What it is
ADK (Agent Development Kit) is Google's code-first framework for building agents. It is optimised for Gemini but model-agnostic, and agents can run locally, in a container on Cloud Run or GKE, or on Vertex AI Agent Engine, Google's managed agent hosting.
Version 2.0 for Python became generally available on 19 May 2026. It added a workflow runtime that treats agents, tools and functions as nodes in a graph, with routing, fan-out and fan-in, loops, retries and human-in-the-loop pauses, plus a Task API for structured delegation between agents. Google says 2.0 is designed to stay compatible with 1.x agents.
Besides Python there are separate SDKs for Java, Kotlin, Go and TypeScript. Agents can also be defined in YAML without code, and a local web UI is provided for testing.
Our review
ADK is a strong choice for teams already committed to Gemini, and a reasonable but not obvious one for everyone else. The Python package has about 22,000 GitHub stars, shipped v2.12.0 on 13 October 2026 and, per its README, releases roughly every two weeks.
The 2.0 workflow runtime is what moves it into LangGraph's territory. Branching, loops, retries and tool confirmation are built in, and the Task API gives agents a structured way to hand work to one another. Evaluation tooling and the local dev UI are included rather than sold separately, which is a real difference from LangChain, where the comfortable debugging lives in paid LangSmith.
The pace has costs. With 392 open issues and pull requests and a release every fortnight, examples written for 1.x often no longer match the docs, and the README recommends pinning installs with Google's constraints files to avoid transitive dependency breakage. Python 3.11 or newer is required. Other providers work through model adapters, but the quickstart, the built-in Google Search and Vertex tools, and the one-command deploy targets are all Google's.
No security advisories have been published against the repository so far. The framework is free, so the real bill is Gemini tokens and whatever compute runs the agent; Agent Engine adds managed sessions and memory at Google Cloud usage rates, which is convenient but makes leaving Google harder than leaving the framework.
Where it earns its place
A platform team at a retailer already on Google Cloud
Ship a Gemini-based order-support agent with human approval on refunds
Build the flow as an ADK workflow with tool confirmation on the refund step, test it in the dev UI, then deploy to Cloud Run or Agent Engine. The framework is free; cost is Gemini tokens plus Cloud Run or Agent Engine compute.
A Java shop with existing Spring services
Add agent features without introducing a Python service
The separate ADK Java SDK lets the team stay in one language. Feature parity with Python lags, so check that the workflow features they need have reached the Java SDK before committing.
An engineering manager comparing agent frameworks for a 12-month roadmap
Pick a framework that will not need replacing next year
ADK is backed by Google, ships every two weeks and includes evaluation at no cost, but much of its convenience assumes Gemini and Vertex AI. That is a sensible bet for a Google Cloud customer and a lock-in risk for a multi-cloud one.
Before you commit
- Setup
- pip install google-adk on Python 3.11 or newer, ideally with the constraints file for your Python version; the quickstart reaches a working Gemini agent and the local web UI in a few minutes with an API key.
- Integrations
- Gemini API and Vertex AI natively, other models through adapters; tools from Python functions, OpenAPI specs and MCP servers; deploys to Cloud Run, GKE or Vertex AI Agent Engine.
- Data and privacy
- Apache 2.0 with no commercial limits. Prompts and tool results go to whichever model endpoint you configure, and sessions are stored wherever your session service points (memory, a database, or Agent Engine). No GitHub security advisories published.
- Cost at scale
- The framework is free; production cost is model tokens plus compute. Agent Engine is billed as Google Cloud usage on the Vertex AI pricing page, so a busy agent there is a cloud line item rather than a licence.
- Limits
- 392 open issues and pull requests; releases roughly every two weeks; non-Python SDKs trail the Python feature set.
What's good
- Graph workflows, retries and human approval built in since 2.0
- Evaluation tools and dev UI included free
- SDKs for Python, Java, Kotlin, Go and TypeScript
- Short path to Cloud Run or Agent Engine
What's not
- Best-supported tools and deploy targets are Google products
- Fortnightly releases leave 1.x tutorials out of date
- Requires Python 3.11+ and pinned constraints for stable installs
Best for
Teams building agents on Gemini or deploying to Google Cloud.
Alternatives
Google Agent Development Kit (ADK): common questions
- Is Google Agent Development Kit (ADK) worth it?
- A serious, fast-moving agent framework whose shortest road always leads to Gemini and Google Cloud, which is either the point or the catch. We rate it 4.2 out of 5.
- How much does Google Agent Development Kit (ADK) cost?
- Free and open source under the Apache 2.0 licence. You pay for the models and Google Cloud services you use.
- What are the downsides of Google Agent Development Kit (ADK)?
- Best-supported tools and deploy targets are Google products; fortnightly releases leave 1.x tutorials out of date; requires Python 3.11+ and pinned constraints for stable installs.
- Who is Google Agent Development Kit (ADK) best for?
- Teams building agents on Gemini or deploying to Google Cloud.
- How much does Google Agent Development Kit (ADK) cost for a team or in production?
- The framework is free; production cost is model tokens plus compute. Agent Engine is billed as Google Cloud usage on the Vertex AI pricing page, so a busy agent there is a cloud line item rather than a licence.
- Is Google Agent Development Kit (ADK) safe for company data?
- Apache 2.0 with no commercial limits. Prompts and tool results go to whichever model endpoint you configure, and sessions are stored wherever your session service points (memory, a database, or Agent Engine). No GitHub security advisories published.
- What are the best Google Agent Development Kit (ADK) alternatives?
- The closest alternatives we have reviewed are LangGraph, OpenAI Agents SDK and Microsoft Agent Framework.
Answers checked Oct 23, 2026.
Make Google Agent Development Kit (ADK)? Show this rating on your site
