Google Agent Development Kit (ADK) is an open-source, code-first Python framework for building, evaluating, and deploying AI agents. With 18.6K GitHub stars, ADK is Google's official agent framework — optimized for Gemini models but fully model-agnostic, supporting deployment from local development to Google Cloud at scale.

framework python google gemini agents multi-agent cloud

Google ADK was released as part of Google's push to make agent development accessible through software engineering best practices rather than prompt-centric approaches. The framework emphasizes a code-first philosophy where agents are defined in Python with clear structure, tools are regular functions, and deployment scales from adk run on a laptop to Vertex AI Agent Engine in production. ADK 2.0 Alpha introduced graph-based orchestration for complex multi-agent workflows, signaling Google's commitment to the framework as a first-class development platform.

• Code-First Development — Agents defined in Python (agent.py) with tools as regular functions; also supports YAML config
• Gemini-Optimized — Default models like gemini-3-flash-preview with native thinking capabilities
• Model-Agnostic — Configurable authentication for any LLM provider
• Multi-Agent Orchestration — Hierarchical agents, agents-as-tools, graph-based workflows (ADK 2.0 Alpha)
• Rich Tool Ecosystem — Search, Code Execution, BigQuery forecast, GKECodeExecutor, OAuth flows
• ADK Web — Open-source developer UI for interactive agent testing and debugging
• CLI Tools — adk create, adk run, adk web for project lifecycle management
• Google Cloud Integration — Vertex AI Agent Engine, Cloud Run, GKE deployment
• ADK Skills — AI-powered coding assistance in development environments

ADK follows a modular project structure with clear separation of concerns:

Building an agent with Google ADK:

from google.adk.agents.llm_agent import Agent
from google.adk.runners import Runner
from google.adk.sessions import InMemorySessionService
 
# Define a tool as a regular Python function
def get_current_weather(city: str) -> dict:
    """Get the current weather for a city."""
    # Real API call would go here
    return {
        "city": city,
        "temperature": "22C",
        "condition": "Sunny",
        "humidity": "45%"
    }
 
def search_news(topic: str) -> dict:
    """Search for recent news on a topic."""
    return {
        "topic": topic,
        "articles": ["Article 1", "Article 2", "Article 3"]
    }
 
# Define the agent
root_agent = Agent(
    model="gemini-2.0-flash",
    name="assistant",
    description="A helpful assistant with weather and news tools.",
    instruction=(
        "You are a helpful assistant. Use available tools to "
        "answer questions about weather and current events."
    ),
    tools=[get_current_weather, search_news],
)
 
# Run interactively
session_service = InMemorySessionService()
runner = Runner(agent=root_agent, app_name="my_app", session_service=session_service)

ADK integrates deeply with Google Cloud services:

• ADK 2.0 Alpha — Graph-based workflows for advanced multi-agent orchestration
• Agent Config — YAML-based agent authoring with CLI deployment to Agent Engine
• ADK Web — Open-sourced developer UI for interactive debugging
• ADK Skills — AI-powered coding tools for development environments
• Tool Enhancements — BigQuery forecast, GKECodeExecutor, tool confirmation flows, OAuth
• Multi-language — Java, Go, and npm support alongside Python
• Samples Repository — 54+ contributors with agents for RAG, financial advisor, travel concierge

• GitHub Repository (Python)
• Documentation
• Samples Repository
• ADK Web UI

• LangChain — General-purpose LLM framework
• PydanticAI — Type-safe agent framework (uses Pydantic, like ADK)
• Agno — High-performance agent runtime
• Haystack — Pipeline-based AI orchestration