Reference architecture

From enterprise signals to governed action

A component model for systems that observe operational change, assemble trustworthy context, reason over options, coordinate tools and humans, and learn from outcomes.

The architecture gap between traditional enterprise AI and agentic systems

Architecture flow

Enterprise systems & events → context fabric → reasoning & planning → orchestration → tools, humans & actions → feedback

Verified data products, event channels, spatial intelligence, and knowledge graphs converge in a context fabric. Models and agents reason over that bounded context. Orchestration separates planning from execution, routes consequential choices to humans, and records outcomes. Trust and governance apply across every stage.

Core components

ComponentRole
Enterprise systemsERP, CRM, GIS, SCM, IoT, data platforms, and operational applications that supply facts or execute work.
Verified Data LayerTrusted products, contracts, lineage, quality, metadata, observability, and stewardship.
Event-Driven ArchitectureStreams, topics, event mesh, change-data capture, operational triggers, replay, and recovery.
Spatial IntelligenceGIS, routing, proximity, service areas, network constraints, movement, and digital twins.
Knowledge Graph LayerEntities, relationships, ontologies, policies, dependencies, provenance, and temporal context.
AI Reasoning & PlanningModels and agents that evaluate goals, constraints, evidence, options, uncertainty, and risks.
AI OrchestrationApproved tool calls, workflow execution, API coordination, budgets, human checkpoints, and compensation.
Trust & GovernanceIdentity, access, policy, risk thresholds, approvals, audit, compliance, explanation, and monitoring.
Feedback & OutcomesResults, incidents, exceptions, overrides, user feedback, policy violations, and performance signals.

Design principles

  • Treat AI actions as governed enterprise operations.
  • Separate reasoning from tool execution through policy-aware orchestration.
  • Make trust, provenance, uncertainty, and policy context machine-readable.
  • Use spatial and graph context when decisions depend on routes, proximity, relationships, or dependency chains.
  • Provide safe state, human escalation, emergency stop, rollback, and compensation.
  • Capture feedback from every material action to improve architecture, controls, and outcomes.

Architecture assurance

Use the Practitioner Suite to define the target maturity, map controls to components, collect operating evidence, and test the architecture before certification.