Skip to content

deployment_patterns.md

Latest commit

 

History

History
190 lines (130 loc) · 6.06 KB

deployment_patterns.md

File metadata and controls

190 lines (130 loc) · 6.06 KB

Deployment Guide

FSI Foundry deploys multi-agent AI systems to Amazon Bedrock AgentCore — a fully managed runtime with auto-scaling, built-in observability, and native AWS integration.

Multi-Use-Case and Multi-Framework Support

The platform supports deploying multiple use cases and multiple frameworks to the same AWS account and region with complete isolation:

  • Workspace-based state isolation: Each use case/framework combination gets its own Terraform workspace ({USE_CASE_ID}-{FRAMEWORK_SHORT}-{AWS_REGION})
  • Resource naming isolation: All resources include the use case ID and framework short name in their names
  • Framework isolation: Deploy the same use case with different frameworks (e.g., LangGraph and Strands) simultaneously
  • Independent lifecycle: Deploy, update, and destroy use cases independently
  • No state conflicts: Destroying one use case/framework doesn't affect others

Interactive Deployment (Recommended)

Use the interactive CLI for guided deployment:

./applications/fsi_foundry/scripts/main/deploy.sh

This will guide you through:

  1. Selecting a use case from the registry
  2. Selecting an AI framework (LangGraph or Strands)
  3. Selecting an AWS region
  4. Running the deployment

Command-Line Deployment

For scripted or CI/CD deployments, set environment variables and run the deploy script:

# Set required environment variables
export USE_CASE_ID="kyc_banking"
export FRAMEWORK="langchain_langgraph"
export AWS_REGION="us-east-1"

# Full deployment (infrastructure + application)
./applications/fsi_foundry/scripts/deploy/full/deploy_agentcore.sh

# Test
./applications/fsi_foundry/scripts/main/test.sh

# Cleanup
./applications/fsi_foundry/scripts/main/cleanup.sh

Environment Variables

Variable Description Example
USE_CASE_ID Use case identifier kyc_banking
FRAMEWORK AI framework langchain_langgraph or strands
AWS_REGION Target AWS region us-east-1
AWS_PROFILE AWS CLI profile (optional) default

Automatic Sample Data Upload

All deployment scripts automatically upload sample data from applications/fsi_foundry/data/samples/{USE_CASE_ID}/ to the S3 bucket created during deployment. This ensures that test entities are available immediately for functional testing.

Control Plane Deployment (One-Click)

For deployments via the AVA Control Plane UI:

  1. Navigate to FSI Foundry in the sidebar
  2. Select a use case and click Deploy
  3. Choose the framework and parameters
  4. Click Deploy — the CI/CD pipeline handles the rest

The pipeline automatically provisions infrastructure, builds the Docker image, deploys the AgentCore runtime, and builds/deploys the per-use-case frontend UI.

→ Full AgentCore Deployment Details

Multi-Use-Case Deployment and State Isolation

Terraform Workspace-Based State Isolation

Each use case/framework deployment gets its own Terraform workspace, which provides:

  1. Isolated State Files: Each workspace maintains its own terraform.tfstate file
  2. Independent Lifecycle: Deploy, update, and destroy use cases independently
  3. Resource Separation: All resources are prefixed with use case ID and framework short name
  4. Framework Isolation: Deploy the same use case with different frameworks simultaneously

Workspace Naming Convention

Workspaces follow the pattern: {USE_CASE_ID}-{FRAMEWORK_SHORT}-{AWS_REGION}

Use Case Framework Region Workspace Name
kyc_banking langchain_langgraph us-east-1 kyc_banking-langgraph-us-east-1
kyc_banking strands us-east-1 kyc_banking-strands-us-east-1
fraud_detection langchain_langgraph us-west-2 fraud_detection-langgraph-us-west-2

Framework Short Names

Framework ID Short Name
langchain_langgraph langgraph
strands strands

Multi-Framework Deployment Example

# Deploy KYC with LangGraph
export USE_CASE_ID=kyc_banking FRAMEWORK=langchain_langgraph AWS_REGION=us-east-1
./applications/fsi_foundry/scripts/deploy/full/deploy_agentcore.sh

# Deploy KYC with Strands (same region, isolated stack)
export USE_CASE_ID=kyc_banking FRAMEWORK=strands AWS_REGION=us-east-1
./applications/fsi_foundry/scripts/deploy/full/deploy_agentcore.sh

Destroying a Single Deployment

Only the targeted use case/framework combination is affected:

export USE_CASE_ID=kyc_banking FRAMEWORK=langchain_langgraph AWS_REGION=us-east-1
./scripts/cleanup/cleanup_agentcore.sh
# The strands deployment remains untouched

Prerequisites

  • AWS Account with appropriate permissions
  • AWS CLI >= 2.28.9
  • Terraform >= 1.0
  • Python >= 3.11
  • Docker with buildx support
  • Node.js >= 22 (for UI builds)
  • jq for JSON processing
  • Amazon Bedrock access with Claude models enabled

Verify Prerequisites

aws --version        # >= 2.28.9
terraform --version  # >= 1.0
docker --version
node --version       # >= 22
jq --version

Testing Deployments

Interactive Testing

./applications/fsi_foundry/scripts/main/test.sh

Command-Line Testing

FRAMEWORK=langchain_langgraph AWS_REGION=us-east-1 \
  bash ./applications/fsi_foundry/scripts/use_cases/kyc_banking/test/test_agentcore.sh

The test scripts automatically:

  1. Retrieve deployment outputs (runtime ARN, endpoints)
  2. Run functional tests against the deployed agent
  3. Report test results with pass/fail status

After Deployment

  1. Test Your Deployment — Use the test scripts or the Control Plane's Test Deployment drawer
  2. Clean Up Resources — ./applications/fsi_foundry/scripts/main/cleanup.sh

Architecture Details