How get StreamingResponseUpdate or ResponseItem StructData

[The-Mojoo]

I am using Response and Stream, and I want to get data like:
event: response.output_text.delta data: {"type":"response.output_text.delta","item_id":"msg_67c9fdcf37fc8190ba82116e33fb28c507b8b0ad4e5eb654","output_index":0,"content_index":0,"delta":"Hi"}
or
{"type":"response.output_text.delta","item_id":"msg_67c9fdcf37fc8190ba82116e33fb28c507b8b0ad4e5eb654","output_index":0,"content_index":0,"delta":"Hi"}
primitive data structure, save in my database for logging purposes.
I know I can output each property individually to assemble them, but I still prefer ToJson() or System.Text.Json.JsonSerializer.Serialize(update) .

await foreach (var update in client.CreateResponseStreamingAsync(question, options))
{
    switch (update)
    {
        case StreamingResponseCompletedUpdate chatCompletion:
            // System.Text.Json.JsonSerializer.Serialize(chatCompletion)
            foreach (var responseItme in chatCompletion.Response.OutputItems)
            {
                switch (responseItme)
                {
                    case FunctionCallResponseItem functionCall:
                    // System.Text.Json.JsonSerializer.Serialize(functionCall) 
                    break;
                }
            }
            break;
    }
}

[jsquire]

Hi @The-Mojoo: Thanks for reaching out and your inquiry. The trick to getting raw SSE data is calling GetRawPagesAsync on the enumerable returned from the CreateResponseStreamingAsync. For example:

var client = new OpenAIResponseClient("gpt-5", "<< YOUR API KEY >>");

// Generally, samples show iterating over this collection to process each update.
AsyncCollectionResult<StreamingResponseUpdate> responseUpdates = client.CreateResponseStreamingAsync("Say 'this is a test.'");

// Instead, unwrap into each raw page.
await foreach (var page in responseUpdates.GetRawPagesAsync())
{
    // The content stream is available on the raw response.  Note that this is unbuffered, so 
    // you cannot access via the "Content" property - you must use the stream.
    var sseContentStream =  page.GetRawResponse().ContentStream;
}

More context and examples can be found in #813 and #515.

[The-Mojoo]

Hi @The-Mojoo: Thanks for reaching out and your inquiry. The trick to getting raw SSE data is calling GetRawPagesAsync on the enumerable returned from the CreateResponseStreamingAsync. For example:

var client = new OpenAIResponseClient("gpt-5", "<< YOUR API KEY >>");

// Generally, samples show iterating over this collection to process each update.
AsyncCollectionResult responseUpdates = client.CreateResponseStreamingAsync("Say 'this is a test.'");

// Instead, unwrap into each raw page.
await foreach (var page in responseUpdates.GetRawPagesAsync())
{
// The content stream is available on the raw response. Note that this is unbuffered, so
// you cannot access via the "Content" property - you must use the stream.
var sseContentStream = page.GetRawResponse().ContentStream;
}
More context and examples can be found in #813 and #515.

Yes, I have obtained the original SSE JSON data format you mentioned, but if I use GetRawPagesAsync(), I will lose the ability to operate on StreamingResponseUpdate.
Actually, I just want to get the data structure of OpenAIResponse from StreamingResponseCompletedUpdate, but JsonSerializer.Serialize(anyitem) always gives the following error related to 'Patch'. Is there any way for me to get the complete JSON data of OpenAIResponse while keeping the logic intact?

[jsquire]

@The-Mojoo: The recommended path would be to use ModelReaderWriter directly to serialize:

var anyItem = << your target instance >>;

// MRW serialized to a BinaryData instance.
var binaryData = ModelReaderWriter.Write(anyItem, ModelReaderWriterOptions.Json);

// If you need the raw JSON string, just call ToString()
var jsonString = binaryData.ToString();

Alternatively, you can use JsonModelConverter with JsonSerializer:

var anyItem = << your target instance >>;

JsonSerializerOptions options = new JsonSerializerOptions()
{
    Converters = { new JsonModelConverter() }
};

var jsonString = JsonSerializer.Serialize(anyItem, options);

For more context on what's going on, see the discussion in #817.

[The-Mojoo]

@The-Mojoo: The recommended path would be to use ModelReaderWriter directly to serialize:

var anyItem = << your target instance >>;

// MRW serialized to a BinaryData instance.
var binaryData = ModelReaderWriter.Write(anyItem, ModelReaderWriterOptions.Json);

// If you need the raw JSON string, just call ToString()
var jsonString = binaryData.ToString();
Alternatively, you can use JsonModelConverter with JsonSerializer:

var anyItem = << your target instance >>;

JsonSerializerOptions options = new JsonSerializerOptions()
{
Converters = { new JsonModelConverter() }
};

var jsonString = JsonSerializer.Serialize(anyItem, options);
For more context on what's going on, see the discussion in #817.

Actually, my main point is that I want to know how to work with the OpenAIResponse class while obtaining GetRawResponse().ContentStream. Or is it that the SSE data source format and the instantiated class cannot coexist?

Back to the question itself, I have already written the logic to operate on OpenAIResponse. I just want to add the complete information to the Response.
However, as mentioned above, in var parser = SseParser.Create(page.GetRawResponse().ContentStream!);, it indeed contains the content I want, but in this way, I need to manually extract attributes like type step by step from the data.

[jsquire]

@The-Mojoo: OpenAIResponse is the complete payload that is returned when you make a non-streaming call. The OpenAI service does not return information in this format when you're making a streaming call. Thus, it is not possible to directly translate the streaming events into OpenAIResponse. To do so, you would have to parse the streaming events that the service emits and extract each relevant bit of information until you had the full set. Then, you would use OpenAIResponsesModelFactory to create an instance with all of the relevant information.

I wouldn't recommend this approach, as it is quite a bit of complexity and I don't see any benefit. If you want to work with OpenAIResponse then your best path forward is to call the non-streaming version.

[The-Mojoo]

@The-Mojoo: OpenAIResponse is the complete payload that is returned when you make a non-streaming call. The OpenAI service does not return information in this format when you're making a streaming call. Thus, it is not possible to directly translate the streaming events into OpenAIResponse. To do so, you would have to parse the streaming events that the service emits and extract each relevant bit of information until you had the full set. Then, you would use OpenAIResponsesModelFactory to create an instance with all of the relevant information.

I wouldn't recommend this approach, as it is quite a bit of complexity and I don't see any benefit. If you want to work with OpenAIResponse then your best path forward is to call the non-streaming version.

Thanks, I upgraded to version 2.7 and saw that you added JsonIgnore to Patch, so it can now be serialized normally.
System.Text.Json.JsonSerializer.Serialize(OpenAIResponse)