Google ADK
What is Google ADK?
Google ADK is an agent development framework for developers who need to build, run, and extend production-grade agents. It combines Build Agents, Run Agents, Components, Visual Builder, and Evaluation, with graph-based workflows, collaborative agents, and dynamic workflows for multi-step orchestration. It works with Gemini, Gemma, Claude, Ollama, vLLM, LiteLLM, LiteRT-LM, Google Search, and Google Cloud, and Google Cloud is a named user.
Last verifiedHow we evaluate
At a glance
- Google ADK is best for developers who need to ship production agents with flexible workflows and model choice.
What does Google ADK do?
ADK handles agent development by giving you a structured framework for building, running, and extending production-grade agents. It combines Build Agents, Run Agents, Components, and a Visual Builder so teams can start quickly and then add complexity as needed. The framework also supports graph-based workflows, collaborative agents, and dynamic workflows, which makes it easier to orchestrate multi-step tasks without turning the codebase into a pile of ad hoc logic. Behind the scenes, ADK is designed for scale and flexibility: it can work with almost any generative model, including Gemini, Claude, Ollama, vLLM, LiteLLM, and locally running models, and it can deploy on your own infrastructure or on Google Cloud through Agent Runtime, Cloud Run, or GKE. The docs describe support for 4 languages, a 2026 release cycle, and production deployment that adds managed infrastructure, authentication, observability, and security without changing agent code. Google Cloud is a named user of the framework.
Why use Google ADK?
- It supports deploy-anywhere setups, so teams can run agents on their own infrastructure or on Google Cloud without rewriting code.
- It works with many model providers, so buyers can avoid locking agent logic to a single model stack.
- Production deployment can inherit managed infrastructure, authentication, observability, and security from Google Cloud.
- The docs show tools for interacting with agents and AI-assisted coding, which can shorten the path from prototype to working system.
Who is Google ADK for?
- Platform engineers who need a framework that can move from local development to production deployment.
- AI application developers who want structured agent workflows without building orchestration from scratch.
- Teams building multi-agent systems who need collaborative and dynamic task handling.
- Developers who want to connect agents to multiple model providers and local models.
- Organizations using Google Cloud who want agents that can deploy into managed infrastructure.
What are Google ADK's key features?
Build Agents
Create production agents with the ADK framework and Gemini, then package them for deployment on Google Cloud instead of one-off demos.
Run Agents
Execute agents in the Agent Runtime so they can handle real requests consistently, with runtime behavior designed for production use on Google Cloud.
Components
Assemble agents from reusable components, which keeps prompts, tools, and logic easier to maintain as workflows grow beyond a single script.
Integrations
Connect agents to Gemini, Gemma, Claude, Ollama, vLLM, LiteLLM, LiteRT-LM, Google Search, and Google Cloud for model choice and external data access.
Evaluation
Test agent behavior with evaluation workflows so teams can measure outputs before release and catch regressions in production-facing agent systems.
Graph-based workflows
Model agent steps as graph-based workflows, making branching logic and multi-step execution easier to trace than linear prompt chains.
Collaborative agents
Coordinate multiple agents in shared workflows, which helps split tasks across specialized roles and manage more complex production processes.
Dynamic workflows
Build workflows that change at runtime, letting agents adapt steps based on inputs, tool results, or model decisions without rewriting the whole flow.
What does Google ADK integrate with?
- Gemini
- Gemma
- Claude
- Ollama
- vLLM
- LiteLLM
- LiteRT-LM
- Google Search
- Google Cloud
What are Google ADK's use cases?
Platform engineers ship agents
Platform engineers who need a framework that can move from local development to production deployment use Google ADK to build and launch agents with Build Agents and Agent Runtime. They can validate behavior with Evaluation before handing off to managed infrastructure, reducing the gap between a prototype and a production-ready service.
Multi-agent task coordination
Teams building multi-agent systems use Google ADK to coordinate specialized agents around shared work, using Collaborative agents and Dynamic workflows to route tasks as conditions change. Graph-based workflows help them keep complex handoffs understandable while still supporting real-world, branching execution.
Model-flexible agent development
Developers who want to connect agents to multiple model providers and local models use Google ADK to assemble agent logic once and swap backends as needed, using Integrations and Components to connect Gemini, Claude, Ollama, or vLLM. That flexibility helps them test model choices without rewriting orchestration.
Google Cloud deployment path
Organizations using Google Cloud who want agents that can deploy into managed infrastructure use Google ADK to move from design to deployment with Visual Builder and Agent Runtime. Platform teams can standardize how agents are assembled, then run them in a managed environment with less custom plumbing.
How does Google ADK work?
- Start by defining your first agent in Build Agents, then assemble its behavior from reusable Components so the core logic is explicit before you scale the workflow.
- Connect the models and services you need through Integrations, including Gemini, Claude, Ollama, vLLM, LiteLLM, or Google Search, to give the agent real inputs and outputs.
- Map the agent's decision path with Graph-based workflows or Visual Builder, so branching steps, handoffs, and tool calls stay understandable during development.
- Test the agent with Evaluation and Run Agents to check responses, catch regressions, and confirm the workflow behaves as expected before wider rollout.
- Deploy through Agent Runtime and keep iterating on Dynamic workflows or Collaborative agents as requirements change, so the same system can adapt without rebuilding orchestration.
Frequently asked questions
What is Google ADK?
Google ADK is an agent development framework for developers who need to build, run, and extend production-grade agents. It combines Build Agents, Run Agents, Components, and a Visual Builder, with graph-based workflows, collaborative agents, and dynamic workflows for multi-step orchestration. It works with Gemini, Claude, Ollama, vLLM, LiteLLM, and local models, and Google Cloud is a named user.
What is Google ADK used for? Who is it for?
Google ADK is used for Build Agents, Run Agents, and Components. It's built for Platform engineers, AI application developers, and Teams building multi-agent systems.
Does Google ADK have an API and what does it integrate with?
Google ADK doesn't publish a public API. It integrates with Gemini, Gemma, Claude, Ollama, vLLM, and 4 more.
Editor's read
Check whether your deployment plan depends on Google Cloud-managed infrastructure, authentication, observability, and security, since those production features are described as part of Google Cloud deployment. Also verify your model stack early: ADK supports many providers and local models, but the integration set should match your runtime and governance requirements.
Explore other Agent Frameworks
