AI / GenAI / Agentic AI

AI modeled as enterprise architecture - not a bolt-on.

Models, agents, retrieval, memory, tools, guardrails, evaluation and agent identity are first-class parts of the solution model, with the same rigor as data and infrastructure.

Model an AI solution See data architecture

Agentic AI

An agentic request, modeled end to end.

Identity, gateway, orchestration, retrieval, tools, memory and human approval - each an addressable, governable part of the architecture.

User

Identity

AI Gateway

Agent Orchestrator

RAG

Enterprise Search

Tools

Memory

Approval

LLM

Response

From the reference model

The AI plane of a real reference architecture.

These are the AI components of the illustrative Enterprise AI Knowledge & Decision Platform - each projected from the same model that drives the interactive blueprint.

AI / Agent Gateway

High

AI gateway

Provider: AWS
Environments: DEV · STG · PROD · DR
Classification: confidential

Guardrail / Policy

High

Policy engine

Provider: AWS
Environments: STG · PROD · DR
Classification: confidential

Agent Orchestrator

Critical

Agent runtime

Provider: AWS
Environments: DEV · STG · PROD · DR
Classification: confidential

RAG Engine

High

Retrieval service

Provider: AWS
Environments: DEV · STG · PROD · DR
Classification: confidential

Vector Store

High

Vector database

Provider: AWS
Environments: DEV · STG · PROD · DR
Classification: confidential

Embedding Model

Medium

Model endpoint

Provider: AWS
Environments: DEV · STG · PROD · DR
Classification: confidential

Tools

Medium

Tool registry

Provider: AWS
Environments: DEV · STG · PROD
Classification: confidential

Agent Memory

Medium

State store

Provider: AWS
Environments: DEV · STG · PROD · DR
Classification: confidential

Human Approval

High

Approval gate

Provider: AWS
Environments: STG · PROD
Classification: confidential

Model Gateway

Critical

Model router

Provider: AWS
Environments: DEV · STG · PROD · DR
Classification: confidential

LLM - Primary

High

Model provider

Provider: AWS
Environments: PROD · DR
Classification: confidential

LLM - Secondary

Medium

Model provider

Provider: Azure
Environments: PROD · DR
Classification: confidential

Fallback Model

Medium

Model provider

Provider: AWS
Environments: PROD · DR
Classification: confidential

Projected from the reference model

Every AI component above is part of the same Enterprise Solution Model behind thereference blueprint - with guardrails, evaluation, model routing and fallback modeled as first-class concerns.

Explore the reference blueprint

Supported concerns

The AI architecture surface it covers.

GenAI
Agentic AI
RAG
LLM gateways
Model routing
Embeddings
Vector databases
Agent orchestration
Agent memory
Tool use
Guardrails
Evaluation
Human approval
AI observability
AI cost / token economics
Agent identity & permissions

Guardrails & evaluation

Policy checks on inputs and outputs, plus evaluation harnesses - modeled as controls, not afterthoughts.

Agent identity & permissions

Agents are principals with scoped permissions and audit trails - treated as first-class identities.

Cost & token economics

Inference and token cost enter the cost model, so AI scale is funded and bounded, not a surprise.

Design the AI system before you deploy it.

Start an architecture conversation