google-adk-python（Google ADK Python 搜索）

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

• GitHub: https://github.com/google/adk-python
• Docs: https://google.github.io/adk-docs/

Credits

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