Multi-agent orchestration
Compose specialized agents into hierarchical systems and delegate subtasks through model-driven transfer or explicit agent-tool invocation.
Agent Development Kit (ADK)
Freemium
Agent Development Kit (ADK) is an open-source framework for building, evaluating, and deploying conversational and non-conversational AI agents. It supports multi-agent systems, tools, structured workflows, and production runtimes across Python, TypeScript, Go, Java, and Kotlin.
What is Agent Development Kit (ADK)?
Agent Development Kit (ADK) is an open-source framework for building, managing, evaluating, and deploying AI-powered agents. It supports conversational and non-conversational applications, including individual tool-using agents, structured workflows, and multi-agent systems.
The framework is built around agents, tools, callbacks, sessions, memory, artifacts, events, models, and runners. Agents can use language models for reasoning or act as deterministic workflow controllers. Tools extend agents to external APIs, search, code execution, and other services; sessions manage interaction history and state; memory supports context across sessions; and artifacts store files or binary data associated with an interaction.
ADK also provides development and runtime capabilities for building, testing, evaluating, observing, and deploying agents. Its documented ecosystem includes an Agents CLI, web interface, visual builder, API server, evaluation tools, logging, metrics, traces, and deployment options such as Cloud Run and GKE. It is available in Python, TypeScript, Go, Java, and Kotlin.
What can Agent Development Kit (ADK) do?
Structured graph workflows
Create sequential, parallel, loop, routing, and custom-template workflows that combine deterministic execution with adaptive model reasoning.
Tools and external actions
Extend agents with custom function tools, built-in tools, other agents, MCP tools, OpenAPI tools, search, code execution, and service calls.
Context and artifact management
Use sessions, events, state, memory, callbacks, and artifacts to manage interaction history, longer-term context, execution events, files, and binary data.
Development and evaluation tooling
Use the Agents CLI and runtime tools to scaffold, run, test, evaluate, and debug agents, with documented support for evaluation criteria, simulations, and custom metrics.
Runtime and deployment options
Develop in Python, TypeScript, Go, Java, or Kotlin, then use documented agent-runtime deployment paths including Cloud Run and GKE.
Use Cases
“Research assistants”
Create an agent that searches for information and uses tools to help users investigate topics, with session context and artifact handling available when the workflow needs them.
“Multi-step business workflows”
Represent a process with sequential, parallel, loop, or routing steps when some actions require fixed execution paths and others require language-model reasoning.
“Specialized agent teams”
Divide a complex task among purpose-built agents and coordinate their work through hierarchical delegation or explicit agent-tool calls.
“Tool-connected applications”
Build agents that interact with APIs, OpenAPI services, MCP tools, custom functions, code execution, or other services rather than only returning conversational text.
“Production agent operations”
Develop and evaluate an agent, run it through an agent runtime, deploy it to documented environments such as Cloud Run or GKE, and inspect logs, metrics, or traces.
Frequently Asked Questions
Is ADK only for conversational chatbots?
No. ADK supports conversational and non-conversational agents. Agents can participate in workflows, call tools, execute code, manage artifacts, and coordinate with other agents.
Which programming languages does ADK support?
The site lists Python, TypeScript, Go, Java, and Kotlin, with installation examples and API references for these ecosystems.
Can an ADK agent use external tools or services?
Yes. Documented options include custom function tools, agent tools, MCP tools, OpenAPI tools, search, code execution, and calls to external APIs or services.
How does ADK handle conversation context?
Sessions contain the context and history of an interaction through events and state. Memory is a separate mechanism for recalling information across sessions, while artifacts manage files or binary data associated with a session or user.
What deployment options are documented?
The documentation includes an agent runtime and deployment guidance for Cloud Run and GKE, along with runtime configuration and testing for deployed agents.
Quick Facts
- Category
- Developer Tool
- Product type
- Open-source agent development framework
- Primary users
- Developers building AI agents and multi-agent applications
- Supported languages
- Python, TypeScript, Go, Java, Kotlin
- Core workflows
- Multi-agent orchestration, graph workflows, tool use, evaluation, and deployment
- Source domain
- adk.dev
Agent Development Kit (ADK) Alternatives
Microsoft Agent Framework
learn.microsoft.com
Microsoft Agent Framework is a developer framework for building, orchestrating, and deploying AI agents and multi-agent workflows. It supports implementations in Python and .NET.
Claude Platform
platform.claude.com
Claude Platform gives developers programmatic access to Claude models and managed agent infrastructure for building agents and applications. It includes a REST API, official client SDKs, a web Console, and documentation for direct integrations.
Context
context.ai
Context is an enterprise AI agents platform for building, deploying, and improving agents on customer infrastructure, with workspace, runtime, context, evaluation, connectors, and IdP access controls.
Gemini Enterprise Agent Platform
cloud.google.com
Gemini Enterprise Agent Platform, formerly Vertex AI, is Google Cloud’s platform for developers and technical teams to build, deploy, govern, and optimize AI agents, generative AI applications, and machine learning models.
OneReach.ai
onereach.ai
OneReach.ai GSX orchestrates, runs, and governs enterprise AI agents at scale.
Octomind Cloud
octomind.run
Octomind Cloud runs AI agents on persistent cloud machines for research, writing, web work, software development, and scheduled routines. Users give instructions in plain language and can receive results through the web panel, Telegram, WhatsApp, or Slack.