Framework · Google
Google ADK at a glance
Google's Agent Development Kit, with SDKs for Python, TypeScript, Go, Java and Kotlin and evaluation built into the core. It is Gemini-first but reaches other models through adapters, and deploys to Agent Runtime on Google Cloud.
Best for: Teams working across several languages that want Gemini-first agents, built-in evals and a direct route to Google Cloud.
Key facts
- Maintainer
- Languages
- Python, TypeScript, Go, Java, Kotlin
- Licence
- Apache-2.0
- Latest version
Python 2.10.0· September 25, 2026- Repository
- google/adk-python
- Hosted option
- Agent Runtime on Gemini Enterprise Agent Platform (formerly Vertex AI Agent Engine); also Cloud Run, GKE
Quickstart
Install the package, set the credentials, then run the smallest agent from the official docs. Model IDs are the ones the docs use and may not be enabled on your account.
Python
pip install google-adk && adk create my_agent (run with: adk run my_agent or adk web --port 8000)from google.adk.agents.llm_agent import Agent
# Mock tool implementation
def get_current_time(city: str) -> dict:
"""Returns the current time in a specified city."""
return {"status": "success", "city": city, "time": "10:30 AM"}
root_agent = Agent(
model='gemini-flash-latest',
name='root_agent',
description="Tells the current time in a specified city.",
instruction="You are a helpful assistant that tells the current time in cities. Use the 'get_current_time' tool for this purpose.",
tools=[get_current_time],
)TypeScript
npm install @google/adk (inferred from package name; not copied from docs)import { LlmAgent, GOOGLE_SEARCH } from '@google/adk';
const agent = new LlmAgent({
name: 'researcher',
model: 'gemini-flash-latest',
instruction: 'You help users research topics thoroughly.',
tools: [GOOGLE_SEARCH],
});Features
| Any model | Yes | Gemini first; other models through LiteLLM and local adapters |
|---|---|---|
| MCP client | Yes | McpToolset, with optional confirmation per tool |
| MCP server | Yes | Expose an agent with to_mcp_server() |
| Multi-agent | Yes | Sub-agent trees, plus graph workflows since ADK 2.0 |
| Durable runs | Yes | Resume stopped agents; Agent Runtime keeps long-running state |
| Human approval | Yes | Tool Confirmation via require_confirmation=True |
| Tracing | Yes | OpenTelemetry-based, with Cloud Trace on Google Cloud |
| Evals | Yes | Evaluation is a core ADK feature |
Recent changes
- ADK 2.0 added graph-based workflows.
- A Kotlin SDK joined Python, TypeScript, Go and Java.
- The docs moved to adk.dev.
- Vertex AI Agent Engine was renamed Agent Runtime on Gemini Enterprise Agent Platform.