azure-ai-openai-dotnet

Azure.AI.OpenAI (.NET)

Client library for Azure OpenAI Service providing access to OpenAI models including GPT-4, GPT-4o, embeddings, DALL-E, and Whisper.

Installation

dotnet add package Azure.AI.OpenAI

# For OpenAI (non-Azure) compatibility
dotnet add package OpenAI

Historical example baseline: 2.1.0; verify and pin compatible Azure.AI.OpenAI/OpenAI versions. Newer v1 API examples below may require newer packages.

Environment Variables

AZURE_OPENAI_ENDPOINT=https://<resource-name>.openai.azure.com
AZURE_OPENAI_API_KEY=<api-key>                    # For key-based auth
AZURE_OPENAI_DEPLOYMENT_NAME=<deployment-name>    # Azure deployment, not necessarily the model ID

Client Hierarchy

AzureOpenAIClient (top-level)
├── GetChatClient(deploymentName)      → ChatClient
├── GetEmbeddingClient(deploymentName) → EmbeddingClient
├── GetImageClient(deploymentName)     → ImageClient
├── GetAudioClient(deploymentName)     → AudioClient
└── GetAssistantClient()               → AssistantClient

Authentication

API Key Authentication

using Azure;
using Azure.AI.OpenAI;

AzureOpenAIClient client = new(
    new Uri(Environment.GetEnvironmentVariable("AZURE_OPENAI_ENDPOINT")!),
    new AzureKeyCredential(Environment.GetEnvironmentVariable("AZURE_OPENAI_API_KEY")!));

Microsoft Entra ID (Recommended for Production)

using Azure.Identity;
using Azure.AI.OpenAI;

AzureOpenAIClient client = new(
    new Uri(Environment.GetEnvironmentVariable("AZURE_OPENAI_ENDPOINT")!),
    new DefaultAzureCredential());

Using OpenAI SDK Directly with Azure

using Azure.Identity;
using OpenAI;
using OpenAI.Chat;
using System.ClientModel.Primitives;

#pragma warning disable OPENAI001

BearerTokenPolicy tokenPolicy = new(
    new DefaultAzureCredential(),
    "https://cognitiveservices.azure.com/.default");

ChatClient client = new(
    model: "gpt-4o-mini",
    authenticationPolicy: tokenPolicy,
    options: new OpenAIClientOptions()
    {
        Endpoint = new Uri("https://YOUR-RESOURCE.openai.azure.com/openai/v1")
    });

Chat Completions

Basic Chat

using Azure.AI.OpenAI;
using OpenAI.Chat;

AzureOpenAIClient azureClient = new(
    new Uri(endpoint),
    new DefaultAzureCredential());

ChatClient chatClient = azureClient.GetChatClient("gpt-4o-mini");

ChatCompletion completion = chatClient.CompleteChat(
[
    new SystemChatMessage("You are a helpful assistant."),
    new UserChatMessage("What is Azure OpenAI?")
]);

Console.WriteLine(completion.Content[0].Text);

Async Chat

ChatCompletion completion = await chatClient.CompleteChatAsync(
[
    new SystemChatMessage("You are a helpful assistant."),
    new UserChatMessage("Explain cloud computing in simple terms.")
]);

Console.WriteLine($"Response: {completion.Content[0].Text}");
Console.WriteLine($"Tokens used: {completion.Usage.TotalTokenCount}");

Streaming Chat

await foreach (StreamingChatCompletionUpdate update 
    in chatClient.CompleteChatStreamingAsync(messages))
{
    if (update.ContentUpdate.Count > 0)
    {
        Console.Write(update.ContentUpdate[0].Text);
    }
}

Chat with Options

ChatCompletionOptions options = new()
{
    MaxOutputTokenCount = 1000,
    Temperature = 0.7f,
    TopP = 0.95f,
    FrequencyPenalty = 0,
    PresencePenalty = 0
};

ChatCompletion completion = await chatClient.CompleteChatAsync(messages, options);

Multi-turn Conversation

List<ChatMessage> messages = new()
{
    new SystemChatMessage("You are a helpful assistant."),
    new UserChatMessage("Hi, can you help me?"),
    new AssistantChatMessage("Of course! What do you need help with?"),
    new UserChatMessage("What's the capital of France?")
};

ChatCompletion completion = await chatClient.CompleteChatAsync(messages);
messages.Add(new AssistantChatMessage(completion.Content[0].Text));

Structured Outputs (JSON Schema)

using System.Text.Json;

ChatCompletionOptions options = new()
{
    ResponseFormat = ChatResponseFormat.CreateJsonSchemaFormat(
        jsonSchemaFormatName: "math_reasoning",
        jsonSchema: BinaryData.FromBytes("""
            {
                "type": "object",
                "properties": {
                    "steps": {
                        "type": "array",
                        "items": {
                            "type": "object",
                            "properties": {
                                "explanation": { "type": "string" },
                                "output": { "type": "string" }
                            },
                            "required": ["explanation", "output"],
                            "additionalProperties": false
                        }
                    },
                    "final_answer": { "type": "string" }
                },
                "required": ["steps", "final_answer"],
                "additionalProperties": false
            }
            """u8.ToArray()),
        jsonSchemaIsStrict: true)
};

ChatCompletion completion = await chatClient.CompleteChatAsync(
    [new UserChatMessage("How can I solve 8x + 7 = -23?")],
    options);

using JsonDocument json = JsonDocument.Parse(completion.Content[0].Text);
Console.WriteLine($"Answer: {json.RootElement.GetProperty("final_answer")}");

Reasoning Models (o1, o4-mini)

ChatCompletionOptions options = new()
{
    ReasoningEffortLevel = ChatReasoningEffortLevel.Low,
    MaxOutputTokenCount = 100000
};

ChatCompletion completion = await chatClient.CompleteChatAsync(
[
    new DeveloperChatMessage("You are a helpful assistant"),
    new UserChatMessage("Explain the theory of relativity")
], options);

Azure AI Search Integration (RAG)

using Azure.AI.OpenAI.Chat;

#pragma warning disable AOAI001

ChatCompletionOptions options = new();
options.AddDataSource(new AzureSearchChatDataSource()
{
    Endpoint = new Uri(searchEndpoint),
    IndexName = searchIndex,
    Authentication = DataSourceAuthentication.FromApiKey(searchKey)
});

ChatCompletion completion = await chatClient.CompleteChatAsync(
    [new UserChatMessage("What health plans are available?")],
    options);

ChatMessageContext context = completion.GetMessageContext();
if (context?.Intent is not null)
{
    Console.WriteLine($"Intent: {context.Intent}");
}
foreach (ChatCitation citation in context?.Citations ?? [])
{
    Console.WriteLine($"Citation: {citation.Content}");
}

Embeddings

using OpenAI.Embeddings;

EmbeddingClient embeddingClient = azureClient.GetEmbeddingClient("text-embedding-ada-002");

OpenAIEmbedding embedding = await embeddingClient.GenerateEmbeddingAsync("Hello, world!");
ReadOnlyMemory<float> vector = embedding.ToFloats();

Console.WriteLine($"Embedding dimensions: {vector.Length}");

Batch Embeddings

List<string> inputs = new()
{
    "First document text",
    "Second document text",
    "Third document text"
};

OpenAIEmbeddingCollection embeddings = await embeddingClient.GenerateEmbeddingsAsync(inputs);

foreach (OpenAIEmbedding emb in embeddings)
{
    Console.WriteLine($"Index {emb.Index}: {emb.ToFloats().Length} dimensions");
}

Image Generation (DALL-E)

using OpenAI.Images;

ImageClient imageClient = azureClient.GetImageClient("dall-e-3");

GeneratedImage image = await imageClient.GenerateImageAsync(
    "A futuristic city skyline at sunset",
    new ImageGenerationOptions
    {
        Size = GeneratedImageSize.W1024xH1024,
        Quality = GeneratedImageQuality.High,
        Style = GeneratedImageStyle.Vivid
    });

Console.WriteLine($"Image URL: {image.ImageUri}");

Audio (Whisper)

Transcription

using OpenAI.Audio;

AudioClient audioClient = azureClient.GetAudioClient("whisper");

AudioTranscription transcription = await audioClient.TranscribeAudioAsync(
    "audio.mp3",
    new AudioTranscriptionOptions
    {
        ResponseFormat = AudioTranscriptionFormat.Verbose,
        Language = "en"
    });

Console.WriteLine(transcription.Text);

Text-to-Speech

BinaryData speech = await audioClient.GenerateSpeechAsync(
    "Hello, welcome to Azure OpenAI!",
    GeneratedSpeechVoice.Alloy,
    new SpeechGenerationOptions
    {
        SpeedRatio = 1.0f,
        ResponseFormat = GeneratedSpeechFormat.Mp3
    });

await File.WriteAllBytesAsync("output.mp3", speech.ToArray());

Function Calling (Tools)

ChatTool getCurrentWeatherTool = ChatTool.CreateFunctionTool(
    functionName: "get_current_weather",
    functionDescription: "Get the current weather in a given location",
    functionParameters: BinaryData.FromString("""
        {
            "type": "object",
            "properties": {
                "location": {
                    "type": "string",
                    "description": "The city and state, e.g. San Francisco, CA"
                },
                "unit": {
                    "type": "string",
                    "enum": ["celsius", "fahrenheit"]
                }
            },
            "required": ["location"]
        }
        """));

ChatCompletionOptions options = new()
{
    Tools = { getCurrentWeatherTool }
};

ChatCompletion completion = await chatClient.CompleteChatAsync(
    [new UserChatMessage("What's the weather in Seattle?")],
    options);

if (completion.FinishReason == ChatFinishReason.ToolCalls)
{
    foreach (ChatToolCall toolCall in completion.ToolCalls)
    {
        Console.WriteLine($"Function: {toolCall.FunctionName}");
        Console.WriteLine($"Arguments: {toolCall.FunctionArguments}");
    }
}

Key Types Reference

Best Practices

1. Use Entra ID in production — Avoid API keys; use DefaultAzureCredential
2. Reuse client instances — Create once, share across requests
3. Handle rate limits — Implement exponential backoff for 429 errors
4. Stream for long responses — Use CompleteChatStreamingAsync for better UX
5. Set appropriate timeouts — Long completions may need extended timeouts
6. Use structured outputs — JSON schema ensures consistent response format
7. Monitor token usage — Track completion.Usage for cost management
8. Validate tool calls — Always validate function arguments before execution

Error Handling

using System.ClientModel;

try
{
    ChatCompletion completion = await chatClient.CompleteChatAsync(messages);
}
catch (System.ClientModel.ClientResultException ex) when (ex.Status == 429)
{
    Console.WriteLine("Rate limited. Retry after delay.");
    await Task.Delay(TimeSpan.FromSeconds(10));
}
catch (System.ClientModel.ClientResultException ex) when (ex.Status == 400)
{
    Console.WriteLine($"Bad request: {ex.Message}");
}
catch (System.ClientModel.ClientResultException ex)
{
    Console.WriteLine($"Azure OpenAI error: {ex.Status} - {ex.Message}");
}

Related SDKs

Reference Links

When to Use

This skill is applicable to execute the workflow or actions described in the overview.

Limitations

• Use this skill only when the task clearly matches the scope described above.
• Do not treat the output as a substitute for environment-specific validation, testing, or expert review.
• Stop and ask for clarification if required inputs, permissions, safety boundaries, or success criteria are missing.

Examples are separate fragments, not a compiled application; add required imports, variables and model-specific deployment names. Chat/audio/image features and options differ by deployment/API version: Whisper transcription and speech generation generally require different deployments. Check refusal, finish reason and content count before indexing; structured output does not guarantee successful non-refusal responses. Treat retrieved text/tool arguments as untrusted and authorize side effects independently. Bound retries and honor retry hints; sleeping once is not a retry loop. Do not log prompts, citations or raw service errors containing sensitive data. Model names, package versions and preview APIs here are historical illustrations, not current availability claims.