Claudia

Unofficial Anthropic Claude API client for .NET.

We have built a C# API similar to the official Python SDK and TypeScript SDK. It supports netstandard2.1, net6.0, and net8.0.

In addition to the pure client SDK, it also includes a C# Source Generator for performing Function Calling.

Installation

This library is distributed via NuGet, supporting .NET Standard 2.1, .NET 6(.NET 7) and .NET 8 or above.

PM> Install-Package Claudia

It can also be used with Unity Game Engine both Runtime and Editor. For instructions on how to use it, please refer to the Unity section.

You can also use it with AWS Bedrock. Check the AWS Bedrock section for more details.

Usage

For details about the API, please check the official API reference.

using Claudia;

var anthropic = new Anthropic
{
    ApiKey = Environment.GetEnvironmentVariable("ANTHROPIC_API_KEY") // This is the default and can be omitted
};

var message = await anthropic.Messages.CreateAsync(new()
{
    Model = "claude-3-5-sonnet-20240620", // you can use Claudia.Models.Claude3_5Sonnet string constant
    MaxTokens = 1024,
    Messages = [new() { Role = "user", Content = "Hello, Claude" }]
});

Console.WriteLine(message);

Claudia is designed to have a similar look and feel to the official client, particularly the TypeScript SDK. However, it does not use object, dynamic, or Dictionary, and everything is strongly typed. By utilizing C# 9.0 Target-typed new expressions and C# 12 Collection expressions, it can be written in a simple manner.

Streaming Messages

We provide support for streaming responses using Server Sent Events (SSE).

using Claudia;

var anthropic = new Anthropic();

var stream = anthropic.Messages.CreateStreamAsync(new()
{
    Model = "claude-3-opus-20240229",
    MaxTokens = 1024,
    Messages = [new() { Role = "user", Content = "Hello, Claude" }]
});

await foreach (var messageStreamEvent in stream)
{
    Console.WriteLine(messageStreamEvent);
}

If you need to cancel a stream, you can pass the CancellationToken to CreateStreamAsync.

The Stream returns an IAsyncEnumerable<IMessageStreamEvent>, allowing it to be enumerated using await foreach. The implementation types of IMessageStreamEvent can be found in IMessageStreamEvent.cs.

For example, outputs the text results.

await foreach (var messageStreamEvent in stream)
{
    if (messageStreamEvent is ContentBlockDelta content)
    {
        Console.WriteLine(content.Delta.Text);
    }
}

Request & Response types

This library includes C# definitions for all request params and response fields. You may import and use them like so:

using Claudia;

var request = new MessageRequest()
{
    Model = Models.Claude3Opus,
    MaxTokens = 1024,
    Messages = [new() { Role = Roles.User, Content = "Hello, Claude" }]
};

Documentation for each method, request param, and response field are available in docstrings and will appear on hover in most modern editors.

All of MessageRequest definitions are here MessageRequest.cs and MessageResponse definitions are here MessageResponse.cs.

Also, commonly used constants are defined. For example, Models.Claude3Opus is claude-3-opus-20240229, and constants like Roles.User and Roles.Assistant are used for roles like "user" and "assistant". Please refer to Constant.cs for all the constants. In addition, the system prompt used in Claude's official chat UI is defined as SystemPrompts.Claude3.

Counting Tokens

You can see the exact usage for a given request through the usage response property, e.g.

var message = await anthropic.Messages.CreateAsync(...)

// Usage { InputTokens = 11, OutputTokens = 18 }
Console.WriteLine(message.Usage);

Streaming Helpers

By integrating with R3, the new Reactive Extensions library, it becomes possible to handle Streaming Events in various ways.

// convert to array.
var array = await stream.ToObservable().ToArrayAsync();

// filterling and execute.
await stream.ToObservable()
    .OfType<IMessageStreamEvent, ContentBlockDelta>()
    .Where(x => x.Delta.Text != null)
    .ForEachAsync(x =>
    {
        Console.WriteLine(x.Delta.Text);
    });

// branching query
var branch = stream.ToObservable().Publish();

var messageStartTask = branch.OfType<IMessageStreamEvent, MessageStart>().FirstAsync();
var messageDeltaTask = branch.OfType<IMessageStreamEvent, MessageDelta>().FirstAsync();

branch.Connect(); // start consume stream

Console.WriteLine((await messageStartTask));
Console.WriteLine((await messageDeltaTask));

// Sum Usage
var totalUsage = await stream.ToObservable()
    .Where(x => x is MessageStart or MessageDelta)
    .Select(x => x switch
    {
        MessageStart ms => ms.Message.Usage,
        MessageDelta delta => delta.Usage,
        _ => throw new ArgumentException()
    })
    .AggregateAsync((x, y) => new Usage { InputTokens = x.InputTokens + y.InputTokens, OutputTokens = x.OutputTokens + y.OutputTokens });

Console.WriteLine(totalUsage);

Handling errors

When the library is unable to connect to the API, or if the API returns a non-success status code (i.e., 4xx or 5xx response), a subclass of ClaudiaException will be thrown:

try
{
    var msg = await anthropic.Messages.CreateAsync(new()
    {
        Model = Models.Claude3Opus,
        MaxTokens = 1024,
        Messages = [new() { Role = "user", Content = "Hello, Claude" }]
    });
}
catch (ClaudiaException ex)
{
    Console.WriteLine((int)ex.Status); // 400(ErrorCode.InvalidRequestError)
    Console.WriteLine(ex.Name);        // invalid_request_error
    Console.WriteLine(ex.Message);     // Field required. Input:...
}

Error codes are as followed:

public enum ErrorCode
{
    /// <summary>There was an issue with the format or content of your request.</summary>
    InvalidRequestError = 400,
    /// <summary>There's an issue with your API key.</summary>
    AuthenticationError = 401,
    /// <summary>Your API key does not have permission to use the specified resource.</summary>
    PermissionError = 403,
    /// <summary>The requested resource was not found.</summary>
    NotFoundError = 404,
    /// <summary>Your account has hit a rate limit.</summary>
    RateLimitError = 429,
    /// <summary>An unexpected error has occurred internal to Anthropic's systems.</summary>
    ApiError = 500,
    /// <summary>Anthropic's API is temporarily overloaded.</summary>
    OverloadedError = 529
}

Retries

Certain errors will be automatically retried 2 times by default, with a short exponential backoff. Connection errors (for example, due to a network connectivity problem), 408 Request Timeout, 409 Conflict, 429 Rate Limit, and >=500 Internal errors will all be retried by default.

You can use the MaxRetries option to configure or disable this:

// Configure the default for all requests:
var anthropic = new Anthropic
{
    MaxRetries = 0, // default is 2
};

// Or, configure per-request:
await anthropic.Messages.CreateAsync(new()
{
    MaxTokens = 1024,
    Messages = [new() { Role = "user", Content = "Hello, Claude" }],
    Model = "claude-3-opus-20240229"
}, new()
{
    MaxRetries = 5
});

Timeouts

Requests time out after 10 minutes by default. You can configure this with a Timeout option:

// Configure the default for all requests:
var anthropic = new Anthropic
{
    Timeout = TimeSpan.FromSeconds(20) // 20 seconds (default is 10 minutes)
};

// Override per-request:
await anthropic.Messages.CreateAsync(new()
{
    MaxTokens = 1024,
    Messages = [new() { Role = "user", Content = "Hello, Claude" }],
    Model = "claude-3-opus-20240229"
}, new()
{
    Timeout = TimeSpan.FromSeconds(5)
});

On timeout, an TimeoutException is thrown.

Note that requests which time out will be retried twice by default.

Default Headers

We automatically send the anthropic-version header set to 2023-06-01.

If you need to, you can override it by setting default headers on a per-request basis.

Be aware that doing so may result in incorrect types and other unexpected or undefined behavior in the SDK.

await anthropic.Messages.CreateAsync(new()
{
    MaxTokens = 1024,
    Messages = [new() { Role = "user", Content = "Hello, Claude" }],
    Model = "claude-3-opus-20240229"
}, new()
{
    Headers = new() { { "anthropic-version", "My-Custom-Value" } }
});

Customizing the HttpClient

The Anthropic client uses a standard HttpClient by default for communication. If you want to customize the behavior of the HttpClient, pass an HttpMessageHandler. Additionally, if you don't want to dispose the HttpClient when disposing the Anthropic client, you can set the disposeHandler flag to false.

public class Anthropic : IDisposable
{
    public HttpClient HttpClient => httpClient;

    public Anthropic()
        : this(new HttpClientHandler(), true)
    {
    }

    public Anthropic(HttpMessageHandler handler)
        : this(handler, true)
    {
    }

    public Anthropic(HttpMessageHandler handler, bool disposeHandler)
    {
        this.httpClient = new HttpClient(handler, disposeHandler);
        this.httpClient.Timeout = System.Threading.Timeout.InfiniteTimeSpan; // Timeout is ignored, Anthropic client uses timeout settings from Timeout(or override per request) property
    }

    public void Dispose()
    {
        httpClient.Dispose();
    }
}

Furthermore, you can retrieve the HttpClient used for requests via the HttpClient property. This allows you to modify settings such as DefaultRequestHeaders.

// disable keep-alive
anthropic.HttpClient.DefaultRequestHeaders.ConnectionClose = true;

You can change the HttpClient.BaseAddress to change the API address(e.g., for proxies).

// request to http://myproxy/messages instead of https://api.anthropic.com/v1/messages
anthropic.HttpClient.BaseAddress = new Uri("http://myproxy/");

Accessing raw Response data (e.g., headers)

By defining a DelegatingHandler in accordance with the HttpClient convention, it is possible to hook into the request pipeline before and after the request. This allows for logging, adding telemetry, and checking response header information. Upon observing the headers, you can see that additional information related to RateLimit is included.

Let's create a DelegatingHandler that retrieves this information and throws a special exception with the RateLimit information when a RateLimit occurs.

public class RateLimitDetailsHandler : DelegatingHandler
{
    protected override async Task<HttpResponseMessage> SendAsync(HttpRequestMessage request, CancellationToken cancellationToken)
    {
        var response = await base.SendAsync(request, cancellationToken);

        if ((int)response.StatusCode == (int)ErrorCode.RateLimitError)
        {
            var requestLimit = int.Parse(response.Headers.GetValues("anthropic-ratelimit-requests-limit").First());
            var requestRemaining = int.Parse(response.Headers.GetValues("anthropic-ratelimit-requests-remaining").First());
            var requestReset = DateTime.Parse(response.Headers.GetValues("anthropic-ratelimit-requests-reset").First());
            var tokenLimit = int.Parse(response.Headers.GetValues("anthropic-ratelimit-tokens-limit").First());
            var tokenRemaining = int.Parse(response.Headers.GetValues("anthropic-ratelimit-tokens-remaining").First());
            var tokenReset = DateTime.Parse(response.Headers.GetValues("anthropic-ratelimit-tokens-reset").First());

            var error = await response.Content.ReadFromJsonAsync<ErrorResponseShape>(AnthropicJsonSerializerContext.Default.Options, cancellationToken);
            var message = error!.ErrorResponse.Message;

            throw new AnthropicRateLimitException(requestLimit, requestRemaining, requestReset, tokenLimit, tokenRemaining, tokenReset, message);
        }

        return response;
    }
}

public class AnthropicRateLimitException : Exception
{
    public int RateLimitRequestsLimit { get; }
    public int RateLimitRequestsRemaining { get; }
    public DateTime RateLimitRequestsReset { get; }
    public int RateLimitTokensLimit { get; }
    public int RateLimitTokensRemaining { get; }
    public DateTime RateLimitTokensReset { get; }


    public AnthropicRateLimitException(
        int rateLimitRequestsLimit,
        int rateLimitRequestsRemaining,
        DateTime rateLimitRequestsReset,
        int rateLimitTokensLimit,
        int rateLimitTokensRemaining,
        DateTime rateLimitTokensReset,
        string message) : base(message)
    {
        RateLimitRequestsLimit = rateLimitRequestsLimit;
        RateLimitRequestsRemaining = rateLimitRequestsRemaining;
        RateLimitRequestsReset = rateLimitRequestsReset;
        RateLimitTokensLimit = rateLimitTokensLimit;
        RateLimitTokensRemaining = rateLimitTokensRemaining;
        RateLimitTokensReset = rateLimitTokensReset;
    }
}

The DelegatingHandler (HttpMessageHandler) can be passed to the constructor of Anthropic.

var anthropic = new Anthropic(new RateLimitDetailsHandler() { InnerHandler = new HttpClientHandler() });

Upload File

Message.Content accepts multiple Content objects. However, if a single string is passed, it is automatically converted into an array of text.

// this code
Content = "Hello, Claude"
// is convert to following
Content = [new Content
{
    Type = "text",
    Text = "Hello, Claude"
}]

When passing an image, set both the image and Text in the Content.

var imageBytes = File.ReadAllBytes(@"dish.jpg");

var anthropic = new Anthropic();
var message = await anthropic.Messages.CreateAsync(new()
{
    Model = "claude-3-opus-20240229",
    MaxTokens = 1024,
    Messages = [new()
    {
        Role = "user",
        Content = [
            new()
            {
                Type = "image",
                Source = new()
                {
                    Type = "base64",
                    MediaType = "image/jpeg",
                    Data = imageBytes
                }
            },
            new()
            {
                Type = "text",
                Text = "Describe this image."
            }
        ]
    }],
});
Console.WriteLine(message);

The above code can be simplified. If a string is passed to the Content constructor, it is set as text, and if ReadOnlyMemory<byte> is passed, it is set as an image.

var message = await anthropic.Messages.CreateAsync(new()
{
    Model = Models.Claude3Opus,
    MaxTokens = 1024,
    Messages = [new()
    {
        Role = Roles.User,
        Content = [
            new(imageBytes, "image/jpeg"),
            "Describe this image."
        ]
    }],
});
Console.WriteLine(message);

Currently, there are four types of uploadable binaries: image/jpeg, image/png, image/gif, and image/webp. For example, if you want to upload a markdown file, it's best to read its contents and send it as text. If you want to upload a PDF, you can either convert it to text or an image before sending. Presentation files like pptx can also be sent as images, and Claude will interpret the content and extract the text for you.

System and Temperature

Other optional properties of MessageRequest include System, Metadata, StopSequences, Temperature, TopP, and TopK.

var message = await anthropic.Messages.CreateAsync(new()
{
    Model = Models.Claude3Haiku,
    MaxTokens = 1024,
    System = SystemPrompts.Claude3,
    Temperature = 0.4,
    Messages = [
        new() { Role = Roles.User, Content = "Hello, Claude" },
    ],
});

SystemPrompts.Claude3 is a string constant for the System Prompt used in the Official Chat UI. Of course, you can also set any arbitrary System Prompt.

Save / Load

All request and response models can be serialized using System.Text.Json.JsonSerializer. Additionally, AnthropicJsonSerializerContext has pre-generated serializers available through Source Generator, enabling even higher performance.

List<Message> chatMessages;

void Save()
{
    var json = JsonSerializer.Serialize(chatMessages, AnthropicJsonSerializerContext.Default.Options);
    File.WriteAllText("chat.json", json);
}

void Load()
{
    chatMessages = JsonSerializer.Deserialize<List<Message>>("chat.json", AnthropicJsonSerializerContext.Default.Options)!;
}

Function Calling

Claude supports Function Calling.

Tool use

Tool use(function calling) is new style of function calling. Currently it is beta and need to add anthropic-beta flag in header.

var anthropic = new Anthropic();
anthropic.HttpClient.DefaultRequestHeaders.Add("anthropic-beta", "tools-2024-04-04");

With Claudia, you only need to define static methods annotated with [ClaudiaFunction], and the C# Source Generator automatically generates the necessary code.

public static partial class FunctionTools
{
    /// <summary>
    /// Retrieve the current time of day in Hour-Minute-Second format for a specified time zone. Time zones should be written in standard formats such as UTC, US/Pacific, Europe/London.
    /// </summary>
    /// <param name="timeZone">The time zone to get the current time for, such as UTC, US/Pacific, Europe/London.</param>
    [ClaudiaFunction]
    public static string TimeOfDay(string timeZone)
    {
        var time =  TimeZoneInfo.ConvertTimeBySystemTimeZoneId(DateTime.UtcNow, timeZone);
        return time.ToString("HH:mm:ss");
    }
}

The partial class includes the generated .AllTools, .Tools.[Methods] and .InvokeToolAsync(MessageResponse).

Function Calling requires two requests to Claude. The flow is as follows: "Initial request to Claude with available tools in System Prompt -> Execute functions based on the message containing the necessary tools -> Include the results in a new message and send another request to Claude."

var anthropic = new Anthropic();
anthropic.HttpClient.DefaultRequestHeaders.Add("anthropic-beta", "tools-2024-04-04");

var input = new Message { Role = Roles.User, Content = "What time is it in Los Angeles?" };
var message = await anthropic.Messages.CreateAsync(new()
{
    Model = Models.Claude3Haiku,
    MaxTokens = 1024,
    Tools = FunctionTools.AllTools, // use generated Tools
    Messages = [input],
});

// invoke local function
var toolResult = await FunctionTools.InvokeToolAsync(message);

var response = await anthropic.Messages.CreateAsync(new()
{
    Model = Models.Claude3Haiku,
    MaxTokens = 1024,
    Tools = FunctionTools.AllTools,
    Messages = [
        input,
        new() { Role = Roles.Assistant, Content = message.Content },
        new() { Role = Roles.User, Content = toolResult! }
    ],
});

// The current time in Los Angeles is 10:45 AM.
Console.WriteLine(response.Content.ToString());

The return type of ClaudiaFunction can also be specified as Task<T> or ValueTask<T>. This allows you to execute a variety of tasks, such as HTTP requests or database requests. For example, a function that retrieves the content of a specified webpage can be defined as shown above.

public static partial class FunctionTools
{
    // ...

    /// <summary>
    /// Retrieves the HTML from the specified URL.
    /// </summary>
    /// <param name="url">The URL to retrieve the HTML from.</param>
    [ClaudiaFunction]
    static async Task<string> GetHtmlFromWeb(string url)
    {
        // When using this in a real-world application, passing the raw HTML might consume too many tokens.
        // You can parse the HTML locally using libraries like AngleSharp and convert it into a compact text structure to save tokens.
        using var client = new HttpClient();
        return await client.GetStringAsync(url);
    }
}

Note that the allowed parameter types are bool, sbyte, byte, short, ushort, int, uint, long, ulong, decimal, float, double, string, DateTime, DateTimeOffset, Guid, TimeSpan and Enum.

The return value can be of any type, but it will be converted to a string using ToString(). If you want to return a custom string, make the return type string and format the string within the function.

Legacy style

The Anthropic Cookbook provides examples of Function Calling. To achieve this, complex XML generation and parsing processing, as well as execution based on the parsed results, are required.

With Claudia, you only need to define static methods annotated with [ClaudiaFunction], and the C# Source Generator automatically generates the necessary code, including parsers and system messages.

public static partial class FunctionTools
{
    /// <summary>
    /// Retrieve the current time of day in Hour-Minute-Second format for a specified time zone. Time zones should be written in standard formats such as UTC, US/Pacific, Europe/London.
    /// </summary>
    /// <param name="timeZone">The time zone to get the current time for, such as UTC, US/Pacific, Europe/London.</param>
    [ClaudiaFunction]
    public static string TimeOfDay(string timeZone)
    {
        var time =  TimeZoneInfo.ConvertTimeBySystemTimeZoneId(DateTime.UtcNow, timeZone);
        return time.ToString("HH:mm:ss");
    }
}

The partial class includes the generated .SystemPrompt and .InvokeAsync(MessageResponse).

Function Calling requires two requests to Claude. The flow is as follows: "Initial request to Claude with available tools in System Prompt -> Execute functions based on the message containing the necessary tools -> Include the results in a new message and send another request to Claude."

// `FunctionTools.SystemPrompt` contains the XML used to inform Claude about the available tools.
// This XML is generated from the method's XML documentation comments.
/*
In this environment you have access to a set of tools you can use to answer the user's question.
...
You may call them like this:
...
Here are the tools available:
<tools>
  <tool_description>
    <tool_name>TimeOfDay</tool_name>
    <description>Retrieve the current time of day in Hour-Minute-Second format for a specified time zone. Time zones should be written in standard formats such as UTC, US/Pacific, Europe/London.</description>
    <parameters>
      <parameter>
        <name>timeZone</name>
        <type>string</type>
        <description>The time zone to get the current time for, such as UTC, US/Pacific, Europe/London.</description>
      </parameter>
    </parameters>
  </tool_description>
</tools>
*/
// Console.WriteLine(FunctionTools.SystemPrompt);

var input = new Message { Role = Roles.User, Content = "What time is it in Los Angeles?" };
var message = await anthropic.Messages.CreateAsync(new()
{
    Model = Models.Claude3Haiku,
    MaxTokens = 1024,
    System = FunctionTools.SystemPrompt, // set generated prompt
    StopSequences = [StopSequnces.CloseFunctionCalls], // set </function_calls> as stop sequence
    Messages = [input],
});

// Claude returns xml to invoke tool
/*
<function_calls>
    <invoke>
        <tool_name>TimeOfDay</tool_name>
        <parameters>
            <timeZone>US/Pacific</timeZone>
        </parameters>
    </invoke>
*/
// Console.WriteLine(message);

// `FunctionTools.InvokeAsync`, which is automatically generated, parses the function name and parameters from the `MessageResponse`,
// executes the corresponding function, and generates XML to inform Claude about the function execution results.
var partialAssistantMessage = await FunctionTools.InvokeAsync(message);

// By passing this message to Claude as the beginning of the Assistant's response,
// it will provide a continuation that takes into account the function execution results.
/*
<function_calls>
    <invoke>
        <tool_name>TimeOfDay</tool_name>
        <parameters>
            <timeZone>US/Pacific</timeZone>
        </parameters>
    </invoke>
</function_calls>
<function_results>
    <result>
        <tool_name>TimeOfDay</tool_name>
        <stdout>03:27:03</stdout>
    </result>
</function_results>
*/
// Console.WriteLine(partialAssistantMessage);

var callResult = await anthropic.Messages.CreateAsync(new()
{
    Model = Models.Claude3Haiku,
    MaxTokens = 1024,
    System = FunctionTools.SystemPrompt,
    Messages = [
        input, // User: "What time is it in Los Angeles?"
        new() { Role = Roles.Assistant, Content = partialAssistantMessage! } // set as Assistant
    ],
});

// The current time in Los Angeles (US/Pacific time zone) is 03:36:04.
Console.WriteLine(callResult);

For the initial request, specifying StopSequences.CloseFunctionCalls is efficient. Also, if you want to include your own System Prompt, it's a good idea to concatenate it with the generated SystemPrompt.

The return type of ClaudiaFunction can also be specified as Task<T> or ValueTask<T>. This allows you to execute a variety of tasks, such as HTTP requests or database requests. For example, a function that retrieves the content of a specified webpage can be defined as shown above.

public static partial class FunctionTools
{
    // ...

    /// <summary>
    /// Retrieves the HTML from the specified URL.
    /// </summary>
    /// <param name="url">The URL to retrieve the HTML from.</param>
    [ClaudiaFunction]
    static async Task<string> GetHtmlFromWeb(string url)
    {
        // When using this in a real-world application, passing the raw HTML might consume too many tokens.
        // You can parse the HTML locally using libraries like AngleSharp and convert it into a compact text structure to save tokens.
        using var client = new HttpClient();
        return await client.GetStringAsync(url);
    }
}

var input = new Message
{
    Role = Roles.User,
    Content = """
        Could you summarize this page in three line?
        https://docs.anthropic.com/claude/docs/intro-to-claude
"""
};

var message = await anthropic.Messages.CreateAsync(new()
{
    Model = Models.Claude3Haiku,
    MaxTokens = 1024,
    System = FunctionTools.SystemPrompt, // set generated prompt
    StopSequences = [StopSequnces.CloseFunctionCalls], // set </function_calls> as stop sequence
    Messages = [input],
});

var partialAssistantMessage = await FunctionTools.InvokeAsync(message);

var callResult = await anthropic.Messages.CreateAsync(new()
{
    Model = Models.Claude3Haiku,
    MaxTokens = 1024,
    System = FunctionTools.SystemPrompt,
    Messages = [
        input,
        new() { Role = Roles.Assistant, Content = partialAssistantMessage! } // set as Assistant
    ],
});

// The page can be summarized in three lines:
// 1. Claude is a family of large language models developed by Anthropic designed to revolutionize the way you interact with AI.
// 2. This documentation is designed to help you get the most out of Claude, with clear explanations, examples, best practices, and links to additional resources.
// 3. Claude excels at a wide variety of tasks involving language, reasoning, analysis, coding, and more, and the documentation covers key capabilities, getting started with prompting, and using the API.
Console.WriteLine(callResult);

Multiple functions can be defined, and they can be executed multiple times in a single request.

public static partial class FunctionTools
{
    [ClaudiaFunction]
    public static string TimeOfDay(string timeZone) //...

    // Sample of https://github.com/anthropics/anthropic-cookbook/blob/main/function_calling/function_calling.ipynb

    /// <summary>
    /// Calculator function for doing basic arithmetic. 
    /// Supports addition, subtraction, multiplication
    /// </summary>
    /// <param name="firstOperand">First operand (before the operator)</param>
    /// <param name="secondOperand">Second operand (after the operator)</param>
    /// <param name="operator">The operation to perform. Must be either +, -, *, or /</param>
    [ClaudiaFunction]
    static double DoPairwiseArithmetic(double firstOperand, double secondOperand, string @operator)
    {
        return @operator switch
        {
            "+" => firstOperand + secondOperand,
            "-" => firstOperand - secondOperand,
            "*" => firstOperand * secondOperand,
            "/" => firstOperand / secondOperand,
            _ => throw new ArgumentException("Operation not supported")
        };
    }
}

var input = new Message
{
    Role = Roles.User,
    Content = """
        What time is it in Seattle and Tokyo?
        Incidentally multiply 1,984,135 by 9,343,116.
"""
};

var message = await anthropic.Messages.CreateAsync(new()
{
    Model = Models.Claude3Haiku,
    MaxTokens = 1024,
    System = FunctionTools.SystemPrompt, // set generated prompt
    StopSequences = [StopSequnces.CloseFunctionCalls], // set </function_calls> as stop sequence
    Messages = [input],
});

var partialAssistantMessage = await FunctionTools.InvokeAsync(message);

var callResult = await anthropic.Messages.CreateAsync(new()
{
    Model = Models.Claude3Haiku,
    MaxTokens = 1024,
    System = FunctionTools.SystemPrompt,
    Messages = [
        input,
        new() { Role = Roles.Assistant, Content = partialAssistantMessage! } // set as Assistant
    ],
});

// The time in Seattle (US/Pacific time zone) is 8:06:53.
// The time in Tokyo (Asia/Tokyo time zone) is 00:06:53.
// The result of multiplying 1,984,135 by 9,343,116 is 18,524,738,326,760.
Console.WriteLine(callResult);

Note that the allowed parameter types are bool, sbyte, byte, short, ushort, int, uint, long, ulong, decimal, float, double, string, DateTime, DateTimeOffset, Guid, TimeSpan and Enum.

The return value can be of any type, but it will be converted to a string using ToString(). If you want to return a custom string, make the return type string and format the string within the function.

Blazor Sample

By using Claudia with Blazor, you can easily create a Chat UI like the one shown below.

All the code can be found in BlazorApp1.

The key parts are the setup in Program.cs and Home.razor.cs.

// Program.cs

// get ANTHROPIC_API_KEY from user secret
// https://learn.microsoft.com/en-us/aspnet/core/security/app-secrets
Environment.SetEnvironmentVariable("ANTHROPIC_API_KEY", builder.Configuration["ANTHROPIC_API_KEY"]);

// Add Anthropic Client
builder.Services.AddSingleton<Anthropic>();

var app = builder.Build();

// Home.razor.cs

using Claudia;
using Microsoft.AspNetCore.Components;
using Microsoft.AspNetCore.Components.Web;

namespace BlazorApp1.Components.Pages;

public partial class Home
{
    [Inject]
    public required Anthropic Anthropic { get; init; }

    double temperature = 1.0;
    string textInput = "";
    string systemInput = SystemPrompts.Claude3;
    List<Message> chatMessages = new();

    bool running = false;

    async Task SendClick()
    {
        if (running) return;
        if (string.IsNullOrWhiteSpace(textInput)) return;

        running = true;
        try
        {
            chatMessages.Add(new() { Role = Roles.User, Content = textInput });

            var stream = Anthropic.Messages.CreateStreamAsync(new()
            {
                Model = Models.Claude3Opus,
                MaxTokens = 1024,
                Temperature = temperature,
                System = string.IsNullOrWhiteSpace(systemInput) ? null : systemInput,
                Messages = chatMessages.ToArray()
            });

            var currentMessage = new Message { Role = Roles.Assistant, Content = "" };
            chatMessages.Add(currentMessage);

            textInput = ""; // clear input.
            StateHasChanged();

            await foreach (var messageStreamEvent in stream)
            {
                if (messageStreamEvent is ContentBlockDelta content)
                {
                    currentMessage.Content[0].Text += content.Delta.Text;
                    StateHasChanged();
                }
            }
        }
        finally
        {
            running = false;
        }
    }
}

If you need to store the chat message history, you can serialize List<Message> chatMessages to JSON and save it to a file or database.

AWS Bedrock

We provide support for the Anthropic Bedrock API through a separate package.

PM> Install-Package Claudia.Bedrock

To create an AmazonBedrockRuntimeClient from the AWS SDK and specify the Bedrock Model ID using UseAnthropic, set the Model property of RequestMessage to anthropic_version. The rest is the same as a regular Anthropic Client.

// credentials is your own
AWSConfigs.AWSProfileName = "";

var bedrock = new AmazonBedrockRuntimeClient(RegionEndpoint.USEast1);
var anthropic = bedrock.UseAnthropic("anthropic.claude-3-haiku-20240307-v1:0"); // Model Id

var response = await anthropic.Messages.CreateAsync(new()
{
   Model = "bedrock-2023-05-31", // anthropic_version
   MaxTokens = 1024,
   Messages = [new() { Role = "user", Content = "Hello, Claude" }]
});

Console.WriteLine(response);

Streaming Messages work in the same way.

var stream = anthropic.Messages.CreateStreamAsync(new()
{
    Model = "bedrock-2023-05-31", // anthropic_version
    MaxTokens = 1024,
    Messages = [new() { Role = "user", Content = "Hello, Claude" }]
});

await foreach (var item in stream)
{
    Console.WriteLine(item);
}

If you need the raw response, call InvokeModelAsync or InvokeModelWithResponseStreamAsync instead. This allows you to check the status code and headers before retrieving the result with GetMessageResponse or GetMessageResponseAsync.

var bedrock = new AmazonBedrockRuntimeClient(RegionEndpoint.USEast1);

// (string modelId, MessageRequest request)
var response = await bedrock.InvokeModelAsync("anthropic.claude-3-haiku-20240307-v1:0", new()
{
    Model = "bedrock-2023-05-31", // anthropic_version
    MaxTokens = 1024,
    Messages = [new() { Role = "user", Content = "Hello, Claude" }]
});

Console.WriteLine(response.ResponseMetadata.RequestId);

var responseMessage = response.GetMessageResponse();

Console.WriteLine(responseMessage);

Unity

Minimum supported Unity version is 2022.3.12f1. You need to install from NuGet. We recommend using NuGetForUnity.

Install NuGetForUnity
Open Window from NuGet -> Manage NuGet Packages, Search "Claudia" and Press Install.

With this, you can use the Anthropic client in both the Editor and at Runtime.

using Claudia;
using System;
using UnityEngine;

public class NewBehaviourScript : MonoBehaviour
{
    async void Start()
    {
        var anthropic = new Anthropic()
        {
            ApiKey = "YOUR API KEY"
        };

        Debug.Log("Start Simple Call in Unity");

        var message = await anthropic.Messages.CreateAsync(new()
        {
            Model = Models.Claude3Opus,
            MaxTokens = 1024,
            Messages = new Message[] { new() { Role = "user", Content = "Hello, Claude" } }
        });

        Debug.Log("User: Hello, Claude");
        Debug.Log("Assistant: " + message);
    }
}

Source Generators for Function Calling are also supported, but additional work is required.

Setup the C# compiler for unity.
- Add a text file named csc.rsp with the following contents under your Assets/.
  - ```
  -langVersion:10 -nullable
```
Setup the C# compiler for your IDE.
- Install CsprojModifier
- Add a text file named LangVersion.props with the following contents
  - ```
  <Project xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
    <PropertyGroup>
      <LangVersion>10</LangVersion>
      <Nullable>enable</Nullable>
    </PropertyGroup>
  </Project>
```
- Open Project Settings and [C# Project Modifier] section under the [Editor].
- Add the .props file you just created, to the list of [Additional project imports].
- Note:
  - If you are using assembly definition, add your additional csproj in the list of [The project to be addef for import].

If you encount Constructors of types with required members are not supported in this version of your compiler error, LangVersion should change to 11.

using Claudia;
using System;
using UnityEngine;

public class NewBehaviourScript : MonoBehaviour
{
    async void Start()
    {
        var anthropic = new Anthropic()
        {
            ApiKey = "YOUR API KEY"
        };

        Debug.Log("Start Function Calling Demo in Unity");

        var input = new Message
        {
            Role = Roles.User,
            Content = "Multiply 1,984,135 by 9,343,116"
        };

        var message = await anthropic.Messages.CreateAsync(new()
        {
            Model = Models.Claude3Haiku,
            MaxTokens = 1024,
            System = FunctionTools.SystemPrompt,
            StopSequences = new[] { StopSequnces.CloseFunctionCalls },
            Messages = new[] { input },
        });

        var partialAssistantMessage = await FunctionTools.InvokeAsync(message);

        var callResult = await anthropic.Messages.CreateAsync(new()
        {
            Model = Models.Claude3Haiku,
            MaxTokens = 1024,
            System = FunctionTools.SystemPrompt,
            Messages = new[]{
                input,
                new() { Role = Roles.Assistant, Content = partialAssistantMessage! }
            },
        });

        Debug.Log("User: Multiply 1,984,135 by 9,343,116");
        Debug.Log("Assistant: " + callResult.ToString().Trim());
    }
}

public static partial class FunctionTools
{
    /// <summary>
    /// Calculator function for doing basic arithmetic. 
    /// Supports addition, subtraction, multiplication
    /// </summary>
    /// <param name="firstOperand">First operand (before the operator)</param>
    /// <param name="secondOperand">Second operand (after the operator)</param>
    /// <param name="operator">The operation to perform. Must be either +, -, *, or /</param>
    [ClaudiaFunction]
    static double DoPairwiseArithmetic(double firstOperand, double secondOperand, string @operator)
    {
        return @operator switch
        {
            "+" => firstOperand + secondOperand,
            "-" => firstOperand - secondOperand,
            "*" => firstOperand * secondOperand,
            "/" => firstOperand / secondOperand,
            _ => throw new ArgumentException("Operation not supported")
        };
    }
}

WebGL

In environments where using HttpClient for communication fails, such as WebGL, you need to replace the transport layer by using this UnityWebRequestHttpMessageHandler.cs.

// replace handler to UnityWebRequestHttpMessageHandler
var anthropic = new Anthropic(new UnityWebRequestHttpMessageHandler())
{
    ApiKey = "YOUR API KEY",
    ConfigureAwait = true // in Unity Environment with UnityWebRequest, true is recommended.
};

However, to make API calls in a browser environment, you need to set up a proxy or take other measures to support CORS.

License

This library is under the MIT License.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

README.md

README.md

Claudia

Installation

Usage

Streaming Messages

Request & Response types

Counting Tokens

Streaming Helpers

Handling errors

Retries

Timeouts

Default Headers

Customizing the HttpClient

Accessing raw Response data (e.g., headers)

Upload File

System and Temperature

Save / Load

Function Calling

Tool use

Legacy style

Blazor Sample

AWS Bedrock

Unity

WebGL

License

Files

README.md

Latest commit

History

README.md

File metadata and controls

Claudia

Installation

Usage

Streaming Messages

Request & Response types

Counting Tokens

Streaming Helpers

Handling errors

Retries

Timeouts

Default Headers

Customizing the HttpClient

Accessing raw Response data (e.g., headers)

Upload File

System and Temperature

Save / Load

Function Calling

Tool use

Legacy style

Blazor Sample

AWS Bedrock

Unity

WebGL

License