Architecture

Yard is a standalone asset and operations platform. The core object is an asset. Devices, vehicles, machines, sensors, and equipment are asset kinds. Zyvor products plug in as optional connectors — Yard installs and runs alone.

Data model

Organization · User · APIKey · Site · Asset · Capability · Observation · Event · WorkOrder · Incident · SeverityPolicy · ActionRequest · Connector · Automation

Every observation stores source, unit, observed_at, received_at, and quality. Offline data is marked stale rather than healthy.

SeverityPolicy

Rows map capability, automation, or default matches (ordered by priority) to incident severity and runbook text when automations open incidents.

Remote actions require a session, expire, carry an idempotency key, and record an outcome. Connectors advertise the actions they can execute.

Boundaries

Component	Responsibility
Yard	Asset registry, sites, workflows, incidents, shared UI
Device Agent	Hardware discovery, health, local diagnostics
Nodra connector	Decoded industrial telemetry and buffered events
Zyvor Fleet connector	Lifecycle requests and progress (catalog today)
OTA connector	Campaign display; execution stays elsewhere
HTTP / simulator	Zero-dependency evaluation path

Device Agent reports physical capability. Nodra interprets protocols. Fleet owns desired state when wired. Yard preserves those lines and adds the operations surface: health, incidents, and work orders.

Live ops loop

• Background stale ticker marks missed heartbeats without waiting for an Overview refresh
• Console pages subscribe to GET /api/v1/stream (SSE)
• Automations trigger on a literal threshold, a capability's own declared Min/Max range (capability_min/capability_max, so the range lives in one place), or a stale heartbeat — and can open_incident, notify (audit), or call a webhook/slack/ email/pagerduty action, all behind one shared dispatch path so a new action type is one addition, not two
• Opened incidents inherit severity + runbook from severity policies

Registry and map

• Assets support create/edit/delete, capability editing, and bulk CSV/JSON import-export
• MapLibre map uses clustered pin layers for dense sites/assets

Storage

• SQLite by default (go run, CI, Compose)
• Postgres via YARD_DATABASE_URL (Compose --profile postgres)
• Schema changes are versioned, transactional migrations tracked in a schema_migrations table — not ad hoc CREATE TABLE IF NOT EXISTS — so upgrades are ordered and idempotent across both backends
• scripts/backup.sh/restore.sh snapshot either backend (SQLite VACUUM INTO; Postgres pg_dump/pg_restore) — see Deploy → Backups

Related articles

APIEndpoints for every object in the data model→Console featuresBulk IO, map clustering, runbooks→ConnectorsHow optional connectors plug in→