Remove id from FunctionCall and FunctionResponse when using Vertex AI

[kawolum]

** Please make sure you read the contribution guide and file the issues in the rigth place. **
Contribution guide.

Describe the bug
A clear and concise description of what the bug is.

Vertex AI does not accept id for FunctionCall and FunctionResponse. ADK sends FunctionCall and FunctionResponse with id regardless of Vertex AI.

To Reproduce
Steps to reproduce the behavior:

1. Create an agent with tools
2. Use Vertex AI with any Gemini models
3. Make the agent use a tool

Expected behavior
A clear and concise description of what you expected to happen.

Remove id from FunctionCall and FunctionResponse when it is configured for Vertex AI.

Screenshots
If applicable, add screenshots to help explain your problem.

Desktop (please complete the following information):

• OS: [e.g. iOS] Linux
• Java version: 21
• ADK version(see maven dependency): 0.3.0

Additional context
Add any other context about the problem here.

[hahan]

+1 to this. This breaks using Vertex AI with Gemini models. Python works as it ignores these fields but Java ADK enforces these ids.

[prasadskarmarkar]

I've tested this scenario with the latest code and I'm unable to reproduce the reported issue.

Test Details:

• Created a tool-enabled agent using Vertex AI with Gemini
• Successfully executed tool calls end-to-end
• Function calls and responses completed without errors

Could you please verify if this is still occurring with ADK version ? If so, please provide:

• Specific error messages or stack traces
• Sample code to reproduce
• Exact ADK version

Happy to investigate further if there are specific scenarios where this still fails.

[kawolum]

I was able to reproduce this issue with version 0.5.0

Sample code that was used to reproduce the error:

public class TimeAgent {
    public static BaseAgent ROOT_AGENT = initAgent();
    
    private static BaseAgent initAgent() {
        final Client client = client.builder()
                .vertexAI(true)
                // redacted the rest
                .build();

         return LlmAgent.builder().name("test").description("test").instruction("You are an agent")
         .model(new Gemini("gemini-3-flash-preview", client)).tools(FunctionTool.create(TimeAgent.class, "getCurrentTime")).build();
    }
}

[kawolum]

The error message was ClientException: 400 Bad Request. Invalid JSON payload received. Unknown name "id" at 'contents[1].parts[0].function_call': Cannot find field.

This happens when ADK creates the payload for the LLM after including the function response.

[asorian0]

+1 this make google-adk for java not usable while using vertexai

[asorian0]

they issue is reproducible only while using api key and vertexai. right now using service account credentials and vertexai is working as expected, so it seems coupled to that scenario

[hemasekhar-p]

Hi @kawolum,

Thank you for reporting this issue. Based on our investigation, The ADK Java library automatically adds a functionCall.id for each FunctionTool call. Running the example below shows the ID in the output while the tool executes successfully. This confirms that ADK injects the ID, and Vertex AI API accepts it and returns output as expected, no functionality is broken. Please try with the latest Vertex AI version. If you are still facing the same error, please share the complete buggy code and any error messages or stack traces. We will be happy to investigate further.

Code :

package com.api.apitest.chat.febw2;

import java.time.Instant;
import java.util.Map;

import com.google.adk.agents.BaseAgent;
import com.google.adk.agents.LlmAgent;
import com.google.adk.models.Gemini;
import com.google.adk.runner.InMemoryRunner;
import com.google.adk.sessions.Session;
import com.google.adk.tools.FunctionTool;
import com.google.adk.tools.Annotations.Schema;
import com.google.genai.Client;
import com.google.genai.types.Content;
import com.google.genai.types.Part;

public class TimeAgent {
	public static BaseAgent ROOT_AGENT = initAgent();

	public static void main(String[] args) {
		InMemoryRunner runner = new InMemoryRunner(ROOT_AGENT);
		Session session = runner.sessionService().createSession(runner.appName(), "user1").blockingGet();
		Content userMessage = Content.fromParts(Part.fromText("What is the current time?"));
		runner.runAsync(session.userId(), session.id(), userMessage)
				.blockingForEach(event -> System.out.println(event.stringifyContent()));
	}

	private static BaseAgent initAgent() {
		final Client client = Client.builder().vertexAI(true).build();
		return LlmAgent.builder().name("test").description("test").instruction("You are an agent")
				.model(new Gemini("gemini-2.5-flash", client))
				.tools(FunctionTool.create(TimeAgent.class, "getCurrentTime")).build();
	}

	@Schema(description = "Returns the current time")
	public static Map<String, String> getCurrentTime() {
		return Map.of("time", Instant.now().toString());
	}
}

[kawolum]

Hi @hemasekhar-p, thanks for taking a look.

I used your code almost exactly, I only changed how the client is built with location, project and credentials. Still the same error.

I am using ADK 0.5.0 and Google Gen AI 1.38.0 for reference.

Not sure what you meant by latest Vertex AI version.

[kawolum]

Was this fixed with commits after ADK 0.5.0 maybe?

[kawolum]

Here is the stack trace:

Caused by: com.google.genai.errors.ClientException: 400 Bad Request. Invalid JSON pyaload received. Unknown name "id" at 'contents[1].parts[0].function_call': Cannot find field.
Invalid JSON payload received. Unknown name "id" at 'contents[2].parts[0].function_response': Cannot find field.
    at com.google.genai.errors.ApiException.throwFromResponse(ApiException.java:94)
    at com.google.genai.HttpApiResponse.getBody(HttpApiResponse.java:37)
    at com.genai.Models.processResponseForPrivateGenerateContent(Models.java:5066)
    at com.google.genai.AsyncModels.lambda$privateGenerateContent$0(AsyncModels.java:117)

[kawolum]

The root cause is that vertex ai v1 does not support ids for function call and response but v1beta1 does. I am still using v1 and you didn't see the error because you are using v1beta1. I want to stick with v1 because v1beta1 is still beta.