[BUG] The PageSizeLimit property does not work with the AssistantClient.GetMessages call.

[aszti911]

Describe the bug

The PageSizeLimit parameter does not work with the GetMessages() and GetMessagesAsync() methods in the instantiated assistantClient class. No matter how I set the parameter, the response always contains the maximum number of messages.

I tested the endpoint provided by OpenAI (GET https://api.openai.com/v1/threads/{thread_id}/messages) via Postman using the same parameter, and it worked as expected there. This leads me to believe that the issue is not with the endpoint itself, but likely with the library implementation.

Steps to reproduce

Just call the affected methods to reproduce the bug:

Code snippets

MessageCollectionOptions options = new()
            {
                Order = order is not null ? order : MessageCollectionOrder.Ascending
            };
            if (limit.HasValue && limit > 0 && limit <= 100)
            {
                options.PageSizeLimit = limit.Value;
            }

            List<ThreadMessage> result = assistantClient.GetMessages(threadId, options).ToList();

OS

Linux Mint

.NET version

9.0.300

Library version

2.2.0

[jsquire]

Hi @aszti911. Thanks for reaching out and we regret that you're experiencing difficulties. When you call GetMessages, the return is a collection of pages - not a single page. Because you're calling ToList on that collection, each page is enumerated and the messages from each of those pages is added to the list that is returned. As a result, that call will always produce a set with every message.

The return value is optimized for ease of use, so though the results are paged, this is normally abstracted away. You can see the pages and that the page size is being respected by dropping down to the protocol details with something similar to:

var result = assistantClient.GetMessages(threadId, new() { PageSizeLimit = 2 });
var pageCount = 0;

// This unwraps enumerator and gives you visibility into each page.
foreach (var page in result.GetRawPages())
{
    Console.WriteLine($"PAGE:{++pageCount}");
        
    // This pulls the JSON response out of the page and shows you the 
    //  list of messages in this specific page.
    Console.WriteLine(page.GetRawResponse().Content.ToString());

    ConsoleWriteLine("============================================================");
    Console.WriteLine();
}

[aszti911]

Dear @jsquire,

Thank you for the quick response! Based on your explanation, it seems I was not paying enough attention — the PageSizeLimit option does not limit the total number of results, but rather the number of items per page.

In our project, a single thread can contain a large number of messages — sometimes several hundred. I would like to ask if there are any plans to implement a method that would allow setting a true limit on the number of results returned, so we wouldn’t have to retrieve an excessive amount of messages all at once? As I mentioned, the OpenAI endpoint supports this kind of limit, which is why I assumed that the GetMessages method would behave similarly.

[jsquire]

@aszti911: You can do this today by just reading the number of items that you're interested in. Pages are not eagerly retrieved - they're loaded on-demand as you iterate over the items. If you were to set PageSizeLimit to the number of items that you wanted and then read just that amount you would only pay the network costs for the pages needed to fill that. For example:

Explicit

var desiredMessages = 10;
var result = assistantClient.GetMessagesAsync(threadId, new() { PageSizeLimit = desiredMessages });
var messages = new List<ThreadMessage>(desiredMessages);

await foreach (var message in result)
{
    messages.Add(message);

    if (messages.Count >= desiredMessages)
    {
        break;
    }
}

Linq

var desiredMessages = 10;
var result = assistantClient.GetMessages(threadId, new() { PageSizeLimit = desiredMessages });

// Note that the order of these is important.  If you reverse them, you'll
// walk all pages and collect all of the items.
var messages = result
    .Take(desiredMessages)
    .ToList();

Assuming that the service returns the number of items that you asked for, each of these would result in one network operation and retrieve just the first page.

The one thing to be aware of is that the service parameter (limit) represents the maximum number of items that it should return. The API docs do not guarantee a minimum size, regardless of whether or not there are enough messages. Since the client has no insight nor influence over how the service chooses to package the response, you may invoke a additional network operations if the service is returning less data than you were asking for.