Infrastructure as Code (Pillar 1)

Why infrastructure-first: Agents cannot run without a secure foundation. The Enterprise Kit provisions 35 Bicep modules with one subscription-scoped deployment — identity, networking, data, AI services, MCP hosting, and governance controls — in under 30 minutes. No portal clicking. No manual RBAC assignments. Zero Trust from day one.

Overview

Pillar 1 delivers a complete, subscription-scoped Azure deployment using 27 parameterized Bicep modules that follow Azure Cloud Adoption Framework (CAF) Landing Zone patterns. The deployment creates a dedicated resource group and provisions all services with managed identities, private networking, Azure Policy, and Zero Trust controls.

35 Bicep Modules

→ View full Bicep Module Reference with parameters, configurations, outputs & dependencies

#	Module	Azure Service	Purpose
1	managed-identities	Microsoft.ManagedIdentity	User-assigned managed identities + RBAC role assignments
2	network	Microsoft.Network	VNet, subnets, NSGs, Private DNS zones, private endpoints
3	key-vault	Microsoft.KeyVault	Secrets management, CMK encryption wrapper
4	cosmos-db	Microsoft.DocumentDB	Decision Ledger, agent memory, state persistence (serverless + vector)
5	ai-search	Microsoft.Search	RAG knowledge grounding, semantic ranking
6	foundry-project	Microsoft.MachineLearningServices	AI Foundry Hub + Project workspace
7	agent-application	Foundry Agent API	ARM-native agent packaging + deployment
8	app-insights	Microsoft.Insights	Application Insights + Log Analytics workspace
9	api-management	Microsoft.ApiManagement	APIM AI Gateway with token limits + semantic caching
10	content-safety	Microsoft.CognitiveServices	Content Safety for prompt attack detection + PII
11	guardrails	Foundry Guardrails API	Platform-managed content safety + task adherence
12	toolbox	Foundry Toolbox API	Centralized MCP tool registry with Entra auth
13	continuous-eval	Foundry Evaluations API	Scheduled evaluation runs + quality gates
14	azure-policy	Microsoft.Authorization	Custom policy initiative for agent governance
15	entra-agent-identity	Microsoft.Graph	Entra Agent Identity Blueprints + federated credentials
16	runtime-interceptor	Container App	Kill switch + token budget enforcement
17	api-center	Microsoft.ApiCenter	Private MCP tool catalog for server discovery & governance
18	function-mcp	Microsoft.Web/sites	Azure Functions MCP server hosting template
19	container-app-mcp	Microsoft.App	Container Apps MCP server hosting template
20	cost-management	Microsoft.CostManagement	Budget alerts, anomaly detection, and cost allocation tags
21	hub-spoke	Microsoft.Network	Hub-spoke VNet topology with Azure Firewall and DNS
22	hub-spoke-peering	Microsoft.Network	VNet peering between hub and spoke networks
23	data-residency	Microsoft.Authorization	Allowed-locations policy for data sovereignty
24	chaos-studio	Microsoft.Chaos	Resilience testing experiments (network disruption, CPU pressure)
25	traffic-manager	Microsoft.Network	Multi-region agent endpoint failover
26	managed-hsm	Microsoft.KeyVault	FIPS 140-3 Level 3 hardware security module
27	ai-search-index	Microsoft.Search	RAG pipeline: vector index, skillset, blob indexer

Zero Trust Controls

Every module implements specific Zero Trust for AI controls (ZT-01 through ZT-17):

Control	Description	Primary Module(s)
ZT-01	Identity Foundation — agent & user authentication	entra-agent-identity
ZT-02	Least Privilege RBAC — fine-grained role assignments	managed-identities, entra-agent-identity
ZT-03	Network Microsegmentation — subnet isolation & private endpoints	network
ZT-04	Data Protection — encryption at rest & in transit	key-vault
ZT-05	Tool Authorization — toolbox allowlists & invocation policy	toolbox, runtime-interceptor
ZT-06	Content Safety — input/output filtering & guardrails	content-safety
ZT-07	Audit & Decision Ledger — immutable decision records	cosmos-db
ZT-08	Continuous Evaluation — runtime quality & compliance scoring	continuous-eval
ZT-09	Threat Detection — anomaly detection & prompt injection	api-management, content-safety
ZT-10	Data Boundary — geo-fencing & residency policies	azure-policy
ZT-11	Memory Governance — retention, PII masking, cross-agent isolation	cosmos-db, azure-policy
ZT-12	Multi-Agent Trust — inter-agent credential verification	entra-agent-identity
ZT-13	Supply Chain — model provenance & adapter verification	foundry-hub
ZT-14	Incident Response — automated containment & recovery	observability, runtime-interceptor
ZT-15	MCP Security — server auth, transport, tool authorization & fleet monitoring	api-center, function-mcp, container-app-mcp
ZT-16	Compliance Automation — continuous compliance evidence collection & framework mapping	azure-policy, continuous-eval
ZT-17	Data Residency — sovereign data controls, geo-fencing, & cross-border transfer enforcement	azure-policy, network

Deployment Profiles

The kit supports two deployment profiles via parameter files — allowing cost-efficient development without sacrificing production security:

Profile	SKU Strategy	Networking	Use Case
`parameters/dev.bicepparam`	Free/Basic/Serverless tiers	Public endpoints (optional VNet)	Development, POC, demos
`parameters/prod.bicepparam`	Standard/Premium tiers	Full private networking	Production, regulated workloads

Standards Alignment

Framework	How Pillar 1 Aligns
Azure CAF (Ready phase)	Subscription-scoped, RG isolation, tagging, RBAC, Azure Policy initiative
Azure WAF — Security	Private endpoints, CMK, managed identity, NSGs
Azure WAF — Reliability	Geo-redundancy options, autoscale configurations
Azure WAF — Cost Optimization	Dev/Prod SKU profiles via parameter files
Azure WAF — Ops Excellence	IaC, CI/CD pipelines, diagnostic settings on all resources
Microsoft Zero Trust for AI	17 ZT controls (ZT-01 through ZT-17) across all modules, including MCP security, compliance automation, and data residency

Quick Start

Single-Tenant Deployment

# Deploy the full platform (dev profile)
az deployment sub create \
  --location eastus2 \
  --template-file infra/bicep/main.bicep \
  --parameters infra/bicep/parameters/dev.bicepparam

# Deploy with production profile
az deployment sub create \
  --location eastus2 \
  --template-file infra/bicep/main.bicep \
  --parameters infra/bicep/parameters/prod.bicepparam

Multi-LOB Platform Deployment

For enterprise scenarios with multiple lines of business (LOBs), use the hub-spoke architecture with full network, identity, data, and secrets isolation per LOB.

# Step 1: Deploy shared platform hub
az deployment sub create \
  --location eastus2 \
  --template-file infra/bicep/main-platform.bicep \
  --parameters infra/bicep/parameters/platform-dev.bicepparam

# Step 2: Onboard each LOB via self-service CLI
python scripts/onboard_lob.py \
  --name banking \
  --display-name "Banking Division" \
  --env dev \
  --cidr 10.1.0.0/16 \
  --data-classification confidential

# Step 3: Manage LOBs
python scripts/afek.py lob list --env dev
python scripts/afek.py lob status banking --env dev

Multi-LOB Architecture

The multi-LOB platform engineering layer provides complete isolation between business units:

Isolation Layer	Mechanism	Module
Network	Dedicated spoke VNet per LOB with NSG deny-all cross-spoke	`lob-spoke.bicep`
Identity	Per-LOB UAMI with 6 scoped RBAC roles	`lob-identity.bicep`
Secrets	Dedicated Key Vault per LOB with private endpoint	`lob-keyvault.bicep`
Data	LOB-prefixed Cosmos containers + AI Search indexes	`lob-cosmos-containers.bicep`, `lob-search-index.bicep`
AI Workspace	Separate Foundry Project per LOB (child of shared Hub)	`lob-project.bicep`
Cost	Per-LOB budget alerts + LOB-tagged resources	`cost-management.bicep`

Operations & Capacity

Capacity Monitoring — Azure Monitor workbook with token consumption, Cosmos RU, AI Search latency, and budget burn rate per LOB
Capacity Alerts — Automated alerts for Cosmos RU > 80%, AI Search latency > 500ms, token usage anomalies
Failover Automation — scripts/failover.py with check/failover/failback/status commands
KQL Query Library — 9 pre-built queries for capacity planning and cross-LOB analysis
Custom RBAC Roles — "AI Agent Developer" and "Platform Operator" roles scoped to LOB boundaries