Google’s open-source Framework Agent Development Kit (ADK) allows for the development of agents in various formats. Despite being developed by Google, it supports models from various manufacturers and can connect to external tools through the MCP protocol.
The uniqueness of the ADK to Google itself is Vertex AI, directly supported through the Vertex AI Agent Engine. However, the ADK itself is packaged as a container, allowing it to run anywhere. Google specifies that the ADK optimizes work with Gemini 2.5 Pro.
The Vertex AI Agent Engine supports running agents from frameworks like ADK, LangGraph, LlamaIndex, LangChain, Crew.ai, and other frameworks. This service provides agent running, code running, session management, and AI memory management. The service is serverless, charged based on CPU and memory usage during runtime.
For cross-agent running tasks, Google and partners have introduced the Agent2Agent (A2A) protocol, allowing different agents to communicate with each other. Currently, many participants, including AI service providers, database systems, and major SaaS providers, have joined the platform.
TLDR: Google’s ADK enables agent development in various formats and models, directly supporting Vertex AI. The containerized ADK optimizes work with Gemini 2.5 Pro and the Vertex AI Agent Engine facilitates seamless communication between various agents.
Leave a Comment