Skip to content

RUNTIME_README.md

Latest commit

 

History

History
100 lines (74 loc) · 3.12 KB

RUNTIME_README.md

File metadata and controls

100 lines (74 loc) · 3.12 KB

🧭 Navigation:
Home | Vision LLM Theory | Local Testing | Deployment UI | Deployment | CDK Stacks | 🔵 Runtime | S3 Files | Lambda Analyzers | Prompting System

🐳 AgentCore Runtime Container

Docker container running the Strands agent for PDF analysis orchestration with WebSocket streaming.

Contents

runtime/
├── Dockerfile.websocket      # Container definition (WebSocket streaming)
├── build_and_push_websocket.sh # Build and push to ECR
├── requirements.txt          # Python dependencies
└── agent/
    ├── __init__.py
    └── main-websocket.py     # Strands agent entry point (WebSocket)

Quick Start

Important

The ECR stack must be deployed before building the container.

# Build and push (requires ECR stack deployed first)
./build_and_push_websocket.sh

# Then deploy Runtime stack
cd .. && cdk deploy badgers-runtime-websocket

Container Details

Property Value
Base Image python:3.11-slim
Architecture linux/arm64
Port 8080
User bedrock_agentcore (non-root)
Health Check GET /ping every 30s

Dependencies

  • strands-agents — Agent framework
  • bedrock-agentcore[strands-agents] — AgentCore SDK
  • mcp — Model Context Protocol client
  • boto3 — AWS SDK
  • httpx — HTTP client
  • aws-opentelemetry-distro — X-Ray tracing

Environment Variables

Set by Runtime stack:

Variable Description
AWS_DEFAULT_REGION AWS region
GATEWAY_URL MCP Gateway endpoint
COGNITO_CREDENTIALS_SECRET_ARN Secrets Manager ARN for OAuth credentials
AGENTCORE_MEMORY_ID Memory ID for session persistence

Build Options

# Build with custom tag
./build_and_push_websocket.sh v1.0.0

# Manual build
docker build -f Dockerfile.websocket --platform linux/arm64 -t pdf-analysis-agent .

Customization

Modify Agent Behavior

Edit agent/main-websocket.py to change:

  • System prompt
  • Model selection
  • Tool handling
  • Session management

Add Dependencies

  1. Add to requirements.txt
  2. Rebuild and push: ./build_and_push_websocket.sh
  3. Redeploy Runtime stack

Troubleshooting

# View Runtime logs
aws logs tail /aws/bedrock-agentcore/runtimes/<runtime_id> --follow

# Check container locally
docker run -p 8080:8080 pdf-analysis-agent

# Verify ECR image
aws ecr describe-images --repository-name pdf-analysis-agent-<deployment_id>