Streaming with Vertex AI produces empty parts in tool-calling follow-up requests

[piotresky]

Environment details

• Programming language: C#
• OS: Windows 11
• Language runtime version: .NET 8
• Package version: 1.6.1
• Microsoft.Extensions.AI: 10.4.1

Steps to reproduce

1. Configure the client with vertexAI: true
2. Set up a chat client with FunctionInvocation middleware and a tool
3. Send a prompt that triggers a tool call
4. The first LLM call succeeds and returns a function call
5. The follow-up LLM call (with tool results in history) fails with:

Google.GenAI.ClientError: Unable to submit request because it must include at least one parts field,
which describes the prompt input. Learn more:
https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/gemini

using System.ComponentModel;
using Google.Apis.Auth.OAuth2;
using Google.GenAI;
using Microsoft.Extensions.AI;

var vertexClient = new Client(
        project: "my-project",
        vertexAI: true,
        location: "us-central1",
        credential: GoogleCredential.GetApplicationDefault())
    .AsIChatClient("gemini-3-flash-preview")
    .AsBuilder()
    .UseFunctionInvocation()
    .Build();

[Description("Gets the current weather for a city")]
string GetWeather(string city) => $"Weather in {city}: 22°C, sunny";

var options = new ChatOptions { Tools = [AIFunctionFactory.Create(GetWeather)] };

// This fails on the second LLM call (after tool execution) with Vertex AI.
// Works fine with API key mode.
await foreach (var update in vertexClient.GetStreamingResponseAsync(
    "What's the weather in Warsaw?", options))
{
    Console.Write(update.Text);
}

Observed behavior

The follow-up request (after tool execution) contains a Content object with an empty parts array: { "role": "model", "parts": [] }. Vertex AI rejects this. The Gemini API (API key mode) tolerates it and works correctly.

The issue was introduced in 1.6.1. Version 1.6.0 works correctly with Vertex AI.

[shivvaam0001]

Hi @piotresky
Thank you for bringing this to our attention. We truly appreciate you flagging this issue, we will file a bug internally.

Thanks!