Platform Architecture

The architecture behind every decision.

Five purpose-built layers, all connected by an immutable event bus.
From raw sensor pulse to boardroom recommendation. Every step is traceable, auditable, and explainable.

Architecture

Built different. For a reason.

Six architectural decisions that make Digitillis work the way your operation works, not the way enterprise software is expected to work.

Event-Sourced Core

Trace every decision back to the data. Every prediction, action, and outcome is recorded. When an intervention works, or doesn't, you know exactly why.

Manifest-Driven Agents

New AI capabilities activate without touching your operation. No replatforming, no downtime, no risk.

CognitiveMesh

A living knowledge graph linking equipment, failure modes, causal chains, and interventions. Every prediction refines it. Every outcome teaches it. It learns your operation specifically.

Saga Orchestration

Multi-agent workflows with built-in compensation. When a prediction triggers action, a saga coordinates every step (work order, assignment, outcome) with rollback if conditions change.

ARIA: LLM-Optional

A natural-language interface that works without a cloud API key. Deterministic routing handles 70-80% of queries. LLM mode handles complex reasoning. Both answer with grounded evidence.

Multi-Tenant by Design

Schema-per-tenant isolation. Each customer's operational data is cryptographically separate. Tier-gated features, per-tenant configuration, no shared schema risk.

Design Principles

Built on three convictions

Immutable by design

Every state change is an append-only domain event. Nothing is overwritten. Anything can be replayed, audited, and explained from any point in time.

Autonomous, not automated

Agents observe, predict, recommend, and act, within configurable execution tiers. From Observe-Only to Full Autonomous, you control the trust level.

Learning in production

Every prediction outcome feeds back into CognitiveMesh. The platform's knowledge of your specific equipment improves continuously, without retraining cycles.

Layer Deep-Dives

What lives inside each layer

Each layer is purpose-built for its role. Click a layer above to jump directly to its detail, or scroll through all five.

Layer 5 · Experience & Value

Prescriptive, not descriptive.

Every user gets intelligence in their language. ARIA surfaces WHAT happened, WHY it matters, WHEN action is needed, and exactly what to DO, backed by a full evidence package behind every recommendation. Decisions create audit trails, trigger work orders, and feed back into the learning loop.

Persona HubsAI AssistantValue Impact
  • ARIA: LLM-optional copilot, structured and semantic modes
  • Evidence Package: contributing sensors, confidence, causal chain
  • Decision Queue: Approve / Defer / Investigate with full traces
  • Foundry: drag-and-drop composable dashboard canvas
Layer 4 · Agentic

AI that acts. Not just advises.

New AI capabilities activate without touching your operation. No replatforming, no downtime, no risk. Multi-agent sagas coordinate the full loop from prediction to outcome, with built-in compensation if conditions change mid-flight.

AgentsOrchestrationExecution Tiers
  • Manifest-driven: add new agents without platform code changes
  • Saga orchestration with timeout detection and rollback
  • ARIA as the conversational agent, routing across all other agents
  • Circuit breaker: auto-pauses agents after consecutive errors
Layer 3 · Cognitive Intelligence

The mind of the factory.

Production ML models trained on real industry datasets, wrapped in a CognitiveMesh that links equipment, failure modes, causal chains, and intervention outcomes. Every prediction refines the graph. Every outcome teaches it. The platform learns your specific operation over time.

Semantic LayerModelsLearning
  • Proprietary ensemble models across every manufacturing domain
  • Full attribution on every prediction: every insight is explainable and auditable
  • CognitiveMesh: knowledge graph connecting failure modes, causal chains, and outcomes
  • FeatureContracts guarantee the model you trained is exactly what runs in production
Layer 2 · Data & Integration

Immutable. Auditable. Always.

An event-sourced CQRS core appends every state change as an immutable domain event: replayable, tenant-isolated, and schema-validated before it hits the bus. Kafka streams events across the platform. Purpose-built databases handle time-series, graph, and vector data, each optimized for the query patterns manufacturing demands.

Event-SourcedTenant-IsolatedAudit Trail
  • Event-sourced CQRS: full replay capability from any point in time
  • Durable event streaming with schema validation and dead-letter handling
  • Purpose-built databases for time-series, graph, and vector data
  • Complete tenant isolation: your data never shares infrastructure with another customer
Layer 1 · Connectivity

Every signal. Every protocol.

Native support for OPC-UA, MQTT, Modbus TCP/RTU, and S7 means your existing plant equipment connects without gateway translation layers. Edge nodes ingest raw sensor pulses and emit validated, timestamped readings directly into the event stream. Nothing is discarded, everything is traceable.

Industry ProtocolsEdge GatewayEvent Stream
  • OPC-UA, MQTT, Modbus TCP/RTU, Siemens S7: production-ready
  • Edge gateway normalizes signals before they enter the platform
  • Schema validation at ingestion: bad data never reaches storage
  • SAP RFC and CMMS REST connectors for enterprise bidirectional sync

See it running in your factory.

Every layer is production-ready. Schedule a technical walkthrough and see how each layer connects to your existing infrastructure.