Azure OpenAI .NET Client

This project contains an unofficial .NET client library for use with Azure's OpenAI resource. It contains an implementation of a pre-configured injectable OpenAIClient which can be used to communicate to the Azure OpenAI platform instead of relying purely on hand-crafted REST API calls.

NOTE: For the official Azure.AI.OpenAI NuGet package, please use this one.

OpenAIClient settings

The following settings allow the OpenAI client to connect to your Azure resource, authenticate, and work with the right deployment model you've created in the portal.

Setting	Purpose	Example
BaseUri	The fully qualified domain name of the OpenAI resource in Azure.	https://myai.openai.azure.com/
ApiKey	The API key used to authenticate the client to OpenAI in Azure.	1bbcc11a3a233857zz12aa5f2fake99af7d9c
DeploymentName	The name of the deployment for a given OpenAI model.	text-davinci-002

Configuration

The section below shows an example JSON configuration which can be stored in your secrets.json file, appsettings.development.json, or appsettings.json.

"OpenAiClientConfiguration": {
    "BaseUri": "[your_fqdn]",
    "ApiKey": "[your_api_key]",
    "DeploymentName": "[your_deployment_name]"
}

Setup in Program.cs

In your .NET application, you can use the NuGet package to allow for the following extension method which will register a singleton reference of the OpenAIClient instance.

builder.Services.AddOpenAIClient(x => builder.Configuration.Bind(nameof(OpenAIClientConfiguration), x));

OpenAIClient usage

Since the OpenAIClient is registered for Dependency Injection, you can inject it and use it with the GetTextCompletionResponseAsync() method. The CompletionResponse contains the model OpenAI uses with their typical REST API.

public class MyClass
{
    private readonly OpenAIClient _client;

    public MyClass(OpenAIClient client)
    {
        _client = client;
    }

    Task DoWork(string input)
    {
        var completionRequest = new CompletionRequest()
        {
            Prompt = input,
            MaxTokens = 100
        };
        CompletionResponse? completionResponse = await _client.GetTextCompletionResponseAsync(completionRequest);
    }        
}

Streaming support

The client now supports streaming using IAsyncEnumerable<CompletionResponse?> and can be iterated using the await foreach pattern as follows:

// inject the OpenAIClient and call streaming method
var completionRequest = new CompletionRequest()
{
    Prompt = input,
    MaxTokens = 100
};
var stream = _client.StreamTextCompletionResponseAsync(completionRequest);

await foreach (var completionResponse in stream)
{
    Console.Write(completionResponse.Choices.FirstOrDefault().Text);
}

Community Links

Blazor and Azure OpenAI