Skip to main content

For Automation Engineers, OT Architects, and VP Engineering

SCADA-driven work execution: the architecture for turning historian data into ranked daily action.

Most "AI for oil & gas" pitches stop at the dashboard. This page is for the engineering-grade buyer who wants the architecture diagram. Five layers, the integration patterns we support, the OT-security posture, and the specific SCADA vendors we read from. No marketing, no per-well ML romance, just the engineering.

Security & Architecture DetailSee 40+ integrations
SOC 2 Type II in progress · IEC 62443 alignment · on-prem LLM option for governance-sensitive deployments

The 5-layer architecture

Ingest. Detect. Score. Route. Execute & Learn.

Each layer is its own set of agents working purpose-built models. Outputs of one feed inputs of the next. The whole thing is a closed loop: execution outcomes feed back into detection thresholds and scoring weights. The Detect and Score layers are the same engine behind production surveillance: every well watched continuously, every exception ranked in dollars. The operating model this architecture serves is management by exception: the five layers exist so a limited crew only works what deviates and pays.

An architecture that stops at layer two is where most SCADA deployments sit today: detection without scoring, which is why an alarm queue grows faster than anyone can read it. The five ways an exception queue stalls at scale covers what each of the three layers above detection actually fixes.

01 · Ingest

Read-only connections to SCADA historians, ERP, CMMS, GIS, and engineering simulators. OPC UA / MQTT / SQL Server / REST APIs / native vendor connectors for AVEVA PI, Inductive Automation Ignition, Cygnet, eLynX, Honeywell Experion, Emerson Ovation, FreeWave. Data QA agents detect sensor drift, stuck values, and bad readings before they corrupt downstream decisions.

Output
Normalized asset model: one well master, one tag dictionary, one allocation methodology
Integrations / modules
AVEVA PI · Ignition · Cygnet · eLynX · Honeywell · Emerson · OPC UA · MQTT · SQL · REST
02 · Detect

Per-well ML models learn each asset's unique operational baseline across pressures, flow rates, temperatures, runtime, dynacard shape, and intervention history. Predictive failure models flag equipment degradation days before failure. Anomaly detection scores deviation severity; reinforcement learning shrinks false-alarm rate over time. Threshold alarms become anomaly scores with confidence intervals.

Output
Ranked anomaly list with economic-impact estimate per item
Integrations / modules
Per-well ML · Predictive Maintenance · Reinforcement Learning · Data QA
03 · Score

Every flagged item receives a dollar-impact score: production at risk × commodity price × working interest × probability of intervention success × downside risk of deferral. Safety risk is a hard constraint (qualification gating) rather than a tradable weight. Live basis signals re-run scoring when the commodity basis moves materially.

Output
Economic-impact-ranked work list, refreshed nightly + intra-shift on material events
Integrations / modules
Economic Scoring · Price Signals · Field Safety (constraint)
04 · Route

Constraint-based optimization (the same solver class UPS and Amazon use for fleet routing) assembles scored tasks into crew-day plans. Inputs: yard locations, vehicle capacity, geography, time windows, crew qualifications (H2S, OQ, equipment-specific). Mid-shift re-optimization when conditions change: weather, breakdown, new high-priority alert.

Output
Crew-day plan in the truck cab by 6 AM; mobile-first, offline-capable
Integrations / modules
Route Optimization · Field Data Capture · Mobile (iOS/Android, offline cache)
05 · Execute & Learn

Field crews run from the ranked plan. Field Data Capture writes outcomes back (configurable: read-only, partial write-back, or full bidirectional per governance). Closed outcomes feed Reinforcement Learning so tomorrow's scoring is measurably better than today's.

Output
Closed-loop intelligence: the system gets more accurate over time
Integrations / modules
Field Data Capture · Reinforcement Learning · Operations Dashboard

Integration patterns we support

Four deployment topologies, governance-graded.

Different operators have different OT-security and data-residency constraints. These four patterns cover ~95% of real deployments. We can mix and match, e.g., read-only Phase 1 with on-prem LLM, then move to bidirectional in Phase 2. Where the watching happens matters as much as the topology: the staffed room reading these signals is the remote operations center, and the same execution layer serves it.

Read-only Phase 1

Initial deployment reads from existing systems-of-record without writing back. Lowest risk, fastest standup, satisfies most OT governance requirements. Most operators stay in this mode for 30-90 days post-go-live.

Default for first deployment

Configurable write-back

Phase 2 (optional). Field Data Capture writes closed-task outcomes, downtime codes, and field observations back to CMMS / production accounting on a configurable schedule. Bidirectional integration is per-system per-field: fine-grained, not all-or-nothing.

Phase 2 once data quality verified

On-prem LLM option

For operators with strict data-residency requirements (some midstream pipelines, federal contracts, sensitive M&A integration windows). The agentic layer can run on customer infrastructure rather than WorkSync cloud: same architecture, different deployment topology.

When governance requires; scoped with the implementation team

Hybrid SCADA + cloud

Common pattern: SCADA stays on-premises (per OT security best practice, IEC 62443, ISA-99), data exfiltrated through a single read-only egress firewall rule to the WorkSync cloud. No bidirectional control commands ever cross from cloud to OT.

Default architecture for cloud deployments

OT-Security posture

SOC 2 Type II
In progress, report targeted Oct 30, 2026
IEC 62443 alignment
OT security practices
Read-only OT egress
No control writes from cloud
On-prem LLM option
Customer-hosted agent layer
Per-field write governance
Fine-grained, not all-or-nothing
Audit log retention
Configurable per regulatory regime

Full architecture detail and audit reports available on request via /security or to qualified OT-architect reviewers under NDA.

For the architecture review

Bring an OT engineer. We’ll bring the architecture diagram.

Four weeks to stand up under the Impact Guarantee, license fees only when the metrics move. Then you run it for three-plus months and let the number prove itself. The stand-up includes a 90-minute architecture-review session with our OT lead. Bring your automation engineer, your IT architect, and any specific governance requirements.

See Your Wells RankedArchitecture & Security

Reply within 1 business day · 4-week pilot · license fees only when the metrics move

Frequently asked

What automation and IT leads ask first.

How does WorkSync connect to my existing SCADA?

Read-only by default. Native connectors for AVEVA PI, Inductive Automation Ignition, Cygnet, eLynX, Honeywell Experion, Emerson Ovation, FreeWave, and any system that supports OPC UA, MQTT, or SQL Server. Setup is typically under 1 week. We do not require any changes to your SCADA configuration; we read tags and historize them in our normalized asset model.

Does WorkSync write control commands back to SCADA?

No. WorkSync is read-only with respect to OT control systems by design. We never issue setpoint changes, valve commands, or any other control writes to SCADA. Outputs go to the truck-cab tablet (a person decides whether to act), to CMMS work orders (configurable write-back, optional), and to production accounting (configurable write-back, optional). The agentic layer never crosses the IT/OT boundary in the control direction.

What about IEC 62443 / ISA-99 OT security alignment?

WorkSync is architected around IEC 62443 zone-and-conduit principles. SCADA stays in the Level 2 (control) zone; data egresses through a single read-only firewall rule to a Level 4 (enterprise) DMZ; the agentic processing happens entirely in the enterprise / cloud zone. SOC 2 Type II is in progress (report targeted Oct 30, 2026) and we provide architecture detail to OT architects under NDA.

Can the LLM / agentic layer run on-premises?

Yes, for operators with strict data-residency requirements (some midstream pipelines, federal contracts, sensitive M&A integration windows). The on-prem deployment runs the same architecture on customer infrastructure rather than WorkSync cloud, and it is scoped with the implementation team when you plan your pilot. Common pattern: start cloud-hosted for Phase 1, evaluate on-prem migration in Phase 2 if governance review requires.

How long does the integration take?

Integration: under 1 week for SCADA + production accounting + CMMS connections. Standup: 2 weeks total including data validation and per-well ML model training. Field rollout: 4 weeks. Total kickoff to ranked plan in the truck cab is typically 4 weeks.

What if we have a custom or non-standard historian?

In addition to native connectors, we can read from any SQL Server, PostgreSQL, MySQL, time-series DB (InfluxDB, TimescaleDB, ClickHouse), or REST API. Custom integrations are a 2-3 week scope on top of standard deployment. Cygnet, eLynX, FreeWave variants, and Honeywell legacy systems are routinely handled.