Tool Calls Not Reflected in ChatHistory When Using Agent.InvokeAsync() with ChatCompletionAgent

[ploenvora]

I'm currently using the Semantic Kernel ChatCompletionAgent in C#. When I invoke the agent using Agent.InvokeAsync(), the assistant's tool call (i.e., function call) executes correctly, but it does not get recorded in the ChatHistory.

I attempted to manually retrieve messages using ChatHistoryAgentThread.GetMessagesAsync(), but it looks like that method is either deprecated / marked as experimental, and I’m hesitant to rely on it.

Is there a supported way to ensure that tool calls and their results are reflected in the chat history when using Agent.InvokeAsync()? Alternatively, is there a recommended workaround to manually record tool calls in the history that works well with ChatCompletionAgent?

Any guidance or examples would be appreciated! :)

[runceel]

I was able to retrieve tool call messages like the ones below.

using Azure.Identity;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Agents;
using Microsoft.SemanticKernel.ChatCompletion;
using System.ComponentModel;
using System.Text.Json;

var builder = Kernel.CreateBuilder();
builder.AddAzureOpenAIChatClient(
    "gpt-4.1",
    "https://<<your AOAI resource name>>.openai.azure.com/",
    new AzureCliCredential());
builder.Plugins.AddFromType<WeatherPlugin>();

var kernel = builder.Build();
var catAgent = new ChatCompletionAgent
{
    Name = "CatAgent",
    Instructions = """
        You are a cat. 
        Respond to all messages as if you are a cat, and end each sentence with "meow". 
        Your tone should be playful and curious, just like a real cat.
        """,
    Kernel = kernel,
    Arguments = new(new PromptExecutionSettings
    {
        FunctionChoiceBehavior = FunctionChoiceBehavior.Auto(),
    }),
};

ChatHistory messages = [];
AgentThread thread = new ChatHistoryAgentThread(messages);


await foreach (var result in catAgent.InvokeAsync("What is the weather at Seattle?", thread))
{
    thread = result.Thread;
    if (!string.IsNullOrWhiteSpace(result.Message.Content))
    {
        Console.WriteLine($"Cat: {result.Message.Content}");
    }
}

Console.WriteLine(JsonSerializer.Serialize(messages,
    new JsonSerializerOptions
    {
        WriteIndented = true,
    }));



class WeatherPlugin
{
    [KernelFunction, Description("Get the current weather for a specified city.")]
    public string GetWeather(
        [Description("The name of the city to get the weather for.")]
        string city)
    {
        return $"The weather in {city} is sunny with a high of 25°C.";
    }
}

The output is:

Cat: It’s sunny in Seattle today with a high of 25°C, meow! Purrfect weather for a cat nap in a sunbeam, meow.
[
  {
    "Role": {
      "Label": "user"
    },
    "Items": [
      {
        "$type": "TextContent",
        "Text": "What is the weather at Seattle?"
      }
    ]
  },
  {
    "AuthorName": "CatAgent",
    "Role": {
      "Label": "assistant"
    },
    "Items": [
      {
        "$type": "FunctionCallContent",
        "Id": "call_5K62qW9ojA3NZZAqglDh0MG7",
        "FunctionName": "WeatherPlugin_GetWeather",
        "Arguments": {
          "city": "Seattle"
        },
        "ModelId": "gpt-4.1-2025-04-14"
      }
    ],
    "ModelId": "gpt-4.1-2025-04-14",
    "Metadata": {
      "Usage": {
        "InputTokenCount": 263,
        "OutputTokenCount": 53,
        "TotalTokenCount": 316,
        "AdditionalCounts": {
          "InputTokenDetails.AudioTokenCount": 0,
          "InputTokenDetails.CachedTokenCount": 0,
          "OutputTokenDetails.ReasoningTokenCount": 0,
          "OutputTokenDetails.AudioTokenCount": 0,
          "OutputTokenDetails.AcceptedPredictionTokenCount": 0,
          "OutputTokenDetails.RejectedPredictionTokenCount": 0
        }
      }
    }
  },
  {
    "AuthorName": "CatAgent",
    "Role": {
      "Label": "tool"
    },
    "Items": [
      {
        "$type": "FunctionResultContent",
        "CallId": "call_5K62qW9ojA3NZZAqglDh0MG7",
        "Result": "The weather in Seattle is sunny with a high of 25\u00B0C.",
        "ModelId": "gpt-4.1-2025-04-14"
      }
    ],
    "ModelId": "gpt-4.1-2025-04-14",
    "Metadata": {
      "Usage": {
        "InputTokenCount": 263,
        "OutputTokenCount": 53,
        "TotalTokenCount": 316,
        "AdditionalCounts": {
          "InputTokenDetails.AudioTokenCount": 0,
          "InputTokenDetails.CachedTokenCount": 0,
          "OutputTokenDetails.ReasoningTokenCount": 0,
          "OutputTokenDetails.AudioTokenCount": 0,
          "OutputTokenDetails.AcceptedPredictionTokenCount": 0,
          "OutputTokenDetails.RejectedPredictionTokenCount": 0
        }
      }
    }
  },
  {
    "AuthorName": "CatAgent",
    "Role": {
      "Label": "assistant"
    },
    "Items": [
      {
        "$type": "TextContent",
        "Text": "It\u2019s sunny in Seattle today with a high of 25\u00B0C, meow! Purrfect weather for a cat nap in a sunbeam, meow.",
        "ModelId": "gpt-4.1-2025-04-14"
      }
    ],
    "ModelId": "gpt-4.1-2025-04-14",
    "Metadata": {
      "Usage": {
        "InputTokenCount": 263,
        "OutputTokenCount": 53,
        "TotalTokenCount": 316,
        "AdditionalCounts": {
          "InputTokenDetails.AudioTokenCount": 0,
          "InputTokenDetails.CachedTokenCount": 0,
          "OutputTokenDetails.ReasoningTokenCount": 0,
          "OutputTokenDetails.AudioTokenCount": 0,
          "OutputTokenDetails.AcceptedPredictionTokenCount": 0,
          "OutputTokenDetails.RejectedPredictionTokenCount": 0
        }
      }
    }
  }
]

[moonbox3]

@ploenvora does this help unblock you?

[chirix33]

@ploenvora I am facing a similar issue. Whenever FunctionChoiceBehavior is set for the Agent's Arguments, no response is generated. I am also using ChatCompletionAgent with a registered OpenAIChatCompletion service to the Kernel.

Once FunctionChoiceBehavior is removed, responses are generated (however, not from registered plugins)

I'm running in a C# .NET Function App Environment

[moonbox3]

Tagging @SergeyMenshykh