Vertex AI Agent Builder Cheat Sheet

Vertex AI Agent Builder is a platform for building, scaling, and governing enterprise-grade AI agents. It provides the foundation to transform applications and workflows into agentic systems using your enterprise data.

The Three Pillars of Vertex AI Agent Builder

Vertex AI Agent Builder Key Components

Agent Development Kit (ADK)

ADK is an open-source framework for building multi-agent systems. It gives you control over how agents think, reason, and collaborate through guardrails and orchestration controls.

• Build production-ready agents in under 100 lines of Python code (Java support coming soon)

• Bidirectional audio and video streaming for human-like conversations

• Choose your preferred model or deployment target

• Build agents with frameworks like LangChain, LangGraph, AG2, and CrewAI

Agent Garden

Agent Garden is a library in the Google Cloud console with sample agents and tools to speed up development.

• Agents: Prebuilt solutions for specific use cases ready to customize

• Tools: Components that add functionality to your agents (only Google publishes agents to Agent Garden)

Agent Designer (Preview)

Agent Designer is a low-code visual tool in the Google Cloud console for designing and testing agents before moving to code in ADK.

Steps to use Agent Designer:

1. Go to the Agent Designer page in the Google Cloud console

2. Click Create agent to open the canvas

3. Design your agent in the Flow tab (create main agent and subagents with visual representation)

4. Configure agents in the Details panel:

   • Name: Identify the agent

   • Description: Summary of your agent’s purpose

   • Instructions: Guide your agent

   • Model: Select the model

   • Tools: Add tools so the agent can complete tasks

5. Use the Preview tab to test the agent

6. Click Get code to see your agent code and continue development in ADK

Tools available in Agent Designer:

• Google Search: Lets the agent perform web searches (on by default)

• URL context: Lets the model analyze URLs from prompts (on by default)

• Vertex AI Search Data Store: Connect to information indexed in your Vertex AI Search data store

• MCP Server: Add MCP tools by connecting to an MCP server (authentication is None; only supports servers without authentication)

Agent Engine

Agent Engine is a set of services for deploying, managing, and scaling AI agents in production. It handles infrastructure, scaling, security, and monitoring.

Services offered by Agent Engine:

• Runtime: Deploy and scale agents with a managed runtime, customize containers, use VPC-SC compliance, and access models and tools

• Sessions: Store individual interactions for conversation context

• Memory Bank: Store and retrieve information from sessions to personalize interactions

• Code Execution: Run code in a secure, isolated sandbox

• Example Store (Preview): Store and retrieve few-shot examples to improve performance

• Quality and evaluation (Preview): Evaluate agent quality with the Gen AI Evaluation service

• Observability: Track agent behavior with Cloud Trace, Cloud Monitoring, and Cloud Logging

Supported frameworks for Agent Engine:

Enterprise security features:

Agent2Agent (A2A) Protocol

A2A is an open communication standard that lets agents from different ecosystems talk to each other, no matter what framework or vendor built them.

• Agents can publish their capabilities and negotiate how they interact (text, forms, audio/video)

• Enables secure collaboration between agents

• Supported by 50+ partners including Box, Deloitte, Elastic, Salesforce, ServiceNow, UiPath, UKG

Model Context Protocol (MCP) Support

ADK supports MCP, letting agents connect to data sources and capabilities through MCP-compatible tools.

• 100+ pre-built connectors to enterprise systems

• Custom APIs in Apigee

• Application Integration

• Google and Google Cloud services through remote MCP servers

Vertex AI Agent Builder Common Uses

• Build agents your way
  • Create multi-agent workflows with open source frameworks. Start with Agent Garden and use frameworks like ADK, LangGraph, or others, then deploy on Vertex AI.
• Turn workflows into agents
  • Connect agents to ERP, procurement, and HR platforms using 100+ connectors and APIs in Apigee. Reuse workflows from Application Integration to handle document processing, approval routing, data validation, and system updates.
• Connect different agent systems
  •  Use the A2A protocol so agents built on different frameworks can work together on complex tasks without rebuilding systems.
• Improve agent quality
  • Use tracing to see how agents process requests, make decisions, and use tools. Register agents on Agent Engine to Gemini Enterprise for centralized governance and discovery.

Grounding and Data Integration

• RAG: Vertex AI Search offers out-of-the-box RAG; Vector Search combines vector and keyword approaches

• Data sources: Connect to local files, Cloud Storage, Google Drive, Slack, Jira, and more

• Grounding: Use Google Search or data from providers like Cotality, Dun & Bradstreet, HGInsights, S&P Global, and Zoominfo

• Google Maps grounding (experimental): Available for US customers. Access Google Maps data with +100M daily updates covering +250M businesses and places globally

Pricing

Vertex AI Agent Builder uses a pay-as-you-go pricing model. You are charged for:

• Compute resources used by agents deployed on Agent Engine

• Agent memory usage

• Model usage based on input and output tokens (pricing varies by model)

• Tools and pre-built agents (fees depend on the tools used)

A free tier is available for Vertex AI Agent Engine Runtime. For complete and current pricing information, including region-specific rates, visit the Vertex AI pricing page. For offerings in preview, contact your sales team.

References

Vertex AI Agent Builder product page

Vertex AI Agent Builder overview

Agent Designer overview

Vertex AI Agent Engine overview