Google Agent Development Kit (ADK)

Google Agent Development Kit (ADK) is an open-source, code-first framework for building, evaluating and deploying AI agents. It is optimized for Gemini and Google Cloud but designed to work with other model providers and deployment environments. Current documentation covers Python, TypeScript, Go, Java and Kotlin; feature coverage varies by language, so teams should check the support matrix for the implementation they plan to use. This entry links to the canonical Python repository and google-adk package, which use the Apache-2.0 license.

Top features

• Compose LLM agents, custom agents and workflow agents into multi-agent systems. Sequential, parallel and loop patterns handle deterministic coordination, while agents can delegate or call other agents as tools.
• Mix model reasoning with explicit graph workflows, state transitions and application logic when a process needs more predictable execution paths.
• Add built-in, custom, third-party or Model Context Protocol tools. ADK also documents interoperability paths such as Agent2Agent for connecting independently hosted agents.
• Manage working context through sessions, events and state, with services for longer-lived memory and artifacts such as files or generated outputs. Context controls can filter or summarize history and account for tool results.
• Develop locally with the ADK command-line tools and web interface, then test agent behavior with evaluation datasets, trace execution and connect observability integrations.
• Deploy to your own infrastructure or supported Google Cloud targets, including Agent Runtime, Cloud Run and GKE. Model and deployment choices are not tied to a single runtime, although capabilities differ across language SDKs.

Use cases

• Build customer-service or internal operations systems in which specialist agents coordinate under a shared workflow.
• Combine Gemini or another supported model with enterprise APIs, search, databases and MCP servers.
• Create document-processing, research or data-analysis agents that need durable session state, memory and file artifacts.
• Encode business processes as graph-based or sequential workflows with evaluation cases before deployment.
• Move a locally tested agent to Google Cloud or a self-managed container environment while keeping the same core agent design.