Skip to content
Demiton

Platform Overview

Demiton is an enterprise AI infrastructure platform. It provides the execution layer between AI reasoning and real systems — ensuring that when agents suggest actions, those actions execute deterministically, with a complete audit trail, within controlled infrastructure boundaries.

The platform is designed for construction and infrastructure businesses that operate across multiple connected systems: ERPs, field management platforms, payroll providers, identity providers, and operational tools.

What problem does Demiton solve?

Enterprise systems don’t talk to each other reliably. Labour recorded in a field system doesn’t automatically appear in the ERP. Estimates don’t stay aligned with actuals. Payroll totals don’t reconcile with timesheet hours. Project codes drift between systems.

Fixing this today means manual exports, spreadsheets, and expensive integrations that break when APIs change.

Demiton replaces that with a governed execution layer: workflows that retrieve, transform, validate, and post data across systems — automatically, observably, and on a schedule.

What Demiton is not

Demiton is not an AI chatbot that writes to your ERP directly. The AI layer assists with reasoning, retrieval, and insight — but it cannot mutate external systems. All writes happen through deterministic workflows that run inside the execution runtime, triggered by users or schedules, with full audit trail.

This separation is intentional. AI reasoning should inform decisions; infrastructure should execute them.

Platform layers

Adapters — the integration boundary. Each adapter knows how to authenticate, communicate, and normalize data from one external system. Adapters never execute workflow logic.

Business objects — the semantic layer. Instead of working with raw API payloads, workflows operate on business objects: Project, LabourEntry, Vendor, PaymentFile. These objects bind to external resources and provide a consistent model across systems.

Workflows — the execution layer. A workflow is an ordered sequence of steps, each executing one verb (FETCH, TRANSFORM, GOVERN, PUSH). Workflows run asynchronously, with every state transition persisted to the database.

AI layer — the reasoning layer. Users interact with operational data through natural language. The AI layer retrieves relevant context, generates responses, and can suggest workflows — but never executes them directly.

Key design principles

  • Identity first — every action is traceable to a user or system identity
  • Deterministic failure — failures are explicit, logged, and recoverable; no silent retries
  • Simulation before production — all write adapters support dry-run mode
  • Pre-flight commit — the full workflow run graph is persisted to the database before execution begins
  • No workflow logic in HTTP threads — all async work runs through the ARQ worker

Next steps