Azure OpenAI client library for .NET

The Azure OpenAI client library for .NET is an adaptation of OpenAI's REST APIs that provides an idiomatic interface and rich integration with the rest of the Azure SDK ecosystem. It can connect to Azure OpenAI resources or to the non-Azure OpenAI inference endpoint, making it a great choice for even non-Azure OpenAI development.

Use the client library for Azure OpenAI to:

• Create a completion for text
• Create a text embedding for comparisons

Azure OpenAI is a managed service that allows developers to deploy, tune, and generate content from OpenAI models on Azure resources.

Source code | Package (NuGet) | API reference documentation | Product documentation | Samples

Getting started

Prerequisites

If you'd like to use an Azure OpenAI resource, you must have an Azure subscription and Azure OpenAI access. This will allow you to create an Azure OpenAI resource and get both a connection URL as well as API keys. For more information, see Quickstart: Get started generating text using Azure OpenAI Service.

If you'd like to use the Azure OpenAI .NET client library to connect to non-Azure OpenAI, you'll need an API key from a developer account at https://platform.openai.com/.

Install the package

Install the client library for .NET with NuGet:

dotnet add package Azure.AI.OpenAI --prerelease

Authenticate the client

In order to interact with Azure OpenAI or OpenAI, you'll need to create an instance of the OpenAIClient class. To configure a client for use with Azure OpenAI, provide a valid endpoint URI to an Azure OpenAI resource along with a corresponding key credential, token credential, or Azure identity credential that's authorized to use the Azure OpenAI resource. To instead configure the client to connect to OpenAI's service, provide an API key from OpenAI's developer portal.

OpenAIClient client = useAzureOpenAI
    ? new OpenAIClient(
        new Uri("https://your-azure-openai-resource.com/"),
        new AzureKeyCredential("your-azure-openai-resource-api-key"))
    : new OpenAIClient("your-api-key-from-platform.openai.com");

Create OpenAIClient with an Azure Active Directory Credential

Client subscription key authentication is used in most of the examples in this getting started guide, but you can also authenticate with Azure Active Directory using the Azure Identity library. To use the DefaultAzureCredential provider shown below, or other credential providers provided with the Azure SDK, please install the Azure.Identity package:

dotnet add package Azure.Identity

string endpoint = "https://myaccount.openai.azure.com/";
var client = new OpenAIClient(new Uri(endpoint), new DefaultAzureCredential());

Key concepts

The main concept to understand is Completions. Briefly explained, completions provides its functionality in the form of a text prompt, which by using a specific model, will then attempt to match the context and patterns, providing an output text. The following code snippet provides a rough overview (more details can be found in the GenerateChatbotResponsesWithToken sample code):

OpenAIClient client = useAzureOpenAI
    ? new OpenAIClient(
        new Uri("https://your-azure-openai-resource.com/"),
        new AzureKeyCredential("your-azure-openai-resource-api-key"))
    : new OpenAIClient("your-api-key-from-platform.openai.com");

Response<Completions> response = await client.GetCompletionsAsync(
    "text-davinci-003", // assumes a matching model deployment or model name
    "Hello, world!");

foreach (Choice choice in response.Value.Choices)
{
    Console.WriteLine(choice.Text);
}

Thread safety

We guarantee that all client instance methods are thread-safe and independent of each other (guideline). This ensures that the recommendation of reusing client instances is always safe, even across threads.

Additional concepts

Client options | Accessing the response | Long-running operations | Handling failures | Diagnostics | Mocking | Client lifetime

Examples

You can familiarize yourself with different APIs using Samples.

Generate Chatbot Response

The GenerateChatbotResponse method authenticates using a DefaultAzureCredential, then generates text responses to input prompts.

string endpoint = "https://myaccount.openai.azure.com/";
var client = new OpenAIClient(new Uri(endpoint), new DefaultAzureCredential());

string deploymentName = "text-davinci-003";
string prompt = "What is Azure OpenAI?";
Console.Write($"Input: {prompt}");

Response<Completions> completionsResponse = client.GetCompletions(deploymentName, prompt);
string completion = completionsResponse.Value.Choices[0].Text;
Console.WriteLine($"Chatbot: {completion}");

Generate Multiple Chatbot Responses With Subscription Key

The GenerateMultipleChatbotResponsesWithSubscriptionKey method gives an example of generating text responses to input prompts using an Azure subscription key

// Replace with your Azure OpenAI key
string key = "YOUR_AZURE_OPENAI_KEY";
string endpoint = "https://myaccount.openai.azure.com/";
var client = new OpenAIClient(new Uri(endpoint), new AzureKeyCredential(key));

List<string> examplePrompts = new(){
    "How are you today?",
    "What is Azure OpenAI?",
    "Why do children love dinosaurs?",
    "Generate a proof of Euler's identity",
    "Describe in single words only the good things that come into your mind about your mother.",
};

string deploymentName = "text-davinci-003";

foreach (string prompt in examplePrompts)
{
    Console.Write($"Input: {prompt}");
    CompletionsOptions completionsOptions = new CompletionsOptions();
    completionsOptions.Prompts.Add(prompt);

    Response<Completions> completionsResponse = client.GetCompletions(deploymentName, completionsOptions);
    string completion = completionsResponse.Value.Choices[0].Text;
    Console.WriteLine($"Chatbot: {completion}");
}

Summarize Text with Completion

The SummarizeText method generates a summarization of the given input prompt.

string endpoint = "https://myaccount.openai.azure.com/";
var client = new OpenAIClient(new Uri(endpoint), new DefaultAzureCredential());

string textToSummarize = @"
    Two independent experiments reported their results this morning at CERN, Europe's high-energy physics laboratory near Geneva in Switzerland. Both show convincing evidence of a new boson particle weighing around 125 gigaelectronvolts, which so far fits predictions of the Higgs previously made by theoretical physicists.

    ""As a layman I would say: 'I think we have it'. Would you agree?"" Rolf-Dieter Heuer, CERN's director-general, asked the packed auditorium. The physicists assembled there burst into applause.
:";

string summarizationPrompt = @$"
    Summarize the following text.

    Text:
    """"""
    {textToSummarize}
    """"""

    Summary:
";

Console.Write($"Input: {summarizationPrompt}");
var completionsOptions = new CompletionsOptions()
{
    Prompts = { summarizationPrompt },
};

string deploymentName = "text-davinci-003";

Response<Completions> completionsResponse = client.GetCompletions(deploymentName, completionsOptions);
string completion = completionsResponse.Value.Choices[0].Text;
Console.WriteLine($"Summarization: {completion}");

Stream Chat Messages with non-Azure OpenAI

string nonAzureOpenAIApiKey = "your-api-key-from-platform.openai.com";
var client = new OpenAIClient(nonAzureOpenAIApiKey, new OpenAIClientOptions());
var chatCompletionsOptions = new ChatCompletionsOptions()
{
    Messages =
    {
        new ChatMessage(ChatRole.System, "You are a helpful assistant. You will talk like a pirate."),
        new ChatMessage(ChatRole.User, "Can you help me?"),
        new ChatMessage(ChatRole.Assistant, "Arrrr! Of course, me hearty! What can I do for ye?"),
        new ChatMessage(ChatRole.User, "What's the best way to train a parrot?"),
    }
};

Response<StreamingChatCompletions> response = await client.GetChatCompletionsStreamingAsync(
    deploymentOrModelName: "gpt-3.5-turbo",
    chatCompletionsOptions);
using StreamingChatCompletions streamingChatCompletions = response.Value;

await foreach (StreamingChatChoice choice in streamingChatCompletions.GetChoicesStreaming())
{
    await foreach (ChatMessage message in choice.GetMessageStreaming())
    {
        Console.Write(message.Content);
    }
    Console.WriteLine();
}

Use Chat Functions

Chat Functions allow a caller of Chat Completions to define capabilities that the model can use to extend its functionality into external tools and data sources.

You can read more about Chat Functions on OpenAI's blog: https://openai.com/blog/function-calling-and-other-api-updates

NOTE: Chat Functions require model versions beginning with gpt-4 and gpt-3.5-turbo's -0613 labels. They are not available with older versions of the models.

To use Chat Functions, you first define the function you'd like the model to be able to use when appropriate. Using the example from the linked blog post, above:

var getWeatherFuntionDefinition = new FunctionDefinition()
{
    Name = "get_current_weather",
    Description = "Get the current weather in a given location",
    Parameters = BinaryData.FromObjectAsJson(
    new
    {
        Type = "object",
        Properties = new
        {
            Location = new
            {
                Type = "string",
                Description = "The city and state, e.g. San Francisco, CA",
            },
            Unit = new
            {
                Type = "string",
                Enum = new[] { "celsius", "fahrenheit" },
            }
        },
        Required = new[] { "location" },
    },
    new JsonSerializerOptions() {  PropertyNamingPolicy = JsonNamingPolicy.CamelCase }),
};

With the function defined, it can then be used in a Chat Completions request via its options. Function data is handled across multiple calls that build up data for subsequent stateless requests, so we maintain a list of chat messages as a form of conversation history.

var conversationMessages = new List<ChatMessage>()
{
    new(ChatRole.User, "What is the weather like in Boston?"),
};

var chatCompletionsOptions = new ChatCompletionsOptions();
foreach (ChatMessage chatMessage in conversationMessages)
{
    chatCompletionsOptions.Messages.Add(chatMessage);
}
chatCompletionsOptions.Functions.Add(getWeatherFuntionDefinition);

Response<ChatCompletions> response = await client.GetChatCompletionsAsync(
    "gpt-35-turbo-0613",
    chatCompletionsOptions);

If the model determines that it should call a Chat Function, a finish reason of 'FunctionCall' will be populated on the choice and details will be present in the response message's FunctionCall property. Usually, the name of the function call will be one that was provided and the arguments will be a populated JSON document matching the schema included in the FunctionDefinition used; it is not guaranteed that this data is valid or even properly formatted, however, so validation and error checking should always accompany function call processing.

To resolve the function call and continue the user-facing interaction, process the argument payload as needed and then serialize appropriate response data into a new message with ChatRole.Function. Then make a new request with all of the messages so far -- the initial User message, the first response's FunctionCall message, and the resolving Function message generated in reply to the function call -- so the model can use the data to better formulate a chat completions response.

Note that the function call response you provide does not need to follow any schema provided in the initial call. The model will infer usage of the response data based on inferred context of names and fields.

ChatChoice responseChoice = response.Value.Choices[0];
if (responseChoice.FinishReason == CompletionsFinishReason.FunctionCall)
{
    // Include the FunctionCall message in the conversation history
    conversationMessages.Add(responseChoice.Message);

    if (responseChoice.Message.FunctionCall.Name == "get_current_weather")
    {
        // Validate and process the JSON arguments for the function call
        string unvalidatedArguments = responseChoice.Message.FunctionCall.Arguments;
        var functionResultData = (object)null; // GetYourFunctionResultData(unvalidatedArguments);
        // Here, replacing with an example as if returned from GetYourFunctionResultData
        functionResultData = new
        {
            Temperature = 31,
            Unit = "celsius",
        };
        // Serialize the result data from the function into a new chat message with the 'Function' role,
        // then add it to the messages after the first User message and initial response FunctionCall
        var functionResponseMessage = new ChatMessage(
            ChatRole.Function,
            JsonSerializer.Serialize(
                functionResultData,
                new JsonSerializerOptions() {  PropertyNamingPolicy = JsonNamingPolicy.CamelCase }));
        conversationMessages.Add(functionResponseMessage);
        // Now make a new request using all three messages in conversationMessages
    }
}

Generate images with DALL-E image generation models

Response<ImageGenerations> imageGenerations = await client.GetImageGenerationsAsync(
    new ImageGenerationOptions()
    {
        Prompt = "a happy monkey eating a banana, in watercolor",
        Size = ImageSize.Size256x256,
    });

// Image Generations responses provide URLs you can use to retrieve requested images
Uri imageUri = imageGenerations.Value.Data[0].Url;

Troubleshooting

When you interact with Azure OpenAI using the .NET SDK, errors returned by the service correspond to the same HTTP status codes returned for REST API requests.

For example, if you try to create a client using an endpoint that doesn't match your Azure OpenAI Resource endpoint, a 404 error is returned, indicating Resource Not Found.

Next steps

• Provide a link to additional code examples, ideally to those sitting alongside the README in the package's /samples directory.
• If appropriate, point users to other packages that might be useful.
• If you think there's a good chance that developers might stumble across your package in error (because they're searching for specific functionality and mistakenly think the package provides that functionality), point them to the packages they might be looking for.

Contributing

See the OpenAI CONTRIBUTING.md for details on building, testing, and contributing to this library.

This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit cla.microsoft.com.

When you submit a pull request, a CLA-bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., label, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.