AI agent platform built on AWS Bedrock AgentCore and Strands Agents framework. Deploy conversational AI with dynamic tool management, multi-layered memory, and flexible integration patterns.
This platform showcases AWS Bedrock AgentCore's capabilities:
- AgentCore Runtime: Containerized Strands Agent as managed AWS service
- AgentCore Memory: Persistent conversation history with DynamoDB backend
- AgentCore Gateway: SigV4-authenticated Lambda tools (12 tools via 5 functions)
- Turn-based Session Management: Optimized memory persistence with buffering
- Dynamic Tool Filtering: Per-user tool selection with real-time updates
-
Frontend + BFF (Next.js)
- Server-side API routes as Backend-for-Frontend
- Cognito authentication with JWT validation
- SSE streaming from AgentCore Runtime
- Session management and file upload handling
-
AgentCore Runtime
- Strands Agent with Bedrock Claude models
- Turn-based session manager (optimized message buffering)
- Uses AgentCore Memory for conversation persistence
- Integrates with AgentCore Gateway via SigV4
- Calls Built-in Tools via AWS API
- Communicates with other agents via A2A protocol (Work in Progress)
-
AgentCore Gateway
- API Gateway with SigV4 authentication
- Routes requests to 5 Lambda functions (12 tools total)
- Lambda functions use MCP protocol
- Tools: Wikipedia, ArXiv, Google Search, Tavily, Finance
-
AgentCore Memory
- DynamoDB-backed conversation persistence
- Automatic history management across sessions
-
Tool Ecosystem
- Local Tools: Weather, visualization, web search, URL fetcher (embedded in Runtime)
- Built-in Tools: Bedrock Code Interpreter for diagrams/charts (AWS API)
- Gateway Tools: Research, search, and finance data (via AgentCore Gateway + MCP)
- Runtime Tools (Work in Progress): Report Writer with A2A protocol
- Per-user tool selection: Each user can enable/disable specific tools
- Real-time filtering: Tools are filtered before each agent invocation
- Category organization: Local, Built-in, Gateway, Runtime
- Connection monitoring: Real-time status for Gateway tools
Turn-based Session Management:
- Buffers messages within a "turn" to reduce API calls
- Merges consecutive user messages before persistence
- Integrated with AgentCore Memory
- Conversation history retained across sessions
Session Isolation:
- User-specific sessions via Cognito user ID
- Automatic session creation and cleanup
- Fresh start with page refresh
Embedded in AgentCore Runtime container:
- Calculator: Mathematical computations (Strands built-in)
- Weather: Current weather by city (wttr.in API)
- Visualization: Chart generation (Plotly)
- Web Search: DuckDuckGo search
- URL Fetcher: Web content extraction
AWS Bedrock-powered capabilities:
- Diagram Generator: Python code + Bedrock Code Interpreter
- Architecture diagrams (AWS, UML)
- Charts and visualizations
- Server-side rendering with file download
Accessed via AgentCore Gateway with SigV4 auth:
| Lambda Function | Tools | API Keys Required |
|---|---|---|
| mcp-wikipedia | search, get_article | None |
| mcp-arxiv | search, get_paper | None |
| mcp-google-search | web_search, image_search | Google API + Search Engine ID |
| mcp-tavily | search, extract | Tavily API Key |
| mcp-finance | stock_quote, stock_history, news, analysis | None (Yahoo Finance) |
Agent-to-Agent (A2A) protocol communication (Work in Progress):
- Report Writer: Comprehensive research reports
- Create report with outline
- Write sections with markdown
- Generate charts (Python + Code Interpreter)
- Export to DOCX (S3 storage)
- AWS Account with Bedrock access
- AWS CLI configured
- Docker installed
- Node.js 18+ and Python 3.13+
# 1. Clone and setup
git clone <repository-url>
cd sample-strands-agent-chatbot/chatbot-app
./setup.sh
# 2. Configure environment
cd ../agent-blueprint
cp .env.example .env
# Edit .env with your AWS credentials
# 3. Start services
cd ../chatbot-app
./start.shAccess at: http://localhost:3000
# Full deployment (all components)
cd agent-blueprint
./deploy.sh
# Or deploy individually:
# 1. Main application (Frontend + AgentCore Runtime)
cd chatbot-deployment/infrastructure
./scripts/deploy.sh
# 2. AgentCore Gateway (Lambda tools)
cd ../../agentcore-gateway-stack
./scripts/deploy.sh
# 3. Report Writer Runtime (optional)
cd ../agentcore-runtime-a2a-stack/report-writer
./deploy.shTools are configured in chatbot-app/frontend/src/config/tools-config.json:
{
"local_tools": [
{
"id": "calculator",
"name": "Calculator",
"enabled": true,
"isDynamic": false
}
],
"gateway_targets": [
{
"id": "gateway_wikipedia-search",
"name": "Wikipedia",
"enabled": false,
"isDynamic": true,
"tools": [
{
"id": "gateway_wikipedia-search___wikipedia_search",
"name": "Search Articles"
}
]
}
]
}isDynamic: Controls whether users can toggle the tool on/off
Configure Gateway tool API keys in AWS Secrets Manager:
# Tavily API Key
aws secretsmanager put-secret-value \
--secret-id strands-agent-chatbot/mcp/tavily-api-key \
--secret-string "YOUR_KEY"
# Google Search Credentials
aws secretsmanager put-secret-value \
--secret-id strands-agent-chatbot/mcp/google-credentials \
--secret-string '{"api_key":"KEY","search_engine_id":"ID"}'
Configure via UI:
- Model: Claude Sonnet 4, Haiku 4.5, etc.
- Temperature: 0.0 (focused) to 1.0 (creative)
- System Prompt: General, Code, Research, RAG Agent
- Caching: Enable/disable prompt caching
User → CloudFront → ALB → Frontend+BFF (Fargate)
↓ HTTP
AgentCore Runtime
(AWS Bedrock service)
↓
┌─────────────────┼─────────────────┐
│ │ │
↓ SigV4 ↓ A2A ↓ AWS API
AgentCore Gateway Report Writer Built-in Tools
(API Gateway) Runtime (Code Interpreter)
↓
Lambda Functions (5x)
(MCP Protocol)
- Frontend: Next.js 15, TypeScript, Tailwind CSS, shadcn/ui
- BFF: Next.js API Routes (server-side) on Fargate
- Runtime: Strands Agents v1.2.0 + FastAPI (Python 3.13) on AgentCore Runtime
- AI: AWS Bedrock (Claude Haiku 4.5 default)
- AgentCore: Runtime, Memory, Gateway components
- Tools: Lambda Functions (MCP protocol)
- Infrastructure: AWS CDK, CloudFront, Cognito
Embed the chatbot in external applications:
<iframe
src="https://your-domain.com/embed"
width="100%"
height="600"
frameborder="0">
</iframe>See docs/guides/EMBEDDING_GUIDE.md for details.
sample-strands-agent-chatbot/
├── chatbot-app/
│ ├── frontend/ # Next.js (Frontend + BFF)
│ │ └── src/
│ │ ├── app/api/ # API routes (BFF layer)
│ │ ├── components/ # React components
│ │ └── config/ # Tool configuration
│ └── agentcore/ # AgentCore Runtime
│ └── src/
│ ├── agent/ # ChatbotAgent + session management
│ ├── local_tools/ # Weather, visualization, etc.
│ ├── builtin_tools/ # Code Interpreter tools
│ └── routers/ # FastAPI routes
│
└── agent-blueprint/
├── chatbot-deployment/ # Main app stack (Frontend+Runtime)
├── agentcore-gateway-stack/ # Gateway + 5 Lambda functions
├── agentcore-runtime-stack/ # Runtime deployment (shared)
└── agentcore-runtime-a2a-stack/ # Report Writer (optional)
- Browser Built-in Tool: Natural language browser automation (planned)
- Redis Short-term Memory: Currently using AgentCore Memory only
- Multi-model Support: Only Claude models (Bedrock limitation)
- DEPLOYMENT.md: Detailed deployment instructions
- Issues: GitHub Issues
- Troubleshooting: docs/guides/TROUBLESHOOTING.md
MIT License - see LICENSE file for details.
Built with AWS Bedrock AgentCore | AWS Samples
