| name | agent-demo |
| description | Complete end-to-end examples demonstrating Google's AI agent ecosystem. Shows how to combine vertex-ai SDK, google-adk, vertex-agent-engine, and a2a protocol to build real-world agent systems. Use when: (1) Learning the full agent development workflow, (2) Understanding how technologies work together, (3) Building production multi-agent systems, (4) Need reference implementations for common patterns. |
Agent Demonstrations
Complete, runnable examples showing how to build AI agent systems using Google's AI ecosystem. Each example builds on previous ones, demonstrating the journey from simple API calls to production-grade distributed multi-agent systems.
Quick Start
Installation
Install the complete ecosystem:
pip install google-genai google-adk "google-cloud-aiplatform[agent_engines]>=1.112" "a2a-sdk[sql,http-server]"
For detailed installation options, see Installation Guide.
Environment Setup
Set up your Google Cloud environment:
bash scripts/setup_gcloud.sh YOUR_PROJECT_ID
python scripts/validate_environment.py
python scripts/generate_agent.py sdk --name MyAgent
For manual setup, see Setup Guide.
Examples by Complexity
Level 1: Foundation
Basic SDK Agent - Start here
- Simple text generation with Vertex AI SDK
- Image analysis
- Function calling basics
- Skills used: vertex-ai
- Complexity: Foundation
Level 2: Single Agent Systems
ADK Agent with Tools
- Build an agent with Google Search and custom tools
- Streaming responses
- Session management for conversations
- Skills used: vertex-ai, google-adk
- Complexity: Intermediate
Level 3: Multi-Agent Systems
Multi-Agent System
- Hierarchical agent coordination with ADK
- Sub-agents for specialized tasks
- Shared context and memory
- Skills used: vertex-ai, google-adk
- Complexity: Advanced
Level 4: Distributed Systems
Distributed Agents with A2A
- Multiple independent agents communicating via A2A protocol
- Agent discovery and coordination
- Service-to-service agent interaction
- Skills used: vertex-ai, google-adk, a2a
- Complexity: Advanced
Level 5: Production Deployment
Production Deployment
- Deploy agents to Vertex AI Agent Engine
- Monitoring and observability
- Scaling and performance optimization
- A2A endpoints for external communication
- Skills used: vertex-ai, google-adk, vertex-agent-engine, a2a
- Complexity: Production
Learning Paths
Choose your learning path based on your goals:
Path 1: Learning the Basics
Start with foundation and single-agent patterns:
- Basic SDK Agent
- ADK Agent with Tools
Path 2: Building Multi-Agent Systems
Master internal multi-agent coordination:
- Basic SDK Agent
- ADK Agent with Tools
- Multi-Agent System
Path 3: Production-Ready Systems
Complete journey to production deployment:
- Basic SDK Agent
- ADK Agent with Tools
- Multi-Agent System
- Distributed Agents
- Production Deployment
Understanding the Stack
This skill demonstrates four complementary technologies working together:
┌─────────────────────────────────────────────────────────────────┐
│ YOUR APPLICATION │
│ (Built with agent-demo) │
└─────────────────────────────────────────────────────────────────┘
│
┌────────────────────────┼────────────────────────┐
│ │ │
▼ ▼ ▼
┌───────────────┐ ┌─────────────────┐ ┌────────────────┐
│ vertex-ai │ │ google-adk │ │ vertex-agent- │
│ │ │ │ │ engine │
│ Foundation │◄─────│ Agent Builder │─────►│ Deployment │
│ (SDK) │ │ │ │ │
└───────────────┘ └─────────────────┘ └────────────────┘
│
│ exposes via
│
▼
┌─────────────────┐
│ a2a │
│ Communication │
│ Protocol │
└─────────────────┘
LEARNING PATH:
1. Start with vertex-ai → Learn model APIs and capabilities
2. Graduate to google-adk → Build agents with tools and orchestration
3. Deploy with agent-engine → Production infrastructure and scaling
4. Connect with a2a → Distributed multi-agent systems
Technology roles:
- vertex-ai: Low-level SDK for direct model API access
- google-adk: High-level framework for building agents with tools
- vertex-agent-engine: Managed deployment infrastructure
- a2a: Agent-to-agent communication protocol
Common Patterns
The examples demonstrate five key architectural patterns:
- SDK → ADK Migration: Start simple, scale complexity as needed
- Internal Multi-Agent: Coordinated specialists in single application
- Distributed Multi-Agent: Independent services via A2A protocol
- Hybrid Architecture: Internal multi-agent + external A2A interface
- Tool Composition: Built-in + custom tools working together
For detailed pattern explanations, see Common Patterns.
Real-World Use Cases
Examples demonstrate patterns applicable to:
- Customer Service: FAQ bots, triage, knowledge base search
- Research & Analysis: Web research, data analysis, report generation
- Workflow Automation: Email/calendar, task orchestration, document processing
- Development Tools: Code review, documentation, testing assistance
For detailed use case mappings, see Use Cases Guide.
Example Structure
Each example follows a consistent structure:
- Overview: What you'll build and why
- Architecture: System design and component diagram
- Prerequisites: Specific requirements
- Step-by-Step Guide: Detailed implementation
- Testing: How to run and validate
- Troubleshooting: Common issues and solutions
- Next Steps: How to extend
Repository Structure
agent-demo/
├── SKILL.md # This file
├── scripts/
│ ├── setup_gcloud.sh # Automated Google Cloud setup
│ ├── validate_environment.py # Environment validation
│ └── generate_agent.py # Agent template generator
└── references/
├── setup.md # Detailed setup guide
├── patterns.md # Architectural patterns
├── use-cases.md # Real-world applications
├── basic-sdk-agent.md # Level 1: Foundation
├── adk-agent-with-tools.md # Level 2: Single agent
├── multi-agent-system.md # Level 3: Multi-agent
├── distributed-agents.md # Level 4: Distributed
└── production-deployment.md # Level 5: Production
Tips for Success
- Follow the order: Examples build on each other
- Run the code: Don't just read, execute and experiment
- Modify and extend: Try variations to deepen understanding
- Check prerequisites: Ensure your environment is set up correctly
- Read error messages: They often contain helpful debugging info
- Use the development UI: ADK's
adk ui is great for testing
- Start simple: Get basic version working before adding features
Getting Help
If you encounter issues:
- Check the example's troubleshooting section
- Review the related skill documentation:
- Verify your Google Cloud setup: Run
python scripts/validate_environment.py
- Check API quotas and billing
- Look for similar issues in GitHub repos
Additional Resources
What You'll Learn
After completing these examples, you'll understand:
- When to use SDK vs ADK vs Agent Engine
- How to build agents with tools
- Multi-agent coordination strategies
- Distributed system patterns
- Production deployment best practices
Ready to build your own agent systems!
