What is Agent Development Kit (ADK)?
The Agent Development Kit (ADK) is Google’s open-source, code-first framework to build, evaluate, and deploy AI agents. It was announced at Google Cloud Next 2025 and is released under the Apache 2.0 license. ADK is available in Python, Java, Go, and TypeScript and is both model-agnostic and deployment-agnostic, while remaining optimized for Gemini and Google Cloud.
The Agent Development Kit addresses the challenge of taking agent-based applications from prototype to production. Instead of wiring agents together in individual scripts, you define them as versionable code with clear interfaces, built-in evaluation, and observability. This turns experimental agents into maintainable systems that you can test, monitor, and roll out in a controlled way.
Core capabilities
- Multi-agent architectures: Compose specialized agents that collaborate and delegate. ADK 2.0 adds graph workflows that combine deterministic code with adaptive AI reasoning.
- Model independence: Optimized for Gemini, but usable with models from Anthropic, Meta, Mistral, and others through the LiteLLM integration.
- Built-in tooling: Provides a Google Search tool, custom tools, OpenAPI specifications, and MCP (Model Context Protocol) tools. It includes an Agents CLI plus built-in evaluation (user simulation, custom metrics) and observability (logs, metrics, traces).
- Flexible deployment: Run agents on Cloud Run, GKE, or the fully managed Vertex AI Agent Engine Runtime.
Typical use cases
Multi-agent systems: Complex tasks are split across several specialized agents that collaborate and delegate. Graph workflows connect fixed process steps with adaptive reasoning.
Service and support automation: Agents access internal systems through tools, OpenAPI integrations, and MCP, automating requests with traceable behavior.
Research and knowledge agents: Using the Google Search tool and your own data sources, agents deliver grounded answers and support internal research.
Benefits
- Open source under Apache 2.0 with no license cost for the framework
- Model- and deployment-agnostic, reducing vendor lock-in
- Production readiness through built-in evaluation and observability
- EU regions with data residency for the managed Vertex AI Agent Engine Runtime
Integration with innFactory
As a certified Google Cloud Partner, innFactory supports you with the adoption and operation of this service.
Typical Use Cases
Frequently Asked Questions
What is Agent Development Kit (ADK)?
ADK is Google's open-source, code-first framework to build, evaluate, and deploy AI agents. It is model-agnostic, optimized for Gemini and Google Cloud, and available in Python, Java, Go, and TypeScript. The framework is released under the Apache 2.0 license.
When should I use Agent Development Kit (ADK)?
ADK fits when you want to build agent-based applications with code rather than low-code tools, such as multi-agent systems, support automation, or research agents. It is especially useful when you need versioning, testing, and production-grade deployment on Cloud Run, GKE, or Vertex AI Agent Engine.
How much does Agent Development Kit (ADK) cost?
The framework itself is free and open source (Apache 2.0). Costs arise only from deployment and operation: model tokens and the infrastructure you use. On Vertex AI Agent Engine, the runtime is billed per vCPU-hour and GB-hour plus model tokens.
Is ADK locked into Gemini and Google models?
No. ADK is model-agnostic. It is optimized for Gemini and Google Cloud, but the LiteLLM integration lets you connect models from other providers such as Anthropic, Meta, or Mistral. Deployment is not tied to Google Cloud either.
