microsoft/autogen · error · InvalidOperationException

Only one choice is supported in streaming response

Error message

Only one choice is supported in streaming response

What it means

The streaming post-processor converts each StreamingChatMessageContent into a TextMessageUpdate, but only supports choice index 0. If the execution settings request multiple choices (ChoiceCount/n > 1), streamed chunks for choices 1+ carry ChoiceIndex > 0 and this InvalidOperationException is thrown.

Source

Thrown at dotnet/src/AutoGen.SemanticKernel/Middleware/SemanticKernelChatMessageContentConnector.cs:106

            _ => throw new InvalidOperationException("Unsupported content type"),
        });

        if (items.Count() == 1)
        {
            return items.First();
        }
        else
        {
            return new MultiModalMessage(Role.Assistant, items, from: messageEnvelope.From);
        }
    }

    private IMessage PostProcessMessage(IMessage<StreamingChatMessageContent> streamingMessage)
    {
        var chatMessageContent = streamingMessage.Content;
        if (chatMessageContent.ChoiceIndex > 0)
        {
            throw new InvalidOperationException("Only one choice is supported in streaming response");
        }
        return new TextMessageUpdate(Role.Assistant, chatMessageContent.Content, streamingMessage.From);
    }

    private IEnumerable<ChatMessageContent> ProcessMessage(IEnumerable<IMessage> messages, IAgent agent)
    {
        return messages.SelectMany(m =>
        {
            if (m is IMessage<ChatMessageContent> chatMessageContent)
            {
                return [chatMessageContent.Content];
            }
            if (m.From == agent.Name)
            {
                return ProcessMessageForSelf(m);
            }
            else
            {

View on GitHub (pinned to 027ecf0a37)

Solutions

1. Set ResultsPerPrompt = 1 (default) in the prompt execution settings used with streaming.
2. If multiple choices are required, consume IAsyncEnumerable<StreamingChatMessageContent> directly and filter by ChoiceIndex instead of using the connector.
3. Catch InvalidOperationException around the streaming loop and fail with a configuration message pointing at ResultsPerPrompt.

Example fix

// before
var settings = new OpenAIPromptExecutionSettings { ResultsPerPrompt = 3 };
await foreach (var update in agent.StreamAsync(msg, settings)) { }

// after
var settings = new OpenAIPromptExecutionSettings { ResultsPerPrompt = 1 };
await foreach (var update in agent.StreamAsync(msg, settings)) { }

if (executionSettings.ResultsPerPrompt is > 1 && useStreaming)
    throw new InvalidOperationException("Streaming through this connector requires ResultsPerPrompt == 1.");

catch (InvalidOperationException ex) when (ex.Message.Contains("Only one choice"))
{
    logger.LogError("Set ResultsPerPrompt=1 or stream manually filtering by ChoiceIndex");
    throw;
}

When it happens

Trigger: OpenAIPromptExecutionSettings.ResultsPerPrompt (or equivalent ChoiceCount) greater than 1 while streaming a Semantic Kernel agent through this connector.

Common situations: Reusing execution settings tuned for best-of-N sampling; copying SK samples that set ResultsPerPrompt; switching an agent from non-streaming to streaming without reviewing choice settings.

Related errors

• Please set OPENAI_API_KEY environment variable.
• Please set OPENAI_API_KEY environment variable.
• Please set OPENAI_API_KEY environment variable.
• The recent update is not a TextMessage
• The content has more than one choice. Please try another inp

AI-assisted analysis of microsoft/autogen@027ecf0a37 (2026-08-15). Data as JSON: /api/errors/7014af123acdc854. Report an issue: GitHub.