google-adk-python

Build AI agents with Google's Agent Development Kit (ADK) Python - an open-source toolkit for building, evaluating, and deploying AI agents. Features LlmAgent, workflow agents (sequential, parallel, loop), tool integration, multi-agent systems, and deployment to Vertex AI or Cloud Run.

⭐2stars🔀1forks📁View on GitHub🕐Updated Jan 5, 2026⬇️1downloads

When & Why to Use This Skill

The Google ADK (Agent Development Kit) for Python is a powerful open-source framework designed for building, evaluating, and deploying sophisticated AI agents. It provides a code-first approach to agentic workflows, supporting multi-agent systems, complex execution patterns (sequential, parallel, and loops), and seamless integration with Google Search and custom tools. Optimized for the Gemini model family, it simplifies the path from local prototyping to production deployment on Vertex AI and Cloud Run.

Use Cases

  • Multi-Agent System Design: Creating hierarchical agent structures where a coordinator delegates specialized tasks to researcher and writer agents for complex content creation.
  • Automated Research Workflows: Implementing sequential or parallel agents to perform web searches, analyze data, and generate comprehensive reports automatically.
  • Enterprise Tool Integration: Connecting LLMs to internal APIs and custom Python functions with human-in-the-loop approval flows for sensitive business operations.
  • Production AI Deployment: Scaling agentic applications using Vertex AI Agent Engine or Cloud Run to ensure robust, enterprise-grade performance and observability.
namegoogle-adk-python
descriptionBuild AI agents with Google's Agent Development Kit (ADK) Python - an open-source toolkit for building, evaluating, and deploying AI agents. Features LlmAgent, workflow agents (sequential, parallel, loop), tool integration, multi-agent systems, and deployment to Vertex AI or Cloud Run.
licenseMIT

Google ADK Python

Open-source, code-first toolkit for building, evaluating, and deploying AI agents.

When to Use

  • Build AI agents with tool integration
  • Create multi-agent systems with hierarchical coordination
  • Implement workflow agents (sequential, parallel, loop)
  • Integrate with Google Search, Code Execution, or custom tools
  • Deploy to Vertex AI Agent Engine or Cloud Run
  • Implement human-in-the-loop approval flows

Installation

pip install google-adk

Agent Types

LlmAgent

LLM-powered agents with dynamic routing and adaptive behavior.

from google.adk.agents import LlmAgent
from google.adk.tools import google_search

agent = LlmAgent(
    name="search_assistant",
    model="gemini-2.5-flash",
    instruction="You are a helpful assistant that searches the web.",
    tools=[google_search]
)

SequentialAgent

Execute agents in defined order.

from google.adk.agents import SequentialAgent

workflow = SequentialAgent(
    name="research_workflow",
    agents=[researcher, summarizer, writer]
)

ParallelAgent

Run multiple agents concurrently.

from google.adk.agents import ParallelAgent

parallel = ParallelAgent(
    name="parallel_research",
    agents=[web_researcher, paper_researcher]
)

Multi-Agent System

# Specialized agents
researcher = LlmAgent(
    name="Researcher",
    model="gemini-2.5-flash",
    tools=[google_search]
)

writer = LlmAgent(
    name="Writer",
    model="gemini-2.5-flash",
)

# Coordinator
coordinator = LlmAgent(
    name="Coordinator",
    model="gemini-2.5-flash",
    instruction="Delegate tasks to researcher and writer.",
    sub_agents=[researcher, writer]
)

Custom Tools

from google.adk.tools import Tool

def calculate_sum(a: int, b: int) -> int:
    """Calculate the sum of two numbers."""
    return a + b

sum_tool = Tool.from_function(calculate_sum)

Model Support

  • gemini-2.5-flash (recommended)
  • gemini-2.5-pro
  • gemini-1.5-flash
  • gemini-1.5-pro

Best Practices

  1. Code-first for version control and testing
  2. Create specialized agents for specific domains
  3. Use workflow agents for predictable pipelines
  4. Implement confirmation flows for sensitive operations
  5. Test agents systematically

Resources

Credits

Source: https://github.com/mrgoonie/claudekit-skills