lofcz/openaing

A simple C# / .NET library to use with OpenAI's API, including GPT-4 Turbo, Assistants, DALL·E 3, etc. Supports commercial and self hosted API providers such as OpenAI, Azure OpenAI, KoboldCpp, LocalAI and others.

.NET SDK for accessing the OpenAI Compatible APIs such as GPT-4 (Vision), Assistants and DALL-E 3

OpenAiNextGeneration is a simple .NET library to use with various OpenAI compatible providers, such as OpenAI, Azure OpenAI, and KoboldCpp ( v1.45.2+). Supports features such as function calling in conjunction with streaming, caches its own HttpClients.

Supported features compared to OpenAI-API-dotnet:

• Assistants, threads, and messages support.
• Includes new models.
• Parallel function calling.
• Improved memory usage, and function calling in conjunction with streaming.
• Manages its pool of HttpClients.
• Improved credentials passing.
• Supports OpenAI-compatible API providers, such as KoboldCpp.
• Improved Azure OpenAI integration.
• Nullability annotations.
• Calls are guaranteed not to throw, full response is included in the call result. (Fully supported in 3.0+)
• Actively maintained, backed by a company I work for.

Features scheduled for open-sourcing:

• High-level plugin API.
• Approximate token counting for streaming & function calling.

⚡Getting Started

Install the library via NuGet:

Install-Package OpenAiNg

🔮 Quick Inference

var api = new OpenAiNg.OpenAiApi("YOUR_API_KEY");
var result = await api.Completions.GetCompletion("One Two Three One Two");
Console.WriteLine(result);
// should print something starting with "Three"

📖 Readme

• Requirements
• Installation
• Authentication
• Additonal Documentation
• License

Requirements

Unlike the original library, OpenAiNg supports only .NET Core >= 6.0, if you need .NET Standard 2.0 /.NET Framework support, please use OpenAI-API-DotNet.

Getting started

Install from NuGet

Install package OpenAiNg from Nuget. Here's how via the command line:

Install-Package OpenAiNg

Authentication

Pass keys directly to ApiAuthentication(string key) constructor.

You use the APIAuthentication when you initialize the API as shown:

// for example
var api = new OpenAiApi("YOUR_API_KEY"); // shorthand
// or
var api = new OpenAiApi(new APIAuthentication("YOUR_API_KEY")); // create object manually

You may optionally include an OpenAi Organization if multiple Organizations are under one account.

// for example
var api = new OpenAiApi(new ApiAuthentication("YOUR_API_KEY", "org-yourOrgHere"));

Documentation

Every single class, method, and property has extensive XML documentation, so it should show up automatically in IntelliSense. That combined with the official OpenAI documentation should be enough to get started. Feel free to open an issue here if you have any questions.

License

This library is licensed under MIT license.

openaing's Issues

Cannot use a base64 image string when talking with GPT4 Vision

Hey,

I have a base64 encoded image string I want to pass to ChatGPT which I understand is valid, but this library only allows a URL. In my use case this isn't a possibility as the images are not publically accessible.

Can the ChatImage type be extended to allow base64 images?

Thanks!

Have you considered using HTTP/2 for the EndpointClient?

I see that the endpoint client sets MaxConnectionsPerServer and PooledConnectionLifetime to help control connection management, but my understanding is that leveraging HTTP/2 (or higher) can also help with this problem by multiplexing connections:

HttpClient EndpointClient = new(new SocketsHttpHandler
{
	MaxConnectionsPerServer = 10000,
	PooledConnectionLifetime = TimeSpan.FromMinutes(2)
})
{
	Timeout = TimeSpan.FromSeconds(600),
	DefaultRequestVersion = new(2, 0) // default to higher version
};

I'm curious whether you've found a reason to be using 1.1 instead?

Should functionCallHandler in GetResponseFromChatbotAsyncWithFunctions return a `List<FunctionResult>`?

Right now the handler is defined as Func<List<FunctionCall>, Task<FunctionResult?>>. This implies that a single FunctionResult can capture the results of multiple functions. However, I can't see how to do that via the API. I would have expected something like this:

Func<List<FunctionCall>, Task<List<FunctionResult>>>

That way if there are N calls I can return N results. Am I misunderstanding the intended usage pattern?

ResponseFormat type

I'm surprised that ChatRequest.ResponseFormat is type ChatRequestResponseFormats instead of ChatRequestResponseFormatTypes. Can you explain what I'm missing or a cleaner way to do it?

e.g.:

wanting to do:

    ResponseFormat = ChatRequestResponseFormatTypes.Text,

instead of:

var textType = new ChatRequestResponseFormats();
textType.Type = ChatRequestResponseFormatTypes.Text;
...
    ResponseFormat = textType,

(actually I am swapping between text and json but that's not relevant for this example)

Assistants

Are there any example Assistants

Should ChatMessage.Name be nullable?

My understanding is that this property is optional; the doc comment even says so. Shouldn't this be marked as nullable?

Assistants

Are there any example Assistants