Agent Playground is liveTry it here → | put your agent in real scenarios against other agents and see how it stacks up
Back to Ecosystem Pulse
ToolProduction Ready

adk-js

by google

TypeScript agent developer kit with evaluation hooks and deployable runtimes

TypeScript
Updated Jun 18, 2026
Share:
1.2k
Stars
161
Forks

View on GitHub

Overview

Provides a code-first TypeScript toolkit for building, evaluating, and deploying agentic AI applications. Exposes composable primitives for agent behavior, tool integration, and evaluation hooks so teams can iterate on agents with control. Includes SDKs and runtimes optimized for Vertex AI and common LLM providers to simplify deployment and testing. In particular, it integrates with Agent Registry Pattern to manage tools and capabilities, and leverages best practices from the Orchestrator-Worker Pattern to coordinate multi-agent workflows.

Key Benefits

As agents become more autonomous and interdependent, teams need repeatable ways to build, test, and deploy them while capturing behaviour and failure modes. ADK-js supplies the developer ergonomics and evaluation hooks needed to measure agent performance and track changes over time. That makes it easier to add traceability and bring agent track records into trust and governance workflows, aligning with the Human-in-the-Loop approach.

Target Use Cases

Teams building production agentic applications in TypeScript who want integrated evaluation and Vertex AI deployment paths. For architecture guidance, consider the Orchestrator-Worker Pattern to design robust tool coordination and Agent Registry Pattern for centralized capability management.

Applications

  • Iterating on agent policies with built-in evaluation hooks to compare behaviours
  • Deploying TypeScript agents to Vertex AI while preserving logs and evaluation traces
  • Integrating LLM providers and tools into composable agent pipelines for production workflows
Topics
agentagenticagentic-aiagentsagents-sdkaiai-agentsaiagentframeworkgenaigenai-chatbot+6 more
Similar Tools
autogenlangchain
Keywords
multi-agent orchestrationagent evaluationvertex-aiagent reliability