Azure Tenant Grapher - Architecture Deep Dive

📋 System Overview

Azure Tenant Grapher is a comprehensive cloud infrastructure discovery, documentation, and replication tool. It discovers every resource in an Azure tenant, stores the results in a richly-typed Neo4j graph database, and offers extensive tooling for visualization, analysis, documentation, and Infrastructure-as-Code (IaC) generation.

Core Capabilities

🔍 Discovery & Scanning

Comprehensive Azure resource discovery across all subscriptions with identity import from Microsoft Graph API. Supports filtering by subscriptions and resource groups.

💾 Graph Database

Rich Neo4j graph with typed nodes, relationships, and RBAC modeling. Includes extensible relationship engine with modular rules.

📊 Visualization

Interactive 3D graph visualization with filtering, search, and ResourceGroup labels. Desktop GUI with Electron and React.

🏗️ IaC Generation

Generate Terraform, Bicep, and ARM templates from the graph with transformation rules and deployment scripts.

🤖 AI Agent Mode

Natural language queries over the graph using MCP (Model Context Protocol) and AutoGen agents.

🔐 Threat Modeling

Automated DFD creation, threat enumeration using STRIDE methodology, and comprehensive security reports.

Technology Stack

Python 3.8+ Neo4j Azure SDK TypeScript React AutoGen FastAPI

🏛️ Overall System Architecture

graph TB subgraph "External Services" Azure[Azure ARM API] Graph[Microsoft Graph API] OpenAI[Azure OpenAI] end subgraph "Data Layer" Neo4j[(Neo4j Graph DB)] end subgraph "Core Services" Discovery[Azure Discovery Service] Processing[Resource Processing Service] ContainerMgr[Container Manager] AADService[AAD Graph Service] end subgraph "IaC Engine" Traverser[Graph Traverser] Engine[IaC Engine] TFEmitter[Terraform Emitter] BicepEmitter[Bicep Emitter] ARMEmitter[ARM Emitter] Validators[Validators] end subgraph "Agent System" MCP[MCP Server] AgentMode[Agent Mode] ThreatAgent[Threat Modeling Agent] end subgraph "User Interfaces" CLI[CLI Dashboard] SPA[Electron GUI] Viz[3D Visualizer] end Azure --> Discovery Graph --> AADService Discovery --> Processing AADService --> Processing Processing --> Neo4j ContainerMgr --> Neo4j Neo4j --> Traverser Traverser --> Engine Engine --> TFEmitter Engine --> BicepEmitter Engine --> ARMEmitter Engine --> Validators Neo4j --> MCP MCP --> AgentMode MCP --> ThreatAgent OpenAI --> AgentMode OpenAI --> ThreatAgent Neo4j --> CLI Neo4j --> SPA Neo4j --> Viz CLI -.-> ContainerMgr SPA -.-> ContainerMgr style Neo4j fill:#008cc1,color:#fff style Azure fill:#0078d4,color:#fff style Graph fill:#0078d4,color:#fff style OpenAI fill:#10a37f,color:#fff

Architecture Principles

Separation of Concerns: Clear boundaries between discovery, storage, transformation, and presentation layers
Async Processing: Core services use asyncio for concurrent API calls and improved performance
Extensibility: Plugin-based relationship rules, multiple IaC formats via emitters, modular validators
Graph-Native: Neo4j as the source of truth for all resource relationships and queries
Multi-Interface: CLI, GUI, and agent-based interfaces for different user workflows

🧩 Module Architecture

Core Services Module

graph LR subgraph "src/services/" ADS[Azure Discovery Service] RPS[Resource Processing Service] AAD[AAD Graph Service] TenantMgr[Tenant Manager] Identity[Identity Collector] Filter[Discovery Filter Service] end subgraph "src/" RP[Resource Processor] CM[Container Manager] SM[Session Manager] end ADS --> RPS AAD --> RPS Filter --> ADS RPS --> RP RP --> SM SM --> Neo4j[(Neo4j)] CM --> Neo4j Identity --> AAD TenantMgr --> RPS style Neo4j fill:#008cc1,color:#fff

Azure Discovery Service

src/services/azure_discovery_service.py - Discovers Azure resources using Azure SDK with pagination and rate limiting support.

Resource Processing Service

src/services/resource_processing_service.py (110 lines) - Orchestrates resource processing, coordinates AAD import, and manages concurrent processing with progress tracking.

Container Manager

src/container_manager.py (746 lines) - Manages Neo4j Docker container lifecycle including:

Container start/stop operations
Readiness checks and health monitoring
Database backup and restore (via neo4j-admin)
Volume management for test isolation
Connection verification

Relationship Rules Engine

The relationship rules system is modular and extensible. Each rule implements the RelationshipRule interface and can create specific types of relationships in the graph. Rules are registered in src/relationship_rules/__init__.py.

IaC Generation Module

graph TB CLI[CLI Command] --> Handler[IaC Handler] Handler --> Engine[IaC Engine] Engine --> Traverser[Graph Traverser] Engine --> Filters[Subset Filters] Engine --> Transform[Transformation Rules] Engine --> Validators[Validators] Traverser --> Neo4j[(Neo4j)] Engine --> Emitters{Format?} Emitters --> |terraform| TF[Terraform Emitter
2126 lines] Emitters --> |bicep| Bicep[Bicep Emitter
479 lines] Emitters --> |arm| ARM[ARM Emitter
473 lines] Validators --> SubnetVal[Subnet Validator] Validators --> AddrVal[Address Space Validator] Validators --> TFVal[Terraform Validator] TF --> Output[main.tf.json] Bicep --> Output2[main.bicep] ARM --> Output3[template.json] Engine --> DeployScript[deploy.sh] style Engine fill:#667eea,color:#fff style Neo4j fill:#008cc1,color:#fff

Key Components

Graph Traverser (src/iac/traverser.py) - Queries Neo4j and extracts resources with relationships
IaC Engine (src/iac/engine.py) - Orchestrates the entire generation pipeline
Terraform Emitter (src/iac/emitters/terraform_emitter.py) - Converts resources to Terraform HCL/JSON format with 50+ Azure resource types mapped
Validators (src/iac/validators/) - Validates subnet containment, address space conflicts, and Terraform syntax

🌊 Data Flow Architecture

Discovery to Graph Flow

sequenceDiagram participant User participant CLI participant Discovery participant Azure participant AAD participant Processing participant RelRules participant Neo4j User->>CLI: atg scan --tenant-id TENANT_ID CLI->>CLI: Setup Neo4j Container CLI->>Discovery: Start Discovery par Parallel Discovery Discovery->>Azure: List Subscriptions Azure-->>Discovery: Subscription List Discovery->>Azure: Discover Resources (paginated) Azure-->>Discovery: Resources Batch and AAD Import Discovery->>AAD: Get Users AAD-->>Discovery: User List Discovery->>AAD: Get Groups AAD-->>Discovery: Group List end Discovery->>Processing: Process Resources loop For Each Resource Processing->>Neo4j: Create Resource Node Processing->>RelRules: Apply Relationship Rules RelRules->>Neo4j: Create Relationships end Neo4j-->>CLI: Stats & Progress CLI-->>User: Completion Report

Graph to IaC Flow

sequenceDiagram participant User participant CLI participant Handler participant Engine participant Traverser participant Neo4j participant Validators participant Emitter participant FS User->>CLI: atg generate-iac --format terraform CLI->>Handler: generate_iac_command_handler() Handler->>Engine: generate() Engine->>Traverser: traverse() Traverser->>Neo4j: MATCH (r:Resource)... Neo4j-->>Traverser: TenantGraph{resources, relationships} Traverser-->>Engine: TenantGraph Engine->>Engine: Apply Subset Filters Engine->>Engine: Apply Transformation Rules Engine->>Validators: validate_address_spaces() Validators-->>Engine: ValidationResult alt Has Conflicts Engine->>User: ⚠️ Warning: Address space conflicts end Engine->>Emitter: emit(TenantGraph) Emitter->>Emitter: Convert to Terraform Emitter->>FS: Write main.tf.json Emitter->>FS: Write variables.tf.json Emitter->>FS: Write deploy.sh FS-->>User: ✅ IaC Generated: ./output/

Agent Mode Query Flow

sequenceDiagram participant User participant AgentMode participant Workbench participant MCP participant Neo4j participant LLM User->>AgentMode: "How many storage resources?" AgentMode->>Workbench: list_tools() Workbench-->>AgentMode: [get_neo4j_schema, read_neo4j_cypher] AgentMode->>Workbench: call_tool(get_neo4j_schema) Workbench->>MCP: get_neo4j_schema MCP->>Neo4j: CALL db.schema.visualization() Neo4j-->>MCP: Schema JSON MCP-->>AgentMode: Schema AgentMode->>LLM: Generate Cypher Query Note over AgentMode,LLM: Prompt includes schema + question LLM-->>AgentMode: MATCH (r:Resource) WHERE... AgentMode->>Workbench: call_tool(read_neo4j_cypher, query) Workbench->>MCP: read_neo4j_cypher MCP->>Neo4j: Execute Cypher Query Neo4j-->>MCP: Query Results MCP-->>AgentMode: Results JSON AgentMode->>LLM: Format Answer LLM-->>AgentMode: Natural Language Answer AgentMode-->>User: "There are 3 storage resources in the tenant."

🔄 Workflow Diagrams

Scanning Workflow

flowchart TD Start([User: atg scan]) --> CheckDocker{Docker
Available?} CheckDocker -->|No| InstallPrompt[Prompt to Install Docker] InstallPrompt --> Exit1([Exit]) CheckDocker -->|Yes| CheckNeo4j{Neo4j
Running?} CheckNeo4j -->|No| StartNeo4j[Start Neo4j Container] StartNeo4j --> WaitReady[Wait for Readiness] WaitReady --> RunMigrations[Run Database Migrations] CheckNeo4j -->|Yes| RunMigrations RunMigrations --> AuthCheck{Azure
Authenticated?} AuthCheck -->|No| AuthPrompt[Prompt: az login] AuthPrompt --> Exit2([Exit]) AuthCheck -->|Yes| StartDashboard[Launch CLI Dashboard] StartDashboard --> InitDiscovery[Initialize Discovery Service] InitDiscovery --> ListSubs[List Subscriptions] ListSubs --> FilterCheck{Filters
Applied?} FilterCheck -->|Yes| ApplyFilters[Apply Subscription/RG Filters] FilterCheck -->|No| DiscoverAll[Discover All Resources] ApplyFilters --> DiscoverFiltered[Discover Filtered Resources] DiscoverFiltered --> ProcessResources DiscoverAll --> ProcessResources[Process Resources] ProcessResources --> AADCheck{AAD Import
Enabled?} AADCheck -->|Yes| ImportAAD[Import Users & Groups] AADCheck -->|No| CreateNodes ImportAAD --> CreateNodes[Create Neo4j Nodes] CreateNodes --> ApplyRules[Apply Relationship Rules] ApplyRules --> GenMetrics[Generate Metrics] GenMetrics --> Complete([Scan Complete]) style Start fill:#667eea,color:#fff style Complete fill:#28a745,color:#fff style Exit1 fill:#dc3545,color:#fff style Exit2 fill:#dc3545,color:#fff

IaC Generation & Deployment Workflow

flowchart TD Start([User: atg generate-iac]) --> ValidateGraph{Graph
Exists?} ValidateGraph -->|No| Error1[Error: Run scan first] Error1 --> Exit1([Exit]) ValidateGraph -->|Yes| TraverseGraph[Traverse Neo4j Graph] TraverseGraph --> SubsetFilter{Subset
Filter?} SubsetFilter -->|Yes| ApplySubset[Apply Subset Filter] SubsetFilter -->|No| TransformRules ApplySubset --> TransformRules{Transform
Rules?} TransformRules -->|Yes| ApplyTransform[Apply Transformation Rules] TransformRules -->|No| ValidateAddr ApplyTransform --> ValidateAddr[Validate Address Spaces] ValidateAddr --> ConflictCheck{Conflicts
Found?} ConflictCheck -->|Yes| AutoFix{Auto-fix
Enabled?} AutoFix -->|Yes| FixConflicts[Fix Address Conflicts] AutoFix -->|No| LogWarning[Log Warnings] FixConflicts --> SelectFormat ConflictCheck -->|No| SelectFormat LogWarning --> SelectFormat SelectFormat{Format?} -->|terraform| TFEmit[Terraform Emitter] SelectFormat -->|bicep| BicepEmit[Bicep Emitter] SelectFormat -->|arm| ARMEmit[ARM Emitter] TFEmit --> GenFiles[Generate IaC Files] BicepEmit --> GenFiles ARMEmit --> GenFiles GenFiles --> GenScript[Generate deploy.sh] GenScript --> Complete([Generation Complete]) Complete --> UserDeploy{User runs
./deploy.sh?} UserDeploy -->|Yes| AzAuth{Azure
Authenticated?} UserDeploy -->|No| ManualDeploy[Manual Deployment] AzAuth -->|No| RunLogin[az login] RunLogin --> ValidatePlan AzAuth -->|Yes| ValidatePlan[Validate/Plan] ValidatePlan --> ReviewPlan{User
Approves?} ReviewPlan -->|No| CancelDeploy([Cancelled]) ReviewPlan -->|Yes| Deploy[Deploy Resources] Deploy --> Monitor[Monitor Deployment] Monitor --> DeployComplete([Deployment Complete]) style Start fill:#667eea,color:#fff style Complete fill:#28a745,color:#fff style DeployComplete fill:#28a745,color:#fff style Exit1 fill:#dc3545,color:#fff style CancelDeploy fill:#ffc107,color:#000

Undeployment Workflow

flowchart TD Start([User: atg undeploy]) --> ListDeploy[List Active Deployments] ListDeploy --> SelectDeploy{Select
Deployment} SelectDeploy --> LoadState[Load Deployment State] LoadState --> AnalyzeDeps[Analyze Resource Dependencies] AnalyzeDeps --> OrderDeletion[Order Deletion Sequence] OrderDeletion --> Confirm{User
Confirms?} Confirm -->|No| Cancel([Cancelled]) Confirm -->|Yes| DeleteResources[Delete Resources] DeleteResources --> MonitorDeletion[Monitor Deletion Progress] MonitorDeletion --> CheckComplete{All
Deleted?} CheckComplete -->|No| RetryFailed{Retry
Failed?} CheckComplete -->|Yes| CleanState[Clean Deployment State] RetryFailed -->|Yes| DeleteResources RetryFailed -->|No| PartialClean[Partial Cleanup] PartialClean --> Report[Generate Failure Report] CleanState --> Complete([Undeploy Complete]) Report --> Exit([Exit with Warnings]) style Start fill:#667eea,color:#fff style Complete fill:#28a745,color:#fff style Cancel fill:#ffc107,color:#000 style Exit fill:#fd7e14,color:#fff

🤖 Agent Mode & MCP Architecture

Agent Mode System

graph TB User[User] --> CLI[CLI: atg agent-mode] CLI --> Startup[Ensure Neo4j Running] Startup --> InitMCP[Initialize MCP Workbench] InitMCP --> MCPServer[MCP Server Process] MCPServer --> Tools[Available Tools] Tools --> Tool1[get_neo4j_schema] Tools --> Tool2[read_neo4j_cypher] Tools --> Tool3[write_neo4j_cypher] InitMCP --> InitLLM[Initialize Azure OpenAI Client] InitLLM --> Assistant[AutoGen Assistant Agent] Assistant --> Workbench[MCP Workbench] Workbench --> MCPServer User --> |Question| Assistant Assistant --> |1. Get Schema| Tool1 Tool1 --> Neo4j[(Neo4j)] Neo4j --> |Schema| Assistant Assistant --> |2. Generate Query| LLM[Azure OpenAI] LLM --> |Cypher Query| Assistant Assistant --> |3. Execute Query| Tool2 Tool2 --> Neo4j Neo4j --> |Results| Assistant Assistant --> |4. Format Answer| LLM LLM --> |Natural Language| Assistant Assistant --> User style Neo4j fill:#008cc1,color:#fff style LLM fill:#10a37f,color:#fff style Assistant fill:#764ba2,color:#fff

MCP Integration Details

The agent mode leverages the Model Context Protocol (MCP) to provide a standardized interface between the LLM and Neo4j database. Key components:

src/mcp_server.py - Launches the MCP server process (uvx mcp-neo4j-cypher)
src/agent_mode.py - Orchestrates the agent workflow with multi-step tool chaining
AutoGen Integration: Uses AutoGen's AssistantAgent with MCP workbench for tool calling
Two Modes: Interactive REPL and single-question mode

Threat Modeling Agent

graph TB User[User] --> ThreatCLI[atg threat-model] ThreatCLI --> LoadSpec[Load Tenant Spec] LoadSpec --> ThreatAgent[Threat Modeling Agent] ThreatAgent --> DFD[DFD Builder] ThreatAgent --> Enum[Threat Enumerator] ThreatAgent --> ASB[ASB Mapper] ThreatAgent --> Report[Report Builder] DFD --> |Identify Components| Neo4j[(Neo4j)] DFD --> |Generate DFD| Mermaid[Mermaid Diagram] Enum --> |STRIDE Analysis| Threats[Threat Categories] Threats --> Spoofing Threats --> Tampering Threats --> Repudiation Threats --> InfoDisclosure[Information Disclosure] Threats --> DoS[Denial of Service] Threats --> ElevationPriv[Elevation of Privilege] ASB --> |Map to Controls| AzureSecBaseline[Azure Security Baseline] Report --> MDReport[threat_model_report.md] Report --> JSONData[threat_data.json] style Neo4j fill:#008cc1,color:#fff style ThreatAgent fill:#dc3545,color:#fff

🖥️ SPA/GUI Architecture

Electron Application Architecture

graph TB subgraph "Main Process (Node.js)" Main[index.ts
App Lifecycle] IPC[IPC Handlers] ProcMgr[Process Manager] Menu[Application Menu] end subgraph "Renderer Process (React)" App[App.tsx] Tabs[Tab Components] Context[React Context] Hooks[Custom Hooks] end subgraph "CLI Integration" CLI[Python CLI] WebSocket[WebSocket Server] Logs[Log Streaming] end subgraph "Backend (Express)" API[API Routes] Neo4jConn[Neo4j Connection] end Main --> IPC Main --> ProcMgr Main --> Menu IPC <--> App App --> Tabs App --> Context Tabs --> Hooks ProcMgr --> CLI CLI --> WebSocket WebSocket --> Logs Logs --> Context Tabs --> API API --> Neo4jConn Neo4jConn --> Neo4j[(Neo4j)] style Main fill:#764ba2,color:#fff style App fill:#61dafb,color:#000 style Neo4j fill:#008cc1,color:#fff

Key GUI Features

📊 Status Tab

Dashboard showing Neo4j status, resource counts, and system health metrics.

🔍 Scan Tab

Interactive scanning interface with real-time progress tracking and log streaming.

📝 Generate Spec Tab

Generate tenant specifications with format options (YAML/JSON/Markdown).

🏗️ Generate IaC Tab

IaC generation interface with format selection, subset filtering, and transformation rules.

🎨 Visualize Tab

3D graph visualization with search, filtering, and navigation controls.

🤖 Agent Mode Tab

Interactive chat interface for natural language queries over the graph.

🔐 Threat Model Tab

Generate threat models and security reports for the tenant.

⚙️ Config Tab

Manage environment variables and Azure credentials.

IPC Communication Pattern

sequenceDiagram participant Renderer participant Main participant CLI participant Neo4j Renderer->>Main: IPC: start-scan Main->>Main: Spawn CLI Process Main->>CLI: python -m scripts.cli scan loop Real-time Updates CLI->>Main: stdout/stderr Main->>Renderer: IPC: scan-progress Renderer->>Renderer: Update UI end CLI->>Neo4j: Write Resources CLI->>Main: Process Exit Main->>Renderer: IPC: scan-complete Renderer->>Main: IPC: get-stats Main->>Neo4j: Query Stats Neo4j-->>Main: Stats Data Main-->>Renderer: IPC: stats-data Renderer->>Renderer: Update Dashboard

🔐 Threat Model & Security Analysis

This threat model follows the Microsoft Threat Modeling Tool methodology using the STRIDE framework. The analysis focuses on the current architecture: a single-user development tool running on a developer's system using Docker for the database.

System Context & Trust Boundaries

STRIDE Analysis

🎭 Spoofing

Threat	Impact	Mitigation
T1: Malicious process impersonates CLI/GUI to access Neo4j	HIGH - Unauthorized access to sensitive tenant data	✅ Neo4j requires password authentication. ⚠️ Password stored in .env file (file permissions critical)
T2: Stolen Azure credentials used to scan wrong tenant	MEDIUM - Unauthorized tenant discovery	✅ Uses Azure CLI authentication (tokens time-limited). ⚠️ No additional verification of tenant ID
T3: Man-in-the-middle attack on localhost Neo4j connection	LOW - Requires local access	⚠️ bolt:// protocol not encrypted. ✅ localhost-only reduces exposure

🔧 Tampering

Threat	Impact	Mitigation
T4: Malicious modification of graph data in Neo4j	CRITICAL - Corrupted infrastructure documentation leading to failed deployments	⚠️ No audit logging of graph changes. ⚠️ No backup automation by default
T5: Tampering with generated IaC before deployment	CRITICAL - Deployment of malicious infrastructure	❌ No integrity checks on generated files. ✅ User reviews in deploy.sh
T6: Modified .env file with malicious credentials	HIGH - Unauthorized access to Azure resources or LLM usage	⚠️ .env file has user-only permissions. ❌ No integrity verification
T7: Supply chain attack via compromised dependencies	CRITICAL - Arbitrary code execution	✅ uv.lock pins versions. ⚠️ No vulnerability scanning in CI

📝 Repudiation

Threat	Impact	Mitigation
T8: No audit trail for scan operations	MEDIUM - Cannot prove who scanned tenant	✅ Logs stored in logs/ directory. ⚠️ Logs not signed or timestamped
T9: No audit trail for IaC deployments	HIGH - Cannot prove who deployed what resources	✅ Azure Activity Log tracks deployments. ❌ No local deployment records

🔓 Information Disclosure

Threat	Impact	Mitigation
T10: Neo4j data exposed if Docker port is exposed	CRITICAL - Full tenant data exposure	✅ Default docker-compose binds to localhost only. ⚠️ User can modify
T11: Credentials in .env file exposed	CRITICAL - Azure tenant compromise, API key theft	✅ .env in .gitignore. ⚠️ File permissions not enforced. ❌ No encryption
T12: Secrets stored in Neo4j graph	HIGH - Key Vault secrets, connection strings exposed	⚠️ Graph stores full resource properties. ❌ No secret redaction. ⚠️ Database not encrypted at rest by default
T13: Generated IaC files contain sensitive data	HIGH - Secrets, keys, connection strings in plain text	⚠️ Output directory permissions user-only. ❌ No secret detection
T14: Logs contain sensitive information	MEDIUM - Resource names, IDs, configuration details	⚠️ Logs stored locally with user permissions. ❌ No sensitive data filtering

⛔ Denial of Service

Threat	Impact	Mitigation
T15: Resource exhaustion from scanning large tenants	MEDIUM - Developer workstation becomes unresponsive	✅ Rate limiting in discovery service. ✅ Pagination support. ⚠️ No memory limits
T16: Neo4j container consumes excessive resources	MEDIUM - System performance degradation	⚠️ No resource limits in docker-compose. ✅ Can be configured by user
T17: Disk space exhaustion from graph data	LOW - Tool stops working	✅ Backup and cleanup utilities available. ⚠️ No automatic cleanup

⬆️ Elevation of Privilege

Threat	Impact	Mitigation
T18: Docker socket access enables container escape	HIGH - Full system compromise	✅ Tool only manages Neo4j container. ⚠️ User must have Docker access. ❌ No additional sandboxing
T19: Malicious deployment script execution	CRITICAL - Arbitrary Azure resource creation	⚠️ deploy.sh generated with user review expected. ❌ No script validation
T20: Agent mode LLM prompt injection	HIGH - Unauthorized graph queries or modifications	✅ System message restricts agent scope. ⚠️ No query validation. ⚠️ write_neo4j_cypher tool available

Security Recommendations

🔴 Critical Priority

Implement secret detection and redaction in graph ingestion
Add integrity checks for generated IaC files
Enable Neo4j encryption at rest
Implement dependency vulnerability scanning in CI

🟠 High Priority

Add audit logging for all graph modifications
Implement .env file encryption or secure secret management
Add Neo4j bolt+s:// (TLS) connection support
Restrict agent mode to read-only queries

🟡 Medium Priority

Implement automated backup scheduling
Add resource limits to docker-compose
Implement sensitive data filtering in logs
Add deployment record tracking

🟢 Low Priority

Implement disk space monitoring and alerts
Add tenant ID verification prompts
Enhanced script validation for generated files

Assumptions & Constraints

Single User Tool: Designed for use by individual developers on their local machines, not as a multi-tenant service
Trusted Environment: Assumes the developer's workstation is secure and not compromised
Azure Trust: Relies on Azure ARM/Graph API security and authentication mechanisms
Docker Dependency: Requires Docker to be installed and running with appropriate user permissions
Network Access: Requires internet access for Azure APIs, OpenAI, and package downloads