Home

Work in Progress

This is a TypeScript port inspired by the original Python ADK. Features and stability are actively being developed.

What is ADK for TypeScript?

Agent Development Kit (ADK) for TypeScript is a powerful, open-source framework for building, orchestrating, and deploying AI agents.

Inspired by the architectural principles of Google's Python ADK, our project is reimagined for the TypeScript ecosystem. We offer first-class support for Google Gemini models while providing the freedom to integrate with any LLM, tool, or data source. ADK makes it easy to start with simple agents and scales to support complex, multi-agent systems in any Node.js environment.

Why ADK for TypeScript?

Our vision is to empower every developer to build intelligent agents with the same confidence as any modern software. We achieve this through three core pillars:

Unparalleled Developer Experience: We are TypeScript-first, providing full type-safety and autocompletion out of the box. Our intuitive AgentBuilder API lets you create powerful agents with minimal, readable code.
Powerful & Flexible Architecture: ADK is modular by design. Compose multiple agents, equip them with custom tools, and orchestrate complex workflows. Our provider-agnostic foundation means you're never locked into a single ecosystem.
Production-Ready from Day One: Go beyond prototypes. With built-in session management, persistent memory, and first-class OpenTelemetry support for observability, ADK provides the robust foundation needed for real-world applications.

Core Features

🤖 Multi-Agent Systems

Orchestrate teams of specialized agents that collaborate to solve complex problems.

🔀 Flexible Orchestration

Build complex workflows using sequential, parallel, or LLM-driven dynamic routing.

💾 State & Memory

Enable persistent conversations with state management and context preservation across interactions.

🔄 Multi-LLM Support

Go beyond one provider. Seamlessly use and switch between models from Google, OpenAI, Anthropic, and more.

🛠️ Rich Tool Ecosystem

Equip agents with powerful tools via the Model Context Protocol (MCP)—connect to a wide range of MCP servers in the market or build your own.

🧭 Built-in Tracing

Debug with confidence. First-class OpenTelemetry integration provides deep insights into your agent's execution.

📊 Built-in Evaluation

Systematically assess agent performance by testing final responses and execution trajectories.

🚀 Deployment Ready

Containerize and deploy your agents anywhere, from local servers to scalable cloud infrastructure.

What You Can Build

Sophisticated AI Assistants

Build chatbots that go beyond simple Q&A, with persistent memory and access to real-time information.

Autonomous Task Agents

Create agents that can perform multi-step tasks, like analyzing data, booking appointments, or managing files.

Collaborative Agent Workflows

Orchestrate multiple agents to solve complex problems, from automated research to code generation.

AI-Powered Applications

Connect agents to your databases and APIs to create intelligent, interactive frontends for your services.

Design Philosophy

ADK is designed with the following principles:

TypeScript-first: Full type safety and a modern developer experience.
Simple APIs: Common patterns should be one-liners.
Gradual Complexity: Start simple and add power as you need it.
Extensibility: Every core component is designed to be replaced or extended.
Production Ready: Built-in observability, error handling, and scalability.
Excellent Google Integration: While provider-agnostic, we ensure seamless support for Gemini models and Google Cloud services.

Join our community for help, discussions, and updates:

GitHub: Report issues and contribute
Documentation: Comprehensive guides and API reference
Examples: Real-world applications and use cases