Universal.OpenAI.Client 3.0.0

Universal.OpenAI.Client

Breaking Changes

1. Message types are now more specific: SystemMessage, UserMessage, AssistantMessage, ToolMessage, FunctionMessage (deprecated), and UnknownMessage.
2. The Models class now includes Gpt4, Gpt4Turbo, and potentially other new models.
3. ResponseFormat can now be set to JsonObject or a custom schema using FromJsonSchema.
4. Tool usage has been introduced, replacing the older function calling mechanism.
5. The Content property of messages can now be either a string or IEnumerable<ContentBlock>.
6. Embeddings, image generation have been marked as obsoleted. They will be refactored at a later point.
7. Text completion support has been removed.

Usage

Chat Completion

using Universal.OpenAI.Client;
using Universal.OpenAI.Client.Chat;

using (OpenAIClient openAIClient = new OpenAIClient(ApiKey))
{
    ChatCompletion response = await openAIClient.CreateChatCompletionAsync(new CreateChatCompletionRequest()
    {
        Model = Models.Gpt4,
        Messages = new Message[]
        {
            new SystemMessage("You are a chatbot for a company named Red Marble AI. Be as helpful as you can without being too terse or verbose.")
        }
    });

    foreach (var choice in response.Choices)
    {
        Console.WriteLine(choice.Message.Content);
    }
}

JSON Output Enforcement

You can enforce JSON output from the model:

ChatCompletion response = await openAIClient.CreateChatCompletionAsync(new CreateChatCompletionRequest()
{
    Model = Models.Gpt4Turbo,
    Messages = [
        new SystemMessage("Your system message here"),
        new UserMessage("Your user message here")
    ],
    ResponseFormat = ResponseFormat.JsonObject
});

Structured JSON Output

You can specify a schema for structured JSON output:

ChatCompletion response = await openAIClient.CreateChatCompletionAsync(new CreateChatCompletionRequest()
{
    Model = "gpt-4o-2024-08-06",
    Messages = [
        new UserMessage("Your message here")
    ],
    ResponseFormat = ResponseFormat.FromJsonSchema("schema_name", new JsonSchema()
    {
        Type = "object",
        Properties = new Dictionary<string, JsonSchema>()
        {
            ["propertyName"] = new JsonSchema()
            {
                Type = "boolean",
                Description = "Property description"
            }
        },
        AdditionalProperties = false,
        Required = ["propertyName"]
    })
});

Using Tools

You can use tools (like functions) in your chat completions:

ChatCompletion firstTurn = await openAIClient.CreateChatCompletionAsync(new CreateChatCompletionRequest()
{
    Model = "gpt-4o-2024-08-06",
    Messages = [
        new SystemMessage("System message"),
        new UserMessage("User message")
    ],
    Tools = [
        new Tool() {
            Type = ToolTypes.Function,
            Function = new Function()
            {
                Name = "function_name",
                Description = "Function description",
                Parameters = new JsonSchema
                {
                    // Define your function parameters here
                }
            }
        }
    ]
});

// Process the tool call and get the tool response

ChatCompletion secondTurn = await openAIClient.CreateChatCompletionAsync(new CreateChatCompletionRequest()
{
    Model = "gpt-4o-2024-08-06",
    Messages = [
        // Previous messages
        new ToolMessage(toolResponse, toolCallId)
    ]
});