.NET Console Azure OpenAI Chat (SDK)

This repository contains a minimal .NET 9 console app that uses the official OpenAI .NET SDK configured for Azure OpenAI to run a chat session with:

• Streaming output (tokens print as they arrive — no waiting for full responses)
• Persistent conversation history (system, user, and assistant turns)

The sample targets C# 13 and .NET 9.

Features

• Uses the OpenAI NuGet package (2.x) and the ChatClient API
• Configured for Azure OpenAI via endpoint and deployment name
• Streams assistant responses via CompleteChatStreamingAsync(...)
• Preserves full conversation context by appending UserChatMessage and AssistantChatMessage objects
• Exits the loop on a blank user input

Prerequisites

• .NET 9 SDK installed
• An Azure OpenAI resource with:
  • Resource endpoint (e.g., https://your-resource.openai.azure.com/)
  • API key
  • A deployed chat model (deployment name, e.g., a gpt-4o-mini deployment)
• Network access to the Azure OpenAI endpoint

Getting started

1. Clone/download this repo and open the solution.

2. Restore and build (Visual Studio or CLI):

dotnet restore
dotnet build -c Release

3. Set these environment variables:

• Windows (PowerShell):

$env:AZURE_OPENAI_ENDPOINT = "https://your-resource.openai.azure.com/"
$env:AZURE_OPENAI_API_KEY = "<your-api-key>"
$env:AZURE_OPENAI_CHAT_DEPLOYMENT = "<your-deployment-name>"

• macOS/Linux (bash/zsh):

export AZURE_OPENAI_ENDPOINT="https://your-resource.openai.azure.com/"
export AZURE_OPENAI_API_KEY="<your-api-key>"
export AZURE_OPENAI_CHAT_DEPLOYMENT="<your-deployment-name>"

4. Run:

dotnet run --project Azure.OpenAI.Chat.SDK.csproj

You should see a prompt:

Type your message and press Enter. Submit a blank line to exit.
You:

Type a message and watch the assistant reply stream in real time. Submit a blank line to exit.

How it works

The core logic lives in Program.cs:

• Create a client using an API key and Azure endpoint:
  • var client = new OpenAIClient(new ApiKeyCredential(apiKey), new OpenAIClientOptions { Endpoint = new Uri(endpoint) });
• Get a chat client from your Azure deployment:
  • var chat = client.GetChatClient(deployment);
• Maintain a List<ChatMessage> for history:
  • Start with a SystemChatMessage to set the assistant behavior.
  • Append each UserChatMessage and AssistantChatMessage after every turn.
• Stream responses:
  • await foreach (var update in chat.CompleteChatStreamingAsync(messages.ToArray())) { ... }
  • Print Text parts as they arrive and accumulate them into a buffer for history.

Changing the deployment

Update your Azure deployment name via the AZURE_OPENAI_CHAT_DEPLOYMENT environment variable (and ensure the model is deployed to the resource referenced by AZURE_OPENAI_ENDPOINT).

Non‑streaming example (reference)

If you prefer to wait for the complete response (not used in this sample), the SDK supports a non‑streaming call:

var result = await chat.CompleteChatAsync(messages.ToArray());
var completion = result.Value; // ChatCompletion
var text = string.Join("\n",
    Enumerable.Range(0, completion.Content.Count)
              .Select(i => completion.Content[i].Text)
              .Where(t => !string.IsNullOrEmpty(t)));

Troubleshooting

• AZURE_OPENAI_ENDPOINT/API_KEY/CHAT_DEPLOYMENT not set: The app exits with an error; set the environment variables and try again.
• 401/403 errors: Verify the API key, endpoint, and that the key belongs to the target resource.
• 404/BadRequest: Confirm the deployment name exists in the specified Azure OpenAI resource and the endpoint/region is correct.
• Networking issues: Ensure outbound HTTPS access to the Azure OpenAI endpoint.
• Quotas/limits: Check your Azure OpenAI quotas and usage.

Security notes

• Do not commit secrets (API keys) to source control.
• Prefer user or machine‑level environment variables or managed secret stores.

Tech stack

• .NET 9
• C# 13
• OpenAI NuGet package (2.x), configured for Azure OpenAI

Scripts/commands summary

• Restore: dotnet restore
• Build: dotnet build
• Run: dotnet run --project Azure.OpenAI.Chat.SDK.csproj