AppNewofficial publisher

Google Agent Development Kit (ADK)

App : A product you use directly

Google's open-source kit for designing multi-agent systems in Python, Java, Go, or TypeScript, with the A2A protocol and direct deployment to Vertex AI.

The right choice if you are on Google Cloud or if your agents need to communicate with other agents via A2A. Gemini by default, other models available via LiteLLM.

Official site GitHub · 21,993 ★ Docs

Overall score

7.8 /10

Security: one caveat, see the card

Utility

8.0

Reliability

10.0

Security

One caveat

Ease

5 min · Technical

Security reflects where data goes, requested permissions and incidents on record at the verification date. It is not a guarantee of zero risk: check the deployment mode and the connections you enable.

What it does, concretely

Building an AI agent capable of chaining complex tasks, delegating to other agents, and integrating with existing tools requires a solid architecture. Google ADK addresses this need by providing a structured framework to design, test, and deploy multi-agent systems without starting from scratch.

In practice, you define your agents in Python (or Java, Go, TypeScript, Kotlin) by assigning them instructions, tools, and a model. You then orchestrate their sequencing through graph-based workflows that handle branching, loops, parallel execution, and error recovery. A local web UI lets you test and debug each agent before any deployment.

ADK targets developers who want to build production-grade agents: technical teams working on Google Cloud, projects requiring agent-to-agent communication via the A2A protocol, or use cases where reliability and observability matter as much as speed of delivery.

However, if you are looking for a no-code tool or a graphical interface for non-developers, ADK is not the right fit: everything is configured through code, and getting started requires proficiency in at least one of the supported languages and familiarity with agent orchestration concepts.

What it offers

·Graph-based workflows: orchestrate agents in sequence, in parallel, or in loops with explicit execution paths and automatic error recovery.
·Modular multi-agent system: compose hierarchies of specialized agents that delegate tasks to each other via a structured API.
·Built-in A2A protocol: agents can communicate with external agents that implement this open protocol.
·Rich tool ecosystem: use pre-built tools (including Google Search), custom Python functions, OpenAPI specs, or MCP tools.
·Local development web UI: test, evaluate, and debug your agents through a built-in interface accessible with the adk web command.
·Tool confirmation (HITL): add a human validation step before executing sensitive tools.
·Flexible deployment: containerize agents with Docker, deploy to Cloud Run or Vertex AI Agent Engine with a single command.
·Context management: ADK filters irrelevant events, summarizes older conversation turns, and lazy-loads artifacts to keep agents efficient.

Our tips

Use adk web during development to visualize agent exchanges and quickly identify steps that fail or produce unexpected results.
Pin a dependency constraints file matching your Python version (constraints-3.x.txt) at install time to avoid transitive package conflicts.
Start with the Agent class for a simple agent, then introduce Workflow only when you need to chain multiple agents: this limits initial complexity.
Use the tool confirmation (HITL) feature for irreversible actions such as sending emails or modifying databases, to keep human control over critical steps.
The adk eval command lets you run automated test sets against your agents: integrate it into your CI pipeline to catch regressions before deployment.

Watch out

ADK is exclusively developer-oriented: there is no no-code graphical interface, and the learning curve can be significant for anyone unfamiliar with agent orchestration.
Usage costs depend on the chosen model and hosting: the kit itself is free, but every model call and every hour on Vertex AI is billed on consumption, which can be surprising without usage monitoring.

In practice

Price: Free
Commercial use: Commercial use: yes
Available on: cli
Licence: Apache-2.0
FR interface: No

Security and data

Rating: Security: one caveat, see the card
Your data: Local or cloud depending on use
Maintenance: Active
Last release: 23 October 2026

Links and repository checked on 25 October 2026 · added on 27 September 2026

How to install Google Agent Development Kit (ADK)

1. Verify that Python 3.10 or higher is installed on your machine. 2. Install ADK via pip: pip install google-adk. 3. To protect transitive dependencies, download the constraints file matching your Python version and reinstall with: pip install google-adk -c constraints-3.x.txt. 4. Configure your API keys (Gemini or another model) in a .env file at the root of your project. 5. Launch the local development UI with adk web path/to/your/agents to test before any deployment. 6. Restrict the permissions of the service account used to the strict minimum to reduce the exposure surface when tools execute.

pip install google-adk

Official documentation ↗

In the same category

Whole category Automation & agents →