Build a Search-Grounded Agent with Google Vertex AI Agents

1. 1
   Set Up Your GCP Environment

   First, enable the Vertex AI API and authenticate your local environment. This allows the Python SDK to interact with your Google Cloud project. Replace `[YOUR_PROJECT_ID]` with your actual project ID.
2. 2
   Define the Agent and its Tools

   In Python, define the agent's configuration. This includes a display name, core instructions, and the tools it can use. We will equip it with the `GoogleSearchRetriever` tool to ground its responses in search results.
3. 3
   Create the Agent Resource

   Use the Vertex AI client to send your agent definition to Google Cloud, creating a persistent agent resource. This makes the agent available for interaction.
4. 4
   Chat with Your Agent

   Now that the agent is live, you can send it a query. The agent will use Google Search to find relevant information and then generate a grounded response.
5. 5
   Clean Up Resources

   To avoid incurring ongoing costs, delete the agent resource from your project after you're done experimenting.

import os
from google.cloud import aiplatform_v1beta1 as aiplatform

# --- Configuration ---
# Replace with your Project ID and a valid GCP location (e.g., 'us-central1')
PROJECT_ID = "your-gcp-project-id"
LOCATION = "us-central1"

# Ensure you have authenticated via `gcloud auth application-default login`

def main():
    """Main function to create, chat with, and delete a Vertex AI Agent."""
    if PROJECT_ID == "your-gcp-project-id":
        print("Error: Please update PROJECT_ID with your Google Cloud Project ID.")
        return

    api_endpoint = f"{LOCATION}-aiplatform.googleapis.com"
    client_options = {"api_endpoint": api_endpoint}
    client = aiplatform.AgentServiceClient(client_options=client_options)

    # 1. Define Agent Configuration
    search_tool = aiplatform.Tool(
        google_search_retriever=aiplatform.GoogleSearchRetriever()
    )
    agent_config = aiplatform.Agent(
        display_name="My Search Agent",
        instruction="You are a helpful assistant that answers questions using the provided search tool.",
        tools=[search_tool],
        default_model="gemini-1.5-pro-001"
    )

    # 2. Create the Agent
    parent = f"projects/{PROJECT_ID}/locations/{LOCATION}"
    print(f"Creating agent in {parent}...")
    try:
        create_response = client.create_agent(parent=parent, agent=agent_config)
        agent_resource_name = create_response.name
        print(f"Successfully created agent: {agent_resource_name}")

        # 3. Chat with the Agent
        user_query = "What were the key announcements at the last Google I/O?"
        print(f"\nQuerying agent with: '{user_query}'")
        chat_response = client.chat_agent(
            name=agent_resource_name,
            user_input=user_query
        )
        print("\nAgent Response:")
        print(chat_response.content)

    except Exception as e:
        print(f"An error occurred: {e}")
        return
    finally:
        # 4. Clean up the agent resource
        if 'agent_resource_name' in locals():
            print(f"\nDeleting agent: {agent_resource_name}")
            client.delete_agent(name=agent_resource_name)
            print("Cleanup complete.")

if __name__ == "__main__":
    main()