Cisharpai.Azure 0.3.0-alpha.7.164

Cisharpai.Azure

Azure AI provider for the Cisharpai unified LLM client library. Covers both Azure OpenAI Service and Azure AI Inference (model-as-a-service).

Features

Azure OpenAI

• Chat completions with all Azure OpenAI deployments
• Reasoning models (o1, o3, o4, GPT-5) with automatic max_completion_tokens handling and optional ReasoningEffort
• Text embeddings (ada-002, text-embedding-3-small/large)
• JSON Mode and Structured Outputs
• Tool calling / function calling
• Vision (image file paths and base64)
• Streaming (token-by-token via SSE)
• API key and Azure AD (Entra ID) authentication

Azure AI Inference

• Chat completions with Azure AI model catalog (Phi-3, Llama-3, Mistral, etc.)
• Reasoning model detection and appropriate request formatting
• Text and image embeddings (with IImageEmbeddingFeature)
• JSON Mode and Structured Outputs
• Tool calling / function calling
• Vision (image file paths and base64)
• Streaming (token-by-token via SSE)

Quick Start — Azure OpenAI

using Cisharpai;
using Cisharpai.Models;
using Cisharpai.Azure.Extensions;
using Microsoft.Extensions.DependencyInjection;

var services = new ServiceCollection();
services.AddAzureOpenAiClient(options =>
{
    options.Endpoint = "https://myresource.openai.azure.com";
    options.ApiKey = "YOUR_API_KEY";
    options.DeploymentName = "gpt-4o";
    // Optional for o1/o3/o4/gpt-5 deployments:
    // options.ReasoningEffort = "low";
});

var provider = services.BuildServiceProvider();
var client = provider.GetRequiredService<IChatCompletionClient>();

var response = await client.GetChatCompletionAsync(
    new ChatCompletionRequest(
        Messages: [new LlmMessage(LlmRole.User, "Hello!")]));

ReasoningEffort is sent only for reasoning deployments. For provider parameters that are not typed yet, or to override a typed value per request, use ChatCompletionRequest.ExtraParameters; it deep-merges into the final JSON request.

Quick Start — Azure AI Inference

services.AddAzureAiInferenceChatCompletion(options =>
{
    options.Endpoint = "https://mymodel.eastus.models.ai.azure.com";
    options.ApiKey = "YOUR_API_KEY";
    options.ModelId = "Phi-3-mini-4k-instruct";
});

Authentication

Both API key and Azure AD (Bearer token) authentication are supported. Omit ApiKey to use Azure.Identity DefaultAzureCredential with scope https://cognitiveservices.azure.com/.default.

Keyed Services

.NET 8 keyed DI is supported for registering multiple clients:

services.AddAzureOpenAiClient("gpt4o", options => { /* ... */ });
services.AddAzureOpenAiClient("gpt4mini", options => { /* ... */ });

Links

• GitHub
• Provider Feature Matrix