Command Palette

Search for a command to run...

Google ADK Python

Active

An open-source, code-first Python toolkit for building, evaluating, and deploying sophisticated AI agents with flexibility and control.

Visit Live App Source Code

Ken's Contributions

  • Cloud Run Deployment Fix: Identified a critical regression in adk deploy cloud_run where a hardcoded --sandbox-launcher flag failed against the gcloud General Availability (GA) track, unblocking deployment workflows across Google Cloud (#6511).
  • Agent Runtime Session Validation: Diagnosed a ValueError when querying ADK agents hosted on Agent Runtime from Gemini Enterprise app, isolating regex session ID mismatches on fully qualified resource paths (#6036).
  • Agent Gateway SSL Handshake Resolution: Uncovered runtime TLS handshake failures in adk deploy agent_engine caused by unconsumed AGENT_GATEWAY_ROOT_CERTIFICATES build args in container builds, proposing the custom CA certificate trust store integration (#6427).

More contributions on GitHub.

Overview

The Google Agent Development Kit (ADK) is an open-source, code-first Python framework engineered to help developers design, orchestrate, evaluate, and deploy production-grade AI agents. While optimized for Google's Gemini models and Vertex AI infrastructure, ADK is model-agnostic and built for flexibility, control, and enterprise scale.

As agentic AI moves beyond single-turn chatbots into autonomous multi-agent systems, developers require traditional software engineering practices—strict typing, robust session persistence, deterministic routing, and automated evaluation. ADK bridges this gap by offering a clean, idiomatic Python API for building complex cognitive architectures.

Features

  • Code-First Multi-Agent Orchestration: Compose modular agent workflows (Sequential, Parallel, Loop, Switch, and hierarchical multi-agent graphs) with fine-grained execution control.
  • Enterprise State & Session Management: Built-in session persistence across Vertex AI Session Service, databases, and in-memory backends with automatic history compaction and turn resumption.
  • Flexible Tool & Skill Ecosystem: Native support for the Model Context Protocol (MCP), OpenAPI/REST endpoints, custom Python functions, and sandboxed code execution environments.
  • Automated Evaluation Suite: Trajectory grading, session metric evaluation, and LLM-as-a-judge capabilities to rigorously benchmark and harden agent reliability before production release.
  • Streamlined Cloud Deployment: Single-command deployments to Google Cloud Run, Vertex AI Agent Engine (Reasoning Engine), GKE, and local FastAPI servers.

Architecture & Tech Stack

ADK is built on a high-performance asynchronous runtime designed for scalable agent execution and seamless cloud integration.

  • Core Language & Runtime: Python 3.10+ utilizing strict type annotations and asyncio for non-blocking I/O.
  • API & Serving Layer: FastAPI and Starlette powering high-throughput HTTP endpoints and real-time Server-Sent Events (SSE) streaming.
  • Model Layer: Native integration with the Google GenAI SDK and Gemini Enterprise Agent Platform (Vertex AI), with multi-provider routing via LiteLLM.
  • Cloud Infrastructure: Deeply integrated with Google Cloud services, including Cloud Run, Vertex AI Agent Engine, Artifact Registry, and Cloud Build.