vertexai: updating aiplatform

[lgesuellip]

PR Description

Hey team,

The current version of aiplatform doesn’t support Google Search via the v1 endpoint. This PR updates the library version to enable that support through tools.

https://github.com/googleapis/python-aiplatform/blob/v1.86.0/google/cloud/aiplatform_v1/types/tool.py#L78

🆕 New Feature
✅ Test

Testing

I tested it using the following code:

import asyncio
import google.cloud.aiplatform_v1beta1.types as vertex_types
from langchain_google_vertexai import ChatVertexAI
from langchain_core.messages import SystemMessage, HumanMessage


async def company_research_with_search_grounding(ticker: str, cik: str) -> str:
    """Research a company using Gemini with Google Search grounding.
    
    Args:
        ticker: The stock ticker symbol
        cik: The SEC's Central Index Key number
        
    Returns:
        Detailed company information including name and domain
    """
    # Initialize the chat model
    llm = ChatVertexAI(
        project="",
        location="us-central1",
        model="gemini-2.0-flash-exp",
        endpoint_version="v1"
    )

    # System message to guide the model's behavior
    messages = [
        SystemMessage(content="You are a helpful AI assistant specializing in Company Research."),
    ]
    
    # Human query with structured data
    human_message = HumanMessage(
        content=f"""Given the following ticker from US market and the CIK of the company, 
        please provide:
        1. The company's full legal name
        2. The company's primary domain/website
        3. A brief description of their business
        
        ticker: {ticker}
        cik: {cik}
        """
    )
    
    # Create a Tool with Google Search
    google_search_tool = vertex_types.Tool(google_search={})

    # Invoke the model with the search tool
    response = await llm.ainvoke(
        messages + [human_message],
        tools=[google_search_tool],
    )
    
    return response


async def main():
    # Example usage
    result = await company_research_with_search_grounding("AAPL", "320193")
    print(result)
    
    # You can easily research multiple companies asynchronously
    companies = [
        ("AAPL", "320193"),
    ]
    
    results = await asyncio.gather(
        *[company_research_with_search_grounding(ticker, cik) for ticker, cik in companies]
    )
    
    for ticker, result in zip([c[0] for c in companies], results):
        print(f"\n--- {ticker} Research Results ---")
        print(result.content)


if __name__ == "__main__":
    asyncio.run(main())

Note

I think that in a future PR, we should also adapt the types in function_utils to support both versions.
In this case, though, it seems to work fine using the beta type (google.cloud.aiplatform_v1beta1.types as vertex_types) with the v1 endpoint.

What do you think? @lkuligin

Thanks,