and understand natural language, but they can’t perform actions on their own. To take meaningful action, they need access to external tools and systems.<\/p>\n This is where tool calling comes in\u2014a feature supported by many modern generative AI models that allows them to invoke functions based on user intent.<\/p>\n Similar to structured output, the AI extensions make it easy to take advantage of this capability in your applications.<\/p>\n In this example, the Semantic search\u2014and by extension, Retrieval-Augmented Generation (RAG)\u2014is a powerful pattern for building intelligent applications. While the core tasks, like embedding generation, are conceptually straightforward, they often involve repetitive, boilerplate code.<\/p>\n At the heart of this pattern are vector databases, which store numerical representations of data (called embeddings) and enable fast similarity search. They let applications retrieve semantically relevant information\u2014not just exact keyword matches\u2014making them essential for tasks like search, question answering, and recommendations.<\/p>\n The AI and Vector Data extensions are designed to streamline this process by working seamlessly together.<\/p>\n When you combine a vector store with an embedding generator, the extensions make it easy to map your existing C# data models directly to the vector store. These abstractions eliminate much of the boilerplate, so you can focus on your application logic\u2014not infrastructure details.<\/p>\n And because these are just abstractions, they\u2019re inherently modular. You can swap out vector store or embedding providers without changing your application logic\u2014making it easy to adapt to new tools or scale as your needs evolve.<\/p>\n In the example below, we provide plain text. The extensions handle the rest\u2014automatically generating the embedding using the configured generator and storing it in the appropriate vector field, all with minimal code.<\/p>\n Depending on your scenario and data model, you may need more advanced search capabilities. The Vector Data abstractions support a range of powerful search features\u2014including multiple similarity metrics, vector search, hybrid search, and filtering.<\/p>\n Just like with embedding generation, the querying process is streamlined. You can pass in plain text, and the abstractions handle generating the embedding, applying the appropriate similarity metric, and retrieving the most relevant results.<\/p>\n Filtering is also built-in and designed to feel familiar. The filter expressions use a syntax similar to LINQ predicates, so you can leverage your existing C# skills without needing to learn a new query language.<\/p>\n In the example below, we’re searching for products that match a natural language query, filtered by tenant:<\/p>\n Modern .NET applications rely on dependency injection (DI) to manage configuration and lifetimes of services\u2014and the AI and Vector Data extensions are built to align with that model.<\/p>\n Whether you’re wiring up components for local development or configuring services for production, the extensions register cleanly into your existing DI container. This means you can compose and configure AI components just like any other part of your application.<\/p>\n Adoption of these extensions has been strong and continues to grow across the ecosystem.<\/p>\n In just a few months, the extensions have surpassed 3 million downloads<\/strong>, with nearly 100 public NuGet packages<\/strong> combined taking a dependency on them.<\/p>\n We’re seeing adoption across a wide range of official and community projects, including:<\/p>\n Below, we highlight a few of these integrations in more detail.<\/p>\n MCP is an open standard that acts as a universal adapter for AI models. It enables models to interact with external data sources, tools, and APIs through a consistent, standardized interface. This simplifies integration by allowing models to invoke functions or access data without needing custom code for each service.<\/p>\n We partnered with Anthropic to deliver an official MCP C# SDK<\/a>. Built on top of shared AI abstractions like Evaluations play a crucial role in building trustworthy AI applications by helping ensure safety, reliability, and alignment with intended behavior. They allow developers to systematically test and validate AI models against real-world scenarios and quality standards.<\/p>\n The .NET AI Evaluation set of libraries builds on top of the AI Extenions to create powerful evaluation tools that integrate seamlessly into your development workflow, enabling continuous monitoring and improvement of your AI systems.<\/p>\n To dive deeper into these evaluation capabilities and see practical examples, check out the following blog posts.<\/p>\n Since the launch of ChatGPT, chat has become the primary way users interact with language models.<\/p>\n To offer a similar experience in their own applications, developers have often had to build custom chat UI components from scratch.<\/p>\n Telerik has made this much easier by providing a set of ready-to-use chat UI components for Blazor. These components simplify the process of adding conversational interfaces to web apps.<\/p>\n Built on top of the AI Extensions, Telerik\u2019s solution also allows developers to switch between different model providers with minimal code changes\u2014making it both flexible and future-proof.<\/p>\n\n
\n
Why target these abstractions?<\/h3>\n
\n
Key Scenarios and Use Cases<\/h2>\n
Portability across model and vector store providers<\/h3>\n
IChatClient chatClient = \n environment == \"Development\"\n ? new OllamaApiClient(\"YOUR-OLLAMA-ENDPOINT\", \"qwen3\")\n : new AzureOpenAIClient(\"YOUR-AZURE-OPENAI-ENDPOINT\", new DefaultAzureCredential())\n .GetChatClient(\"gpt-4.1\")\n .AsIChatClient();\n\nawait foreach (var message in chatClient.GetStreamingResponseAsync(\"What is AI?\"))\n{\n Console.Write($\"{message.Text}\");\n};\n\nIEmbeddingGenerator<string, Embedding<float>> embeddingGenerator = \n environment == \"Development\"\n ? new OllamaApiClient(\"YOUR-OLLAMA-ENDPOINT\", \"all-minilm\")\n : new AzureOpenAIClient(\"YOUR-AZURE-OPENAI-ENDPOINT\", new DefaultAzureCredential())\n .GetEmbeddingClient(\"text-embedding-3-small\")\n .AsIEmbeddingGenerator();\n\nvar embedding = await embeddingGenerator.GenerateAsync(\"What is AI?\");\n\nVectorStoreCollection<int, Product> collection =\n environment == \"Development\"\n ? new SqliteCollection<int, Product>(\n \"Data Source=products.db\",\n \"products\", \n new SqliteCollectionOptions { EmbeddingGenerator = embeddingGenerator})\n : new QdrantCollection<int, Product>(\n new QdrantClient(\"YOUR-HOSTED-ENDPOINT\"), \n \"products\", \n true, \n new QdrantCollectionOptions { EmbeddingGenerator = embeddingGenerator});\n\nawait collection.UpsertAsync(...);<\/code><\/pre>\nProgressively add functionality<\/h3>\n
ILogger<\/code>, IDistributedCache<\/code>, and OpenTelemetry-compliant tools without reinventing the wheel.<\/p>\nIChatClient chatClient = \n new ChatClientBuilder(...)\n .UseLogging()\n .UseDistributedCache()\n .UseOpenTelemetry()\n .Build();<\/code><\/pre>\nRateLimiter rateLimiter = new ConcurrencyLimiter(new()\n{\n PermitLimit = 1,\n QueueLimit = int.MaxValue\n});\n\nIChatClient client = \n new ChatClientBuilder(...)\n .UseDistributedCache()\n .Use(async (messages, options, nextAsync, cancellationToken) =>\n {\n using var lease = await rateLimiter.AcquireAsync(permitCount: 1, cancellationToken).ConfigureAwait(false);\n if (!lease.IsAcquired)\n throw new InvalidOperationException(\"Unable to acquire lease.\");\n\n await nextAsync(messages, options, cancellationToken);\n })\n .UseOpenTelemetry()\n .Build();<\/code><\/pre>\nHandle different content and structure their output<\/h3>\n
record Item(string Name, float Price);\nenum Category { Food, Electronics, Clothing, Services };\nrecord Receipt(string Merchant, List<Item> Items, float Total, Category Category);\n\nvar imageUri = new Uri(\"https:\/\/host\/someimage.jpg\");\n\nList<AIContent> content = [\n new TextContent(\"Process this receipt\"),\n new UriContent(new Uri(imageUri), mediaType: \"image\/jpeg\")\n];\n\nvar message = new ChatMessage(ChatRole.User, content);\nvar response = await chatClient.GetResponseAsync<Receipt>(message);\n\nresponse.TryGetResult(out var receiptData);\n\nConsole.WriteLine($\"Merchant: {receiptData.Merchant} | Total: {receiptData.Total} | Category: {receiptData.Category}\");\n\/\/Merchant: ECOSPACE | Total: 49.64 | Category: Food<\/code><\/pre>\nTool Calling<\/h3>\n
CalculateTax<\/code> method is registered as an AI-invokable function. The model can automatically decide when to call it based on the user\u2019s request. If you need more control over this behavior, it can be easily configured.<\/p>\nrecord ReceiptTotal(float SubTotal, float TaxAmount, float TaxRate, float Total);\n[Description(\"Calculate tax given a receipt and tax rate\")]\nfloat CalculateTax(Receipt receipt, float taxRate)\n{\n return receipt.Total * (1 + taxRate);\n}\n\nIChatClient functionChatClient = \n chatClient\n .AsBuilder()\n .UseFunctionInvocation()\n .Build();\n\nvar message = new ChatMessage(ChatRole.User, [\n new TextContent(\"Here is information from a recent purchase\"),\n new TextContent($\"{JsonSerializer.Serialize(receiptData)}\"),\n new TextContent(\"What is the total price after tax given a tax rate of 10%?\")\n]);\n\nvar response = await functionChatClient.GetResponseAsync<ReceiptTotal>(message, new ChatOptions {Tools = [AIFunctionFactory.Create(CalculateTax)]});\n\nresponse.TryGetResult(out var receiptTotal);\nConsole.WriteLine($\"SubTotal: {receiptTotal.SubTotal} | TaxAmount: {receiptTotal.TaxAmount} | TaxRate: {receiptTotal.TaxRate} | Total: {receiptTotal.Total}\");\n\/\/ SubTotal: 49.64 | TaxAmount: 4.96 | TaxRate: 0.1 | Total: 54.6<\/code><\/pre>\nSimplified embedding generation<\/h3>\n
\/\/ Using Microsoft.SemanticKernel.Connectors.SqliteVec package\nEmbeddingGenerator<string, Embedding<float>> embeddingGenerator = ...;\nVectorStoreCollection<int, Product> collection = new SqliteCollection<int,Product>(\"Data Source=products.db\", \"products\", new SqliteCollectionOptions {\n EmbeddingGenerator = embeddingGenerator,\n});\n\nawait collection.UpsertAsync(new Product\n{\n Id = 1,\n Name = \"Kettle\",\n TenantId = 5,\n Embedding = \"This kettle is great for making tea, it heats up quickly and has a large capacity.\"\n});\n\nrecord Product\n{\n [VectorStoreKey]\n public int Id { get; set; }\n\n [VectorStoreData]\n public required string Name { get; set; }\n\n [VectorStoreData]\n public int TenantId { get; set; }\n\n [VectorStoreVector(Dimensions: 1536)]\n public string? Embedding { get; set; }\n}<\/code><\/pre>\nPowerful search capabilities<\/h3>\n
EmbeddingGenerator<string, Embedding<float>> embeddingGenerator = ...;\nVectorStoreCollection<int, Product> collection = new SqliteCollection<int,Product>(\"Data Source=products.db\", \"products\", new SqliteCollectionOptions {\n EmbeddingGenerator = embeddingGenerator,\n});\n\nvar query = \"Find me kettles that can hold a lot of water\";\n\nawait foreach (var result in collection.SearchAsync(query, top: 5, new() { Filter = r => r.TenantId == 8 }))\n{\n yield return result.Record;\n}<\/code><\/pre>\nPlugs right into your existing dependency injection configurations<\/h3>\n
builder.Services.AddChatClient(sp => {...})\n .UseLogging()\n .UseCaching()\n .UseOpenTelemetry();\n\nbuilder.Services.AddEmbeddingGenerator(sp => {...})\n .UseLogging()\n .UseCaching()\n .UseOpenTelemetry();\n\n\/\/ SQLite implementation of Vector Data\nbuilder.Services.AddSqliteCollection<int, Product>(\"Products\", \"Data Source=\/tmp\/products.db\");<\/code><\/pre>\nA growing ecosystem<\/h2>\n
\n
Model Context Protocol (MCP) C# SDK<\/h3>\n
AIContent<\/code>, AIFunction<\/code>, and others, the SDK enables MCP clients and servers to easily define tools and invoke them using IChatClient<\/code> implementations.<\/p>\nvar mcpClient = await McpClientFactory.CreateAsync(clientTransport, mcpClientOptions, loggerFactory);\n\nvar tools = await mcpClient.ListToolsAsync();\n\nvar response = await _chatClient.GetResponseAsync<List<TripOption>>(messages, new ChatOptions { Tools = [.. tools ] });<\/code><\/pre>\nEvaluations<\/h3>\n
![\"Ai](\"https:\/\/devblogs.microsoft.com\/dotnet\/wp-content\/uploads\/sites\/10\/2025\/05\/aivectorreport.png\")
<\/p>\n\n
Progress Telerik<\/h3>\n