Microsoft.Extensions.AI IChatClient Implementation

Introduction

This document describes the GenerativeAIChatClient class, which implements the IChatClient interface from Microsoft.Extensions.AI. This implementation allows developers to seamlessly integrate Google's Generative AI models, specifically Gemini, into applications built using Microsoft's AI abstractions and dependency injection patterns. To use this functionality, you will need to install the Google_GenerativeAI.Microsoft NuGet package. This document will detail how to use this package.

Details

The GenerativeAIChatClient class provides a wrapper around the core GenerativeModel from the Google_GenerativeAI SDK, adapting it to the IChatClient interface. This allows for a consistent and familiar way to interact with Gemini models within .NET applications that utilize the Microsoft.Extensions.AI library.

Key Features and Functionality

1. IChatClient Implementation: The GenerativeAIChatClient class implements all the necessary methods and properties of the IChatClient interface, including CompleteAsync, CompleteStreamingAsync, and the Metadata property.
2. Constructor Overloads: The class offers constructors that accept either an API key or an IPlatformAdapter (for platform-specific configurations), providing flexibility in how the underlying GenerativeModel is initialized. A modelName parameter allows you to specify which Gemini model to use (defaults to GoogleAIModels.DefaultGeminiModel).
3. Asynchronous Operations: The CompleteAsync and CompleteStreamingAsync methods are asynchronous, ensuring that your application remains responsive while interacting with the Generative AI models.
4. Streaming Support: The CompleteStreamingAsync method enables streaming responses, allowing you to process and display the generated content as it becomes available.
5. Chat History Handling: The IChatClient interface itself manages the chat history. You provide the current set of ChatMessage objects, and the client handles the context internally.
6. Dependency Injection Integration: The client is designed to be easily registered with .NET's dependency injection container, making it simple to manage and use within your application.

Code Examples

1. Basic Usage (Non-Streaming)

using GenerativeAI.Microsoft;
using Microsoft.Extensions.AI;

// ... in your service registration ...
services.AddScoped<IChatClient>(provider => 
    new GenerativeAIChatClient(Environment.GetEnvironmentVariable("GOOGLE_API_KEY")));

// ... in your application code ...
public async Task<string> GetResponseAsync(string userMessage)
{
    var chatHistory = new List<ChatMessage> { new ChatMessage(AuthorRole.User, userMessage) };
    ChatCompletion completion = await _chatClient.CompleteAsync(chatHistory);
    return completion.Content;
}

2. Streaming Usage

using GenerativeAI.Microsoft;
using Microsoft.Extensions.AI;

// ... in your application code ...
public async IAsyncEnumerable<string> StreamResponseAsync(string userMessage)
{
    var chatHistory = new List<ChatMessage> { new ChatMessage(AuthorRole.User, userMessage) };
    await foreach (var update in _chatClient.CompleteStreamingAsync(chatHistory))
    {
        yield return update.Content;
    }
}

3. Using ChatOptions

using GenerativeAI.Microsoft;
using Microsoft.Extensions.AI;

// ... in your application code ...
public async Task<string> GetResponseWithOptionsAsync(string userMessage, ChatOptions options)
{
    var chatHistory = new List<ChatMessage> { new ChatMessage(AuthorRole.User, userMessage) };
    ChatCompletion completion = await _chatClient.CompleteAsync(chatHistory, options);
    return completion.Content;
}

4. Configuration with Dependency Injection

using GenerativeAI.Microsoft;
using Microsoft.Extensions.AI;
using Microsoft.Extensions.DependencyInjection;

// ...

public void ConfigureServices(IServiceCollection services)
{
    // Get API key from environment variables (recommended).
    string apiKey = Environment.GetEnvironmentVariable("GOOGLE_API_KEY");

    // Basic configuration:
    services.AddScoped<IChatClient>(provider => new GenerativeAIChatClient(apiKey));    
}

Important Considerations

1. Installation: Before using this functionality, make sure you have installed the Google_GenerativeAI.Microsoft NuGet package. You can install it using the NuGet Package Manager or the .NET CLI:

   Install-Package Google_GenerativeAI.Microsoft

2. API Key Management: Store your API key securely. Avoid hardcoding it directly in your code. Environment variables or user secrets are recommended.
3. Error Handling: Implement proper error handling to manage potential exceptions during API calls.
4. Rate Limiting: Be mindful of the API rate limits and implement appropriate strategies to handle them.
5. Context Window: Large conversations might exceed the model's context window. Consider strategies for managing conversation history and truncating older messages if necessary.
6. Model Selection: The modelName parameter allows you to select the appropriate Gemini model for your task. Consider the capabilities and cost implications of different models.

References

• Google_GenerativeAI.Microsoft NuGet Package
• Link to Microsoft.Extensions.AI Documentation