Agent Development Kit (ADK)

Agent Development Kit (ADK)

The Agent Development Kit (ADK) is a flexible and modular open-source Python framework designed to apply software development principles to the creation of AI agents. It simplifies the processes of building, deploying, and orchestrating agent workflows, supporting everything from simple tasks to complex multi-agent systems. While optimized for Google's Gemini models, ADK remains model-agnostic, deployment-agnostic, and compatible with other frameworks, making it versatile for various AI development needs.

Key Features

• Rich Tool Ecosystem: Leverage pre-built tools, custom functions, OpenAPI specifications, MCP tools, or integrate existing tools to empower agents with diverse capabilities. It offers tight integration with the Google ecosystem for enhanced functionality.

• Code-First Development: Define agent logic, tools, and orchestration directly in Python code, providing ultimate flexibility, testability, and versioning control.

• Agent Config: Allow for agent building without writing code through the intuitive Agent Config feature. For more details, refer to the documentation.

• Tool Confirmation (HITL): Implement a human-in-the-loop (HITL) tool confirmation flow to safeguard tool executions with explicit user confirmations and custom inputs. See the guide.

• Modular Multi-Agent Systems: Compose scalable applications by designing hierarchies of specialized agents, enabling complex interactions and task delegation.

• Deploy Anywhere: Containerize agents for easy deployment on Cloud Run or scale them seamlessly using Vertex AI Agent Engine.

Recent Updates

Recent enhancements include:

• Custom Service Registration: A service registry for registering custom implementations in the FastAPI server (commit: 391628f).
• Rewind Functionality: Ability to rewind sessions to previous states before invocations (commit: 9dce06f).
• New CodeExecutor: AgentEngineSandboxCodeExecutor for executing agent-generated code via the Vertex AI Code Execution Sandbox API (commit: ee39a89).

Installation

Stable Release

Install via pip:

pip install google-adk

Releases occur bi-weekly for stability.

Development Version

For cutting-edge features:

pip install git+https://github.com/google/adk-python.git@main

Integrations and Examples

ADK integrates with the Agent2Agent (A2A) protocol for remote multi-agent communication—check the A2A repo and samples.

Single Agent Example

from google.adk.agents import Agent
from google.adk.tools import google_search
 
root_agent = Agent(
    name="search_assistant",
    model="gemini-2.5-flash",
    instruction="You are a helpful assistant. Answer user questions using Google Search when needed.",
    description="An assistant that can search the web.",
    tools=[google_search]
)

Multi-Agent System Example

Define agents like greeter, task_executor, and a coordinator with sub-agents for collaborative task handling.

Development UI and Evaluation

A built-in UI aids in testing, evaluating, debugging, and showcasing agents. Evaluate using:

adk eval samples_for_testing/hello_world samples_for_testing/hello_world/hello_world_eval_set_001.evalset.json

Community and Contributions

Join the community via Reddit or Google Group. The first community meeting was on October 15, 2025. Contributions are welcome—see guidelines and the community repo. Licensed under Apache 2.0.