google/adk-java

Agent Development Kit (ADK) for Java – Quick Overview

What it is: A Google‑maintained, open‑source Java library that lets you write, test, and ship AI agents directly in code. It mirrors the functionality of the Python ADK but is built for the Java ecosystem and integrates tightly with Google Cloud services.

Key capabilities

• Tool ecosystem – Plug in pre‑built tools (e.g., GoogleSearchTool), custom functions, or OpenAPI‑described services so agents can act on the web, databases, or other APIs.
• Code‑first design – Define an agent’s name, description, model, system prompt, and attached tools in Java builder syntax, giving you full version‑control and testability.
• Multi‑agent composition – Combine several specialized agents into hierarchical systems for larger applications.
• Development UI – A built‑in web UI (similar to the Python version) for interactive debugging, evaluation, and showcase of agents.
• A2A integration – Supports remote agent‑to‑agent communication via Google’s A2A protocol.

Typical usage pattern

LlmAgent assistant = LlmAgent.builder()
    .name("search_assistant")
    .description("An assistant that can search the web.")
    .model("gemini-2.0-flash")
    .instruction("You are a helpful assistant. Answer user questions using Google Search when needed.")
    .tools(new GoogleSearchTool())
    .build();

You then run the agent locally or deploy it to Google Cloud, using the same Maven coordinates (com.google.adk:google-adk) that appear on Maven Central.

Getting started

1. Add the google-adk (and optionally google-adk-dev for the UI) dependencies to your pom.xml.
2. Follow the online docs (https://google.github.io/adk-docs) and sample repo (google/adk-samples) for concrete examples.
3. Use the development UI to iterate on prompts, tool calls, and agent logic.

Who it’s for

• Java developers building chat‑style assistants, retrieval‑augmented agents, or orchestrations of multiple LLM‑powered services.
• Teams that need tight integration with Google Cloud (e.g., Vertex AI models, Cloud Search) while keeping the entire stack in Java.

Status The library is in a pre‑GA (pre‑general‑availability) stage – functional but labeled “as‑is” with limited support.

---

Happy Agent Building!