Vertex AI Agent Builder Demo

This repository contains demonstrations and examples for building AI agents using Google Cloud's Vertex AI Agent Builder and related frameworks.

What is Vertex AI Agent Builder?

Vertex AI Agent Builder is a managed service on Google Cloud Platform (GCP) that enables developers to create, deploy, and manage conversational AI agents at scale. It provides a comprehensive platform for building intelligent agents that can:

• Understand natural language queries and provide contextual responses
• Integrate with external data sources and APIs
• Execute custom business logic through tools and functions
• Orchestrate multi-step workflows and reasoning chains
• Access enterprise data through grounding with Google Search, Vertex AI Search, or custom data stores

Key Components

Agent Builder

Agent Builder is the visual development environment that allows you to:

• Design conversation flows using a no-code/low-code interface
• Configure data stores for grounding (websites, structured/unstructured data)
• Set up integrations with Vertex AI Search and other Google Cloud services
• Test and iterate on agent behavior through the built-in playground
• Deploy agents as web apps, chat widgets, or API endpoints

Agent Engine (Reasoning Engine)

Agent Engine (also known as Reasoning Engine) is the programmatic approach that provides:

• Full code control for building agents using Python SDKs
• Support for custom tools, functions, and integrations
• Advanced orchestration and reasoning capabilities
• Deployment as containerized services with versioning and monitoring
• Integration with Vertex AI's machine learning infrastructure

Supported Frameworks

Agent Builder

Agent Builder primarily uses:

• Dialogflow CX: Enterprise-grade conversational AI platform with visual flow design
• Vertex AI Search: For grounding agents with enterprise search capabilities
• Built-in connectors: Pre-built integrations for common data sources and APIs

Agent Engine

Agent Engine supports multiple frameworks for programmatic agent development:

1. Agent Development Kit (ADK)

   • Google's official Python framework for building agents
   • Supports tools, multi-agent orchestration, and custom workflows
   • Seamless deployment to Vertex AI Agent Engine
   • Model flexibility (Gemini, Claude, GPT-4, and others)

2. LangChain

   • Popular open-source framework for LLM applications
   • Extensive ecosystem of integrations and tools
   • Can be deployed to Agent Engine with custom containers

3. CrewAI

   • Framework for orchestrating role-playing autonomous AI agents
   • Supports multi-agent collaboration patterns
   • Compatible with Vertex AI through custom deployment

4. Custom Python Applications

   • Build agents from scratch using Vertex AI Python SDK
   • Full control over agent architecture and behavior
   • Direct integration with Gemini API and other Vertex AI services

Examples in This Repository

Available Examples

1. Python Quickstart for ADK
   • Get started with Google's Agent Development Kit (ADK)
   • Learn how to create, test, and deploy a basic agent
   • Includes Docker setup and deployment options (Cloud Run, Vertex AI Agent Engine)
   • Demonstrates custom tool implementation

Coming Soon

2. Agent with ADK and RAG

   • Build an agent that uses Retrieval-Augmented Generation (RAG)
   • Integrate with vector databases for knowledge retrieval
   • Ground agent responses with custom document collections

3. ADK Decisional Agent with Subagents

   • Create a hierarchical agent architecture
   • Implement decision-making logic to route requests to specialized subagents
   • Demonstrate multi-agent orchestration patterns

4. Agent with Vertex AI Search

   • Integrate Vertex AI Search as a grounding source
   • Query enterprise data and websites
   • Combine search results with generative AI responses

Getting Started

Each example includes detailed instructions for:

• Prerequisites and setup
• Local development with Docker
• Testing and iteration
• Deployment to Google Cloud Platform

Navigate to the individual example directories to get started.

Prerequisites

• Google Cloud Platform account with billing enabled
• Docker installed (for local development)
• gcloud CLI installed and configured
• Appropriate GCP APIs enabled (Vertex AI, Cloud Run, etc.)

Resources

• Vertex AI Agent Builder Documentation
• Agent Development Kit (ADK) Documentation
• Vertex AI Python SDK
• Dialogflow CX Documentation

License

See LICENSE file for details.