Integrate Azure OpenAI Models into .NET Apps

This skill covers Azure.AI.OpenAI, the .NET client library for Azure OpenAI Service (v2.1.0 stable), providing access to GPT-4, GPT-4o, embeddings, DALL-E, and Whisper models. A top-level AzureOpenAIClient exposes GetChatClient, GetEmbeddingClient, GetImageClient, GetAudioClient, and GetAssistantClient, each scoped to a specific deployment name. Two authentication paths are documented: API-key auth via AzureKeyCredential, and Microsoft Entra ID via DefaultAzureCredential (recommended for production); the OpenAI SDK itself can also be pointed at Azure directly using a BearerTokenPolicy scoped to the cognitiveservices.azure.com resource. Chat completions cover basic synchronous calls, async calls with token-usage reporting, streaming responses via CompleteChatStreamingAsync, tunable options (MaxOutputTokenCount, Temperature, TopP, frequency/presence penalties), multi-turn conversations built by appending each assistant reply back into the message list, structured JSON-Schema-constrained outputs via ChatResponseFormat.CreateJsonSchemaFormat with a strict schema, reasoning-model-specific options (ReasoningEffortLevel, larger MaxOutputTokenCount, DeveloperChatMessage) for o1/o4-mini-class models, retrieval-augmented generation by attaching an AzureSearchChatDataSource to chat options and reading back intent and citations from the response's message context, and function/tool calling by defining a ChatTool with a JSON-Schema parameter definition and inspecting ToolCalls when FinishReason is ToolCalls. Beyond chat, it covers single and batch embedding generation (GenerateEmbeddingAsync / GenerateEmbeddingsAsync, reading back float vectors), DALL-E image generation with size/quality/style options, Whisper audio transcription with a verbose response format and language hint, and text-to-speech generation with a selectable voice and speed.

Use it for any .NET application calling Azure-hosted OpenAI models - chat completions (including streaming, RAG, and structured outputs), embeddings, image generation, or audio transcription/synthesis - through Azure OpenAI Service specifically rather than OpenAI's own API directly. Choose Microsoft Entra ID authentication over API keys for production deployments.

Given a deployment name and a list of ChatMessage objects, chat clients return a ChatCompletion with content text, token usage, tool calls, or RAG citations, or a stream of StreamingChatCompletionUpdate objects. Embedding, image, and audio clients take text/image prompts or audio files and return float vectors, image URIs, or transcribed/synthesized audio respectively.

dotnet add package Azure.AI.OpenAI
dotnet add package OpenAI

The OpenAI package is an optional add-on for full OpenAI-SDK compatibility against an Azure endpoint. Authenticates via AzureKeyCredential or DefaultAzureCredential against AZURE_OPENAI_ENDPOINT, with AZURE_OPENAI_API_KEY and AZURE_OPENAI_DEPLOYMENT_NAME as supporting environment variables, and can integrate with Azure AI Search for retrieval-augmented chat.

.NET developers building applications on Azure OpenAI Service who need chat, embeddings, image, or audio model access with production-grade authentication, streaming, RAG via Azure AI Search, function calling, structured JSON output, and support for newer reasoning models like o1 and o4-mini.