[BUG] Azure OpenAI Chat Completions - OnYourData throwing Bad Request

[mikepugh]

Library name and version

Azure.AI.OpenAI - 2.1.0

Describe the bug

I've had the following code running in production since mid 2024:

options.AddDataSource(new AzureSearchChatDataSource()
{
    Endpoint = new Uri(searchEndpoint),
    IndexName = targetIndexName
    Authentication = DataSourceAuthentication.FromApiKey(searchApiKey)
}

And we just started noticing on Monday (3/10/2025) that when we call out for a chat completion via:

 client.CompleteChatStreamingAsync(messages, options, cancellationToken)

The system immediately throws an HTTP 400 Bad Request error. Nothing has changed with regards to our AI Search service setup or configuration.

I've also attempted to upgrade to the 2.2.0-beta.2 version of the library but that did not fix the issue. We are targeting the GPT-4o model in our Azure tenant.

Expected behavior

The chat completion would utilize the Azure AI Search endpoint for RAG functionality within our LLM chat application, and return a response with relevant citations.

Actual behavior

Service request failed.
Status: 400 (Bad Request)

Reproduction Steps

var options = new ChatCompletionOptions();

options.AddDataSource(new AzureSearchChatDataSource()
{
    Endpoint = new Uri(searchEndpoint),
    IndexName = targetIndexName
    Authentication = DataSourceAuthentication.FromApiKey(searchApiKey)
}

await foreach (var completion in client.CompleteChatStreamingAsync(messages, options, cancellationToken))
{
   // handle content updates...
}

Environment

Visual Studio 2022 Enterprise

.NET SDK:
Version: 9.0.103
Commit: 96da45d427
Workload version: 9.0.100-manifests.ea610b94
MSBuild version: 17.12.24+90b52dda6

Runtime Environment:
OS Name: Windows
OS Version: 10.0.26100
OS Platform: Windows
RID: win-x64
Base Path: C:\Program Files\dotnet\sdk\9.0.103\

[github-actions]

Thanks for the feedback! We are routing this to the appropriate team for follow-up. cc @jpalvarezl @ralph-msft @trrwilson.

[mikepugh]

I've just finished creating a bare bones replication of the issue, here is the full code:

using Azure.AI.OpenAI;
using Azure.AI.OpenAI.Chat;
using Azure.Identity;
using Microsoft.Extensions.Configuration;
using OpenAI.Chat;

var builder = new ConfigurationBuilder()
    .AddUserSecrets<Program>();

var config = builder.Build();

var searchApiKey = config["searchKey"];
var searchEndpoint = config["searchEndpoint"];
var indexName = config["indexName"];
var openAIEndpoint = config["openAiEndpoint"];
var modelName = "gpt-4o";

var client = new AzureOpenAIClient(new Uri(openAIEndpoint), new DefaultAzureCredential());

var chatClient = client.GetChatClient(modelName);

var options = new ChatCompletionOptions();

#pragma warning disable AOAI001 
options.AddDataSource(new AzureSearchChatDataSource()
{
    Endpoint = new Uri(searchEndpoint),
    IndexName = indexName,
    Authentication = DataSourceAuthentication.FromApiKey(searchApiKey)
});
#pragma warning restore AOAI001

var msgs = new List<ChatMessage> { new UserChatMessage("What can you tell me about cyber security?") };


// The code throws the HTTP 400 Bad Request error here
var response = await chatClient.CompleteChatAsync(msgs, options);


// Never gets to here..
foreach (var message in response.Value.Content)
{
    Console.WriteLine(message.Text);
}