C# Anthropic Claude Library! You Can Call Claude API Easy and Intuitive

Introduction

For two days, I worked on Anthropic API and completed creating a C# library, called HigLabo.Anthropic. You can easily call Claude AI by this library. This article will explain how to use it.

Background

Almost every day, AI provider introduces a new service. It is so fast that there is no library but only REST API endpoint. So, we must develop DIY library to consume API endpoint. I just created a library that may help someone who is planning to develop Claude API. I am sharing it here.

Download via Nuget.

HigLabo.Anthropic

All source code is on https://github.com/higty/higlabo/tree/master/Net8.

HigLabo.Anthropic is that.

You can see the sample code here.

Using the Code

Consume Messages endpoint receiving as stream via server sent event.

var cl = new AnthropicClient("API KEY");
var result = new MessagesStreamResult();
await foreach (string text in cl.MessagesStreamAsync
  ("How to enjoy coffee?", ModelNames.Claude3Opus, result, CancellationToken.None))
{
    Console.Write(text);
}
if (result.MessageDelta != null)
{
    Console.WriteLine("StopReason: " + result.MessageDelta.Delta.Stop_Reason);
    Console.WriteLine("Usage: " + result.MessageDelta.Usage.Output_Tokens);
}

You can pass all API parameter by using MessagesParameter class.

var cl = new AnthropicClient("API KEY");
var p = new MessagesParameter();
p.AddUserMessage("How to enjoy coffee?");
p.Model = ModelNames.Claude3Opus;
p.Max_Tokens = 1024;
// set other properties...
var result = new MessagesStreamResult();
await foreach (var item in cl.MessagesStreamAsync(p, result, CancellationToken.None))
{
    Console.Write(item);
}
if(result.MessageDelta != null)
{
    Console.WriteLine("StopReason: " + result.MessageDelta.Delta.Stop_Reason);
    Console.WriteLine("Usage: " + result.MessageDelta.Usage.Output_Tokens);
}

You can get all server sent event information by MessagesStreamResult after completion.

This is the code of function calling.

var cl = new AnthropicClient("API KEY");
var tools = new AnthropicTools();
var tool = new AnthropicTool("GetTickerSymbol", 
           "Gets the stock ticker symbol for a company searched by name. 
           Returns str: The ticker symbol for the company stock. Raises 
           TickerNotFound: if no matching ticker symbol is found.");
tool.Parameters.Add(new AnthropicToolParameter("company_name", "string", 
                    "The name of company"));
tools.Add(tool);
var toolXml = tool.ToString();

var p = new MessagesParameter();
p.Messages.Add(new ChatMessage(ChatMessageRole.User, 
               $"What is the current stock price of Microsoft?"));
p.SetTools(tools);
p.Model = ModelNames.Claude3Opus;
p.Max_Tokens = 1024;
p.Stream = true;

var result = new MessagesStreamResult();
await foreach (var item in cl.MessagesStreamAsync(p, result, CancellationToken.None))
{
    Console.Write(item);
}

var calls = AnthropicFunctionCalls.Parse(result.GetText());
if (calls.InvokeList.Count > 0)
{
    Console.WriteLine();
    Console.WriteLine("■Function call list");
    Console.WriteLine(calls.ToString());

    var invoke = calls.InvokeList.Find(el => el.ToolName == "GetTickerSymbol");
    if (invoke != null)
    {
        var companyName = invoke.GetParameterValue("company_name") ?? "";
        var tickerSymbol = GetTickerSymbol(companyName);
    }
}

This is the code to send image and get explanation.

var p = new MessagesParameter();
p.Model = "claude-3-opus-20240229";
p.Max_Tokens = 1024;

var msg = new ChatImageMessage(ChatMessageRole.User);
msg.AddTextContent($"What is this image include?");
msg.AddImageFile(Path.Combine(Environment.CurrentDirectory, "Image", "Rock.jpg"));
p.Messages.Add(msg);
p.Stream = true;

var result = new MessagesStreamResult();
await foreach (var item in cl.MessagesStreamAsync(p, result, CancellationToken.None))
{
    Console.Write(item);
}

Points of Interest

All class design architecture is the same as HigLabo.OpenAI. Please see this article.

You can use the same as HigLabo.OpenAI. HigLabo.Anthropic has:

• AnthropicClient
• XXXParameter
• XXXAsync
• XXXResponse
• RestApiResponse

classes and the behavior is the same. This proves that the class architecture design is absolutely well for REST API client library. Actually, I finished 6 hours work to implement the entire code of HigLabo.Anthropic. You can learn how to design class and library from these two library source code and if you have a plan to implement some REST API client library, it may help you.

History

• 17^th March, 2024: Initial release of HigLabo.Anthropic