Base Agent

A base agent implementation using Google ADK, designed to be easily extended with sub-agents and tools. This package provides a template for building advanced AI agents that can interact with users, delegate tasks to sub-agents, and utilize custom tools.

Features

• Google ADK Integration: Leverages the LlmAgent from Google ADK for powerful language model capabilities.
• Sub-Agent Support: Easily add and manage sub-agents for modular task delegation.
• Custom Tools: Integrate Python functions as tools for the agent to use.
• Vertex AI Ready: Example deployment script for running the agent on Google Vertex AI.

Directory Structure

base-agent/
├── base_agent/
│   ├── __init__.py
│   ├── agent.py           # Main agent definition
│   ├── tools/
│   │   └── hello_world_tool.py  # Example tool
│   └── sub_agents/
│       └── sub_agent/
│           └── agent.py   # Example sub-agent
├── deployment/
│   ├── deploy.py          # Deployment script
│   └── run.py             # Example run script
├── pyproject.toml         # Poetry project file
└── README.md

Installation

1. Clone the repository:

   git clone <repo-url>
   cd tag-agents/base-agent

2. Install dependencies with Poetry:

   poetry install

   Requires Python 3.11 or higher.

Usage

The main agent is defined in base_agent/agent.py:

from google.adk.agents import LlmAgent
from .tools.hello_world_tool import hello_world_tool
from .sub_agents.sub_agent.agent import sub_agent

root_agent = LlmAgent(
    name="root_agent",
    model='gemini-2.5-flash',
    description="You are a helpful assistant that can answer questions and help with tasks.",
    instruction="You are a helpful assistant that can answer questions and help with tasks.",
    sub_agents=[sub_agent],
    tools=[hello_world_tool],
)

Example Tool

hello_world_tool.py:

def hello_world_tool(name: str) -> str:
    """
    This tool prints "Hello, {name}!"
    """
    return f"Hello, {name}!"

Example Sub-Agent

sub_agent/agent.py:

from google.adk.agents import LlmAgent

sub_agent = LlmAgent(
    name="sub_agent",
    model='gemini-2.5-flash',
    description="You are a helpful assistant that can answer questions and help with tasks.",
    instruction="You are a helpful assistant that can answer questions and help with tasks.",
    output_key="sub_agent_output",
)

Deployment

1. Configure Environment Variables:

   • Edit the .env file and replace the environment variables with your GCP project values:
     • GOOGLE_CLOUD_PROJECT
     • GOOGLE_CLOUD_LOCATION
     • GOOGLE_API_KEY
     • STAGING_BUCKET
   • Note: AGENT_ENGINE_ID is set automatically when the agent is deployed; you do not need to set it manually.

2. Deploy the Agent to GCP Agent Engine:

   python deployment/deploy.py

3. Test the Deployed Agent:

   • You can run a test to make sure the deployment works by calling the deployed agent and asking something:

   python deployment/run.py

Customization

• Add new tools: Place Python functions in base_agent/tools/ and add them to the agent's tools list.
• Add sub-agents: Define new sub-agents in base_agent/sub_agents/ and include them in the main agent's sub_agents list.

Author

Damian Martinez Carmona (damian.martinez.carmona@gmail.com)

License

This project is licensed under the MIT License. See the LICENSE file for details.