6.6 KiB
Google ADK Python Skill
You are an expert guide for Google's Agent Development Kit (ADK) Python - an open-source, code-first toolkit for building, evaluating, and deploying AI agents.
When to Use This Skill
Use this skill when users need to:
- Build AI agents with tool integration and orchestration capabilities
- Create multi-agent systems with hierarchical coordination
- Implement workflow agents (sequential, parallel, loop) for predictable pipelines
- Integrate LLM-powered agents with Google Search, Code Execution, or custom tools
- Deploy agents to Vertex AI Agent Engine, Cloud Run, or custom infrastructure
- Evaluate and test agent performance systematically
- Implement human-in-the-loop approval flows for tool execution
Core Concepts
Agent Types
LlmAgent: LLM-powered agents capable of dynamic routing and adaptive behavior
- Define with name, model, instruction, description, and tools
- Supports sub-agents for delegation and coordination
- Intelligent decision-making based on context
Workflow Agents: Structured, predictable orchestration patterns
- SequentialAgent: Execute agents in defined order
- ParallelAgent: Run multiple agents concurrently
- LoopAgent: Repeat execution with iteration logic
BaseAgent: Foundation for custom agent implementations
Key Components
Tools Ecosystem:
- Pre-built tools (google_search, code_execution)
- Custom Python functions as tools
- OpenAPI specification integration
- Tool confirmation flows for human approval
Multi-Agent Architecture:
- Hierarchical agent composition
- Specialized agents for specific domains
- Coordinator agents for delegation
Installation
# Stable release (recommended)
pip install google-adk
# Development version (latest features)
pip install git+https://github.com/google/adk-python.git@main
Implementation Patterns
Single Agent with Tools
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 for information.",
description="Search assistant for web queries",
tools=[google_search]
)
Multi-Agent System
from google.adk.agents import LlmAgent
# Specialized agents
researcher = LlmAgent(
name="Researcher",
model="gemini-2.5-flash",
instruction="Research topics thoroughly using web search.",
tools=[google_search]
)
writer = LlmAgent(
name="Writer",
model="gemini-2.5-flash",
instruction="Write clear, engaging content based on research.",
)
# Coordinator agent
coordinator = LlmAgent(
name="Coordinator",
model="gemini-2.5-flash",
instruction="Delegate tasks to researcher and writer agents.",
sub_agents=[researcher, writer]
)
Custom Tool Creation
from google.adk.tools import Tool
def calculate_sum(a: int, b: int) -> int:
"""Calculate the sum of two numbers."""
return a + b
# Convert function to tool
sum_tool = Tool.from_function(calculate_sum)
agent = LlmAgent(
name="calculator",
model="gemini-2.5-flash",
tools=[sum_tool]
)
Sequential Workflow
from google.adk.agents import SequentialAgent
workflow = SequentialAgent(
name="research_workflow",
agents=[researcher, summarizer, writer]
)
Parallel Workflow
from google.adk.agents import ParallelAgent
parallel_research = ParallelAgent(
name="parallel_research",
agents=[web_researcher, paper_researcher, expert_researcher]
)
Human-in-the-Loop
from google.adk.tools import google_search
# Tool with confirmation required
agent = LlmAgent(
name="careful_searcher",
model="gemini-2.5-flash",
tools=[google_search],
tool_confirmation=True # Requires approval before execution
)
Deployment Options
Cloud Run Deployment
# Containerize agent
docker build -t my-agent .
# Deploy to Cloud Run
gcloud run deploy my-agent --image my-agent
Vertex AI Agent Engine
# Deploy to Vertex AI for scalable agent hosting
# Integrates with Google Cloud's managed infrastructure
Custom Infrastructure
# Run agents locally or on custom servers
# Full control over deployment environment
Model Support
Optimized for Gemini:
- gemini-2.5-flash
- gemini-2.5-pro
- gemini-1.5-flash
- gemini-1.5-pro
Model Agnostic: While optimized for Gemini, ADK supports other LLM providers through standard APIs.
Best Practices
- Code-First Philosophy: Define agents in Python for version control, testing, and flexibility
- Modular Design: Create specialized agents for specific domains, compose into systems
- Tool Integration: Leverage pre-built tools, extend with custom functions
- Evaluation: Test agents systematically against test cases
- Safety: Implement confirmation flows for sensitive operations
- Hierarchical Structure: Use coordinator agents for complex multi-agent workflows
- Workflow Selection: Choose workflow agents for predictable pipelines, LLM agents for dynamic routing
Common Use Cases
- Research Assistants: Web search + summarization + report generation
- Code Assistants: Code execution + documentation + debugging
- Customer Support: Query routing + knowledge base + escalation
- Content Creation: Research + writing + editing pipelines
- Data Analysis: Data fetching + processing + visualization
- Task Automation: Multi-step workflows with conditional logic
Development UI
ADK includes built-in interface for:
- Testing agent behavior interactively
- Debugging tool calls and responses
- Evaluating agent performance
- Iterating on agent design
Resources
- GitHub: https://github.com/google/adk-python
- Documentation: https://google.github.io/adk-docs/
- llms.txt: https://raw.githubusercontent.com/google/adk-python/refs/heads/main/llms.txt
Implementation Workflow
When implementing ADK-based agents:
- Define Requirements: Identify agent capabilities and tools needed
- Choose Architecture: Single agent, multi-agent, or workflow-based
- Select Tools: Pre-built, custom functions, or OpenAPI integrations
- Implement Agents: Create agent definitions with instructions and tools
- Test Locally: Use development UI for iteration
- Add Evaluation: Create test cases for systematic validation
- Deploy: Choose Cloud Run, Vertex AI, or custom infrastructure
- Monitor: Track agent performance and iterate
Remember: ADK treats agent development like traditional software engineering - use version control, write tests, and follow engineering best practices.