A comprehensive hands-on workshop demonstrating production-ready infrastructure for Agentic AI using Amazon Bedrock AgentCore, Strands Agents framework, and Amazon Nova Pro foundation model. This project covers seven modular labs that progressively build expertise in deploying secure, scalable AI agents at enterprise scale.
Workshop Repository: aws-samples/sample-bedrock-agentcore-with-strands-and-nova Target Audience: Solution Architects, Software Designers, Developers Duration: Multi-day workshop (7 labs) Supported Regions: us-east-1, us-west-2, eu-central-1, ap-southeast-2
This workshop demonstrates how to build production-ready AI agents using AWS's comprehensive AgentCore platform. Unlike basic chatbots, these agents can:
- ✅ Execute Python code in isolated environments
- ✅ Automate web browser interactions
- ✅ Securely manage API credentials
- ✅ Remember conversations across sessions
- ✅ Access external APIs through standardized protocols
- ✅ Provide full observability and debugging capabilities
Amazon Bedrock AgentCore is a comprehensive platform that enables deployment and operation of highly effective AI agents securely at scale.
- Framework Flexibility: Works with any agent framework (CrewAI, LangGraph, LlamaIndex, Strands Agents)
- Model Agnostic: Supports any foundation model inside or outside Amazon Bedrock
- Purpose-Built Infrastructure: Optimized for dynamic agent workloads
- Production Controls: Essential security and governance for real-world deployment
| Service | Purpose |
|---|---|
| AgentCore Runtime | Low-latency serverless environments with session isolation, supporting any framework and long-running agents (up to 8 hours) |
| AgentCore Memory | Manages session and long-term memory, providing context while agents learn from past interactions |
| AgentCore Identity | Enables secure access to AWS services and third-party tools (GitHub, Salesforce, Slack) with user consent |
| AgentCore Gateway | Transforms APIs and Lambda functions into agent-ready tools with unified MCP protocol access |
| AgentCore Code Interpreter | Isolated environment to execute agent-generated code in Python, JavaScript, TypeScript |
| AgentCore Browser | Managed web browser instances for web automation workflows |
| AgentCore Observability | Step-by-step visualization with metadata tagging, scoring, trajectory inspection, and debugging |
Strands Agents is a code-first framework for building and running AI agents in just a few lines of code.
- ✅ Lightweight & Customizable: Simple agent loop that's fully extensible
- ✅ Production Ready: Full observability, tracing, and deployment options
- ✅ Model Agnostic: Supports multiple models from various providers
- ✅ Community Tools: Quick start with powerful contributed tools
- ✅ Multi-Agent Support: Agent teams and autonomous self-improving agents
- ✅ Flexible Modes: Conversational, non-conversational, streaming, non-streaming
- ✅ Security First: Responsible agent operation with data protection
Amazon Nova Pro is Amazon's highly capable, multimodal generative AI foundation model that excels at:
- 📝 Text Processing: Advanced language understanding and generation
- 🖼️ Image Analysis: Deep visual understanding and processing
- 🎥 Video Processing: Multimodal video analysis capabilities
- 📊 Document Analysis: Complex document understanding
- 📈 Data Visualization: Deep analytical capabilities
Access: Available through Amazon Bedrock with strong balance of accuracy, speed, and cost.
Foundation lab to familiarize with Strands Agents framework.
Key Learnings:
- Build agents with tools in few lines of code
- Run agents locally with Nova Pro as the brain
- Create custom tools for specific use cases
- Explore conversation history and agent memory
Objective: Integrate Strands Agents with AgentCore Code Interpreter for dynamic Python code execution.
What I Built:
- Default sandboxed Code Interpreter for math operations
- Custom Code Interpreter with public network access
- Stock price retrieval using external APIs
Key Concepts:
- Secure code execution in containerized environments
- Support for Python, JavaScript, TypeScript
- File references from Amazon S3 for large datasets
- Extended execution time (15 minutes default, up to 8 hours)
- Structured data handling (CSV, Excel, JSON)
Use Cases:
- Data analysis and cleaning
- Mathematical calculations at scale
- Multi-step computational problem solving
Objective: Build AI agents capable of web automation using AgentCore Browser.
What I Built:
- Direct browser session access using Playwright
- Autonomous web navigation agents
- Element interaction (click, extract, summarize)
- AWS Builder Center content extraction
Key Concepts:
- Secure, isolated browser environments
- Visual understanding through screenshots
- Session isolation and audit capabilities
- Human intervention with live interactive view
Use Cases:
- Web application testing
- Form submissions and data entry
- Information extraction from websites
- E-commerce transactions
- Website monitoring and updates
Objective: Securely manage API keys and credentials for external services.
What I Built:
- Secure API Key Credential Providers
- Exa Search API integration with proper credential management
- Elimination of hardcoded credentials
Key Concepts:
- Centralized credential management
- AWS IAM (SigV4) authentication
- OAuth 2.0 integration
- API key secure storage
- Independent request verification
Security Benefits:
- No credential exposure in code
- Audit trails for all access attempts
- Integration with corporate identity providers (Okta, Microsoft Entra ID, Amazon Cognito)
Objective: Deploy Model Context Protocol (MCP) servers to AgentCore Runtime as managed services.
What I Built:
- Custom MCP server with web search functionality
- Amazon Cognito authentication setup
- Cloud-native MCP server deployment
- Local Strands Agent integration with deployed MCP server
Key Concepts:
- Framework Agnostic: Works with LangGraph, Strands, CrewAI
- Model Flexibility: Any LLM (Bedrock, Claude, Gemini, OpenAI)
- MCP Protocol Support: Standardized tool communication
- Extended Execution: Up to 8 hours for complex reasoning
- Session Isolation: Dedicated microVM per user session
- Consumption-Based Pricing: Pay only for resources consumed
Objective: Deploy Strands Agents with comprehensive monitoring and debugging capabilities.
What I Built:
- Strands Agent with weather and calculator tools
- AgentCore Runtime deployment with boto3/IAM
- CloudWatch integration for GenAI observability
- Session management and isolation testing
Key Concepts:
- Built-in Tracing: Captures reasoning steps, tool invocations, model interactions
- OpenTelemetry (OTEL) Compatible: Integrates with existing monitoring stacks
- GenAI Semantic Conventions: Framework-aware instrumentation
- Real-Time Dashboards: Session count, latency, token usage, error rates
- Rich Metadata: Tagging, filtering, custom scoring
Observability Features:
- Step-by-step execution visualization
- Trajectory inspection for debugging
- Performance bottleneck identification
- Quality maintenance at scale
Objective: Create agents with persistent memory across conversations and sessions.
What I Built:
- Short-term memory for immediate context
- Long-term memory for user preferences
- Memory-enabled personalized agent experiences
- Context retention across sessions
Memory Types:
- Stores raw interactions within a single session
- Operations:
CreateEvent,ListSessions,ListEvents,GetEvent - Maintains conversation context
- Stores structured insights extracted from conversations
- Asynchronous Extraction: Background processing without interrupting interactions
- Consolidation: Merges new information with existing knowledge
- Operations:
CreateMemory,GetMemoryRecord,ListMemoryRecords,RetrieveMemoryRecords
Benefits:
- Personalized user experiences
- Context-aware responses
- Reduced redundant questions
- Learning from past interactions
Objective: Automatically generate MCP servers from OpenAPI specifications.
What I Built:
- AgentCore Gateway with authentication
- OpenAPI-based gateway targets (Exa Search API)
- Secure credential management for external APIs
- Strands Agent integration with gateway
Key Concepts:
- Transform Existing APIs: Convert REST APIs to agent-ready tools without custom code
- Unified Access: Single endpoint for multiple tool sources
- Comprehensive Authentication: Inbound (IAM/OAuth) and outbound (OAuth/API keys)
- Semantic Search: Contextual tool discovery at scale
- 1-Click Integrations: Salesforce, Slack, Jira, Asana, Zendesk
Gateway Target Types:
- OpenAPI: RESTful services with OpenAPI schemas
- Smithy: AWS service definitions
- Lambda: AWS Lambda functions
- Amazon Bedrock – Foundation model hosting
- Amazon Bedrock AgentCore – All 7 modular services
- AWS Lambda – Serverless compute
- Amazon CloudWatch – Observability and monitoring
- Amazon ECR – Container registry
- Amazon Cognito – User authentication
- Amazon S3 – Data storage
- AWS IAM – Identity and access management
- AWS CloudFormation – Infrastructure as code
- Strands Agents – AI agent framework
- Python 3.11 – Primary language
- Jupyter Notebook – Interactive development
- OpenTelemetry (OTEL) – Observability instrumentation
- Playwright – Browser automation
- Docker – Containerization
- Amazon Nova Pro – Multimodal foundation model
- Model Context Protocol (MCP) – Tool communication standard
- Visual Studio Code – IDE
- uv – Python environment manager
- boto3 – AWS SDK for Python
- ADOT – AWS Distro for OpenTelemetry
The workshop includes comprehensive architecture diagrams for:
- AgentCore Code Interpreter – Isolated code execution flow
- AgentCore Browser – Web automation with visual feedback
- AgentCore Identity – Multi-path authentication architecture
- AgentCore Runtime (MCP) – Containerized MCP server deployment
- AgentCore Runtime (Agent) – Framework deployment pipeline
- AgentCore Observability – OTEL integration and dashboards
- AgentCore Memory – Short-term and long-term memory systems
- AgentCore Gateway – API transformation and tool management
All diagrams demonstrate clear data flow, security boundaries, and service interactions.
By completing this workshop, I gained expertise in:
✅ AgentCore Architecture – Mastered all 7 modular services ✅ Production Agent Deployment – Moved from local to cloud-native deployments ✅ Security Implementation – Proper credential management and session isolation ✅ Observability – Full tracing and debugging capabilities ✅ Agent Memory Management – Short-term and long-term memory strategies ✅ External Tool Integration – MCP servers and OpenAPI transformations ✅ Web Automation – Browser automation with AI agents ✅ Safe Code Execution – Sandboxed code interpretation ✅ Framework Flexibility – Working with any agent framework or model ✅ Scalable Architecture – Consumption-based pricing and auto-scaling
The project includes detailed screenshots from:
- AWS Bedrock Console (Model access configuration)
- CloudWatch Console (Transaction Search setup)
- CloudFormation Console (Code Editor stack)
- Visual Studio Code (Jupyter Notebook interface, kernel selection, code execution)
- Architecture diagrams for all 7 AgentCore services
All diagrams are stored in the /images folder with descriptive filenames.
- AWS Account with appropriate IAM permissions
- Python 3.11+
- Docker installed
- Basic knowledge of AWS services
- Clone the repository:
git clone https://github.com/aws-samples/sample-bedrock-agentcore-with-strands-and-nova cd sample-bedrock-agentcore-with-strands-and-nova