A production application
Orchestration, RAG, workflows, evaluations and interfaces engineered around a measurable business outcome.
Australian enterprise AI engineering
Sovereign Agentic AI
Secure AI agents. Australian data. Enterprise governance.
Build production AI agents that can reason, retrieve information and take action across your organisation - while maintaining control of identity, data, models, tools and approvals.
01Sovereign
02Secure
03Governed
04Operational
Beyond the chatbot
Agentic AI without uncontrolled autonomy
Traditional generative AI responds to a prompt. An AI agent can pursue an objective, retrieve information, call approved tools, reason over multiple steps and perform controlled actions.
That additional capability introduces additional risk. DW Data designs the security, governance and operational controls around the agent - not just the AI model itself.
An enterprise platform
Azure infrastructure, private networking, identity, model access, data services and repeatable deployment automation.
A governed control system
Explicit permissions, tool policies, approvals, model routing, audit evidence and controlled autonomy.
An operated service
Monitoring, incident response, prompt and model lifecycle, cost control, regression testing and access review.
Secure by design
Our architecture principles
Six practical rules shape how DW Data designs enterprise agents for regulated environments.
Australian data first
Keep enterprise data within Australian-controlled environments wherever the workload requires it.
Identity before intelligence
Every user, agent and tool interaction must have an identity and an attributable purpose.
Least privilege by default
Agents receive only the data, operations and time-bound permissions required for the task.
No unrestricted tool access
Enterprise actions are mediated through governed interfaces with validation and policy controls.
Humans retain authority
High-impact or ambiguous actions stop at an explicit approval gate owned by an authorised person.
Everything is observable
Prompts, models, retrieval, tool calls, approvals, costs and outcomes can be traced and evaluated.
Production standard
Prototype AI vs production Agentic AI
The difference is the control system around the model: identity, network boundaries, permissions, evidence and operational ownership.
| Prototype AI | DW Sovereign Agent Platform |
|---|---|
| API key | Managed identity |
| Public endpoints | Private networking |
| Broad access | Least privilege |
| One model | DW Model Gateway |
| Direct tool access | DW Agent Security Gateway |
| Uncontrolled autonomy | DW Agent Autonomy Model |
| Limited logs | Full observability |
| Manual testing | Continuous evaluation |
| Informal prompts | Governed system instructions |
| Demo | Production service |
Deployment profiles
Choose the sovereignty boundary that matches your risk
Sovereignty is an architecture and risk decision, not a single cloud setting. DW Data makes the processing boundary, control assumptions and customer decisions explicit.
Profile 01
Australian Sovereign
For sensitive or regulated workloads requiring the strongest practical Australian residency controls, subject to service availability and security assessment.
- Australian Azure regions and customer-owned subscriptions
- Australian-hosted data stores and telemetry
- Private networks, private endpoints and restricted egress
- Regional model deployment where available
- Customer-controlled secrets, encryption options and model catalogue
- No unnecessary public internet exposure
This profile can support sensitive government architecture patterns. It does not make a solution automatically suitable for PROTECTED or remove the need for agency assessment and authorisation.
Profile 02
Sovereign-Controlled Frontier AI
For customers who need newer frontier capabilities that may not yet be offered as an Australian regional deployment.
- Enterprise data and systems remain in the Australian architecture
- Outbound model access is explicit and policy controlled
- Sensitive content can be redacted, tokenised, transformed or excluded
- Routing depends on classification, use case and customer risk acceptance
- Providers can be changed through the DW Model Gateway
- Responses return through validation before application use
Use of any external or globally processed model remains a customer policy and risk decision. Available models and processing locations change over time.
Sovereign-controlled frontier AI flow
DW Data reference patternAustralian enterprise data
Data classification
Policy gateway
Redaction / minimisation
Approved frontier model
Validation
Australian application environment
Azure-first. Cloud-flexible.
The DW Sovereign Agent Platform
A customer-owned Azure reference architecture spanning Microsoft Foundry, identity, private networking, governed data, model access, tool execution, observability and delivery automation.
DW Sovereign Agent Platform - Azure reference architecture
DW Data reference patternCustomer Australian Azure Environment
Australia East / SoutheastPrivate networkRestricted egressCustomer-owned subscription
User & Experience
Web applicationsTeamsEnterprise appsAPIsCustom apps
Identity & Access
Entra IDConditional AccessRBACManaged Identity
Agent Platform
Microsoft FoundryOrchestrationPlanningMemoryWorkflowEvaluation
DW AI Control Plane
Model GatewayAgent Security GatewayPolicy EngineClassificationGuardrailsHuman approvalAudit
Models, Knowledge & Tools
Azure modelsApproved external modelsFabric / OneLakeAI SearchSQL / ADLSSharePointERP / CRM / APIs
Azure is DW Data's primary reference platform. AWS delivery is available where it better fits the customer's estate or control framework. Exact Azure service and region availability is confirmed during solution design.
DW Agent Security Gateway
The agent never gets unrestricted access
Enterprise agents should not receive broad SQL credentials, shared service accounts or direct access to production systems. Actions flow through a governed security and execution layer.
Agents receive exactly the capabilities they need - not broad access to enterprise systems.
Every sensitive action can be attributed to a user, agent, model, tool and approval decision.
DW Agent Security Gateway
DW Data reference patternUser or system
Authenticated request and user context
Enterprise agent
Defined objective and bounded role
DW Agent Security Gateway
Identity-aware policy and execution layer
Identity validation
Agent identity
User delegation
Tool allow-list
Operation allow-list
Parameter validation
Data classification
Contextual policy
Injection detection
Content validation
Transaction limits
Human approval
Rate limiting
Immutable audit
Anomaly detection
Kill switch
Fabric / SQL
SharePoint
ERP / CRM
APIs / LOB apps
Model abstraction
DW Model Gateway
Do not tightly couple an enterprise application to one model. A governed routing layer keeps model choice explicit, reviewable and changeable.
DW Model Gateway
DW Data reference patternAgent
Workload and context
DW Model Gateway
Classification, use case, cost, latency, capability and sovereignty
AU regional model
Sovereignty
Approved APAC model
Availability
Frontier model
Capability
Specialist model
Task fit
Coding model
Engineering
Embedding model
Retrieval
Sensitive / high-control
AU regional approved model
Internal corporate
Approved model catalogue
Public information
Frontier model where policy allows
Identity-aware
Every agent gets an identity
DW Data does not recommend shared service accounts for autonomous agents. Each agent uses a separate workload identity, delegated user context where required, narrow RBAC scope and time-limited permissions where appropriate.
Every agent gets an identity
DW Data reference patternEmployee
User context
Entra ID
Authentication and Conditional Access
Agent identity
Separate workload identity
Managed identity
Delegated or workload access
Policy
RBAC, scope and duration
Approved tools
Allow-listed operations
Enterprise systems
Audited resources
DW Agent Autonomy Model
Autonomy should be earned, not assumed
Autonomy is assigned by use case and risk, then enforced in architecture. Regulated customers will often operate predominantly at A0-A2.
A0
Advisory
Search, explain or summarise
A1
Drafting
Generate correspondence or recommendations
A2
Approval Required
Prepare an action and request human approval
Common production boundaryA3
Bounded Autonomy
Perform explicitly defined low-risk actions
A4
High Autonomy
Broader actions under exceptional controls
A2 - approval required
DW Data reference patternAgent
Prepares action
Proposed transaction
Structured and validated
Human review
Approve or reject
ApproveReject
Security Gateway
Revalidates policy
Enterprise system
Executes authorised action
Packaged outcomes
Agents built around work, not novelty
Initial solution patterns are tailored to the organisation, its controls and measurable operating outcomes.
Procurement Agent
Find contract obligations and expiry risk, analyse supplier spend and produce sourcing briefs grounded in procurement policy.
Data & Analytics Agent
Help teams discover governed data, generate validated queries, explain datasets and surface anomalies through Fabric.
Knowledge Agent
Answer staff questions from approved policies and records, with security-trimmed retrieval and citations to source documents.
Service Desk Agent
Improve first-contact resolution through guided diagnosis, ticket creation and approved low-risk remediation workflows.
Finance Agent
Investigate transactions, reconcile data, explain variances and draft management commentary while finance retains authority.
Document & Compliance Agent
Compare policies, extract obligations, identify evidence gaps and maintain a traceable record of compliance review activity.
Software Engineering Agent
Understand repositories, prepare code and tests, analyse architecture and create reviewable pull requests inside a controlled environment.
Explore secure coding agentsSelect a workflow to see its control profile
Each use case combines different data, tools, autonomy and approval decisions.
Procurement Agent
- Data accessed
- Contracts, supplier records, policy and approved spend data
- Tools used
- Document Search, Fabric Query API, procurement API
- Autonomy
- A1 - Drafting
- Human approvals
- A procurement officer releases recommendations and any downstream action
- Example outcome
- Earlier expiry visibility, faster policy research and evidence-backed sourcing briefs
A DW Data advantage
Agents + Microsoft Fabric
Fabric provides the governed enterprise data foundation that many useful agents need: OneLake, lakehouses, warehouses, semantic models, metadata, BI and managed data products.
Agents should use semantic and query services that preserve permissions, validation and auditability. They should not simply be handed broad SQL credentials.
Agents + Microsoft Fabric
DW Data reference patternAgent
User context and intent
Policy layer
Classification and least privilege
Semantic / Query Service
Validated DAX, SQL or metadata request
Microsoft Fabric
Governed workspace access
OneLake data products
Warehouse, lakehouse and semantic models
Government and regulated industry
Built for environments where governance matters
Architecture and controls can be mapped to the customer's obligations and assurance process. We use careful, evidence-based language: designed to support, aligned with and subject to customer assessment.
Australian Government
Queensland Government
Critical infrastructure
Regulated enterprise
Financial services
Healthcare
Utilities
Education
Australian control context
Designed around Australian security expectations
DW Data maps architecture decisions and evidence to the frameworks that apply to the organisation. This supports, but does not replace, customer assurance and authorisation.
Queensland
- Information and cyber security policy (IS18)
- Queensland Government Information Security Classification Framework (QGISCF)
- Queensland Privacy Principles
- Public Records Act 2023 and records governance obligations
- ISO/IEC 27001-aligned security management
Australian Government
- Australian Government Information Security Manual (ISM)
- Essential Eight
- Protective Security Policy Framework (PSPF)
- Policy for the responsible use of AI in government
- AI technical standard and Agentic AI addendum
Accreditation statement: Architectures can be designed to operate within IRAP-assessed Azure environments and mapped against relevant ISM controls. Final accreditation, information classification, risk acceptance and system authorisation remain the customer's responsibility and may require independent security assessment.
Evidence, not just software
Governance deliverables
A production engagement leaves the customer with the architecture, controls, registers, evaluations and operating evidence needed to own the service.
Architecture
- Reference and solution architecture
- Data flow and network diagrams
- Identity and integration design
Security
- Threat model and attack surface review
- Agent privilege and secrets design
- Security test evidence
AI Governance
- Use-case, model, agent and tool registers
- DW Agent Cards and autonomy classification
- Data source and instruction registers
Evaluation
- Quality, groundedness and retrieval benchmarks
- Prompt-injection and tool-use tests
- Regression suites and release gates
Operations
- Logging, telemetry and monitoring
- Cost and model lifecycle management
- Incidents, access reviews and change control
DW Agent Card
A governance record for every agent
The Agent Card records ownership, autonomy, data, models, tools, permissions, required approvals and evidence. It gives governance, security and operations teams a common control record.
DW Agent Card
Procurement Contract Agent
- Owner
- Procurement Operations
- Model
- Approved enterprise model
- Data
- Contract and supplier information
- Tools
- Fabric Query API, Procurement API, Document Search
Can
- Search contracts and supplier information
- Analyse obligations, spend and expiry risk
- Generate recommendations and sourcing briefs
Cannot
- Approve purchases
- Change supplier master data
- Create or release payments
Human approval: Purchase recommendation release
Audit: 100% tool invocation logging
Turnkey delivery
From AI idea to governed production agent
DW Data can provide the complete stack: strategy, architecture, secure Azure landing zone, data, models, agents, integrations, security, governance, production and AgentOps.
From AI idea to governed production agent
DW Data reference pattern- 01
Discover & Assess
Use cases, classification, risk and roadmap
- 02
Sovereign AI Landing Zone
Identity, networking, gateways and monitoring
- 03
Production Agent
Application, retrieval, tools, evaluation and deployment
- 04
DW AgentOps
Operate, evaluate, govern and improve
Agentic AI Discovery
For organisations identifying where to start.
- Workshops and use-case prioritisation
- AI, data and security assessment
- Initial architecture and delivery roadmap
Sovereign Agent Platform
For organisations ready to build for production.
- Secure Azure landing zone
- Model and Agent Security Gateways
- Governance framework, monitoring and first production agent
Managed AgentOps
For teams operating production agents.
- Ongoing monitoring and evaluation
- Security, access and model management
- Quality, cost and performance optimisation
Recurring operations
Managed AgentOps
DevOps + SecOps + ModelOps + FinOps for autonomous software.
AgentOps is the ongoing discipline of keeping production agents secure, reliable, cost-effective and aligned with their approved purpose as models, prompts, data and tools change.
Model upgrades
Prompt lifecycle
Policy management
Agent monitoring
Security monitoring
Access reviews
Evaluation
Regression testing
Cost monitoring
Quality monitoring
Incident management
Vulnerability review
Tool changes
Performance optimisation
New agent onboarding
Frequently asked questions
Practical answers for enterprise buyers
Can all AI processing stay within Australia?
It depends on the selected model, service feature and deployment type. Regional deployments can process prompts and responses in the selected Azure region where supported. Global and DataZone deployments have different processing boundaries, and some frontier models may not be available regionally in Australia. DW Data makes that distinction explicit in the architecture and model-routing policy.
Can this be used for government workloads?
Yes. The architecture is designed with government security, privacy, audit and governance expectations in mind. Suitability still depends on the workload, classification, selected services and controls. Customer accreditation, risk acceptance and system-specific security assessment remain necessary.
Can agents access our internal systems?
Yes, through controlled least-privilege integrations. Agents call approved operations through the DW Agent Security Gateway rather than receiving unrestricted credentials to databases, SharePoint, ERP, CRM or production systems.
Can the agent perform actions?
Yes, but autonomy is explicitly classified and governed. Sensitive actions can require a human approval, deterministic validation, transaction limits and a final policy check immediately before execution.
Which models can we use?
The DW Model Gateway supports a customer-approved catalogue across Azure regional models, other Foundry models and explicitly permitted external providers. Routing can consider classification, capability, latency, cost and sovereignty. Available options change with cloud and model availability.
Can we use the latest coding models?
Where customer security and data policy permit it. Repository content can be minimised and classified before routing, and sensitive workloads can be restricted to approved deployments even if that means selecting a different model.
Does our data train public AI models?
For models sold by Azure, Microsoft states that prompts and completions are not used to train or improve base generative AI models. The final answer depends on the selected model service, feature, deployment type and current provider terms, which DW Data confirms during solution design.
Can it integrate with Microsoft Fabric?
Yes. Fabric is a core DW Data capability. Agents can work with governed lakehouses, warehouses, semantic models, metadata and OneLake data products through scoped query services and existing Fabric permissions.
Can it use SharePoint and Microsoft 365?
Yes, subject to Entra permissions, security trimming, data classification and the selected architecture. Access is scoped to the user or agent role and remains auditable.
Can you build on AWS?
Yes. DW Data is Azure-first and cloud-flexible. Azure is the primary reference platform; AWS patterns can be designed where customer strategy, existing controls or service requirements call for it.
From concept to controlled production
Is your organisation ready for agentic AI?
We can assess your use case, data, security requirements and sovereignty constraints and provide a practical production architecture.