Gemini models streamed tool calls are missing tool_call.id, resulting in no final response after tool execution

[Thirukumaran-SF4558]

Description

Hi! While testing DeepAgents with ChatGoogleGenerativeAI or ChatGoogle and Gemini 2.5 Flash, I noticed a difference in behavior between agent.invoke() and agent.streamEvents().

Tool calling works correctly in the non-streaming flow (agent.invoke()), but in the streaming flow (agent.streamEvents()), the generated tool call does not include a tool_call.id. After the tool executes, the model returns only a newline ("\n") instead of generating the expected final assistant response.

Using the exact same code with ChatOpenRouter (Gemini 2.5 Flash) works correctly and includes a tool call ID.

Environment

• DeepAgents: Latest
• @langchain/google-genai
• LangChain
• Model: gemini-2.5-flash

Minimal Reproduction

import { ChatGoogleGenerativeAI } from "@langchain/google-genai";
import { createDeepAgent } from "deepagents";
import { tool } from "langchain/tools";
import { z } from "zod";

const model = new ChatGoogleGenerativeAI({
  model: "gemini-2.5-flash",
  apiKey: process.env.GOOGLE_API_KEY,
});

const echoTool = tool(
  ({ message }) => `Echo: ${message}`,
  {
    name: "echo",
    description: "Echoes back whatever message you give it.",
    schema: z.object({
      message: z.string(),
    }),
  }
);

const agent = createDeepAgent({
  model,
  tools: [echoTool],
  systemPrompt:
    "You are a helpful assistant. Always use the echo tool.",
});

const run = await agent.streamEvents(
  {
    messages: [
      {
        role: "user",
        content: "hi how are you",
      },
    ],
  },
  { version: "v3" }
);

Observed Behavior

The first streamed AI message contains a tool call without a tool_call.id.

{
  "tool_calls": [
    {
      "type": "tool_call",
      "name": "echo",
      "args": {
        "message": "Hello! I am doing well, thank you for asking."
      }
    }
  ]
}

After the tool executes, the next streamed assistant message is:

{
  "content": [
    {
      "type": "text",
      "text": "\n"
    }
  ],
  "tool_calls": []
}

As a result, no final assistant response is streamed.

Expected Behavior

The streamed tool call should include a valid tool_call.id, for example:

{
  "tool_calls": [
    {
      "id": "call_xxxxx",
      "type": "tool_call",
      "name": "echo",
      "args": {
        "message": "Hello! I am doing well, thank you for asking."
      }
    }
  ]
}

After the tool executes, the model should stream the final assistant response.

[syncsiva]

langchain-ai/langchainjs#11258

[syncsiva]

langchain-ai/langchainjs#11262