SYSTEM: OPERATIONAL
·
SYS:INDUSTRIALCLAW.AISTATUS:NOMINALOT/IT CONNECTORS: 150+AUTONOMOUS OPERATION: 15+ DAYSINDUSTRIES: MINING · OIL & GAS · ENERGYGOVERNED AUTONOMY: ENFORCEDAUDIT TRAIL: IMMUTABLEBLAST RADIUS: ZERO

The Platform

How IndustrialClaw works.

The architecture inside the node -- the governance loop, the autonomy framework, the OT connectivity layer, and the knowledge model that turns your plant's operational history into a compounding moat.

New to IndustrialClaw? Start here →

What changes in your stack -- and what does not.

IndustrialClaw sits above your existing OT infrastructure. It does not replace it. Each row shows exactly where the boundary is.

Layer What it does today With IndustrialClaw
PLC / DCS / SCADA Executes reliably. Supervises the process. Generates alarms. Unchanged -- IndustrialClaw proposes through supervisory paths, never bypasses control systems.
Historian / MES / CMMS Records what happened. Schedules maintenance. Tracks production. Source data for agent reasoning. Agents write back to CMMS (work orders), MES (schedule updates).
Software Defined Automation (SDA) IEC 61499 function blocks, OPC-UA, containerised control on standard edge compute. Full action surface: in SDA environments, agents can propose and execute control logic changes through governed supervisory paths.
ChatGPT / Copilot Responds when queried. Advisory only. No OT awareness. Replaced for plant use cases -- IndustrialClaw runs persistently, understands OT context, acts within declared boundaries.
Open-source agent frameworks Capable building blocks. No OT governance, no audit trail, unacceptable blast radius. Replaced -- IndustrialClaw is the enterprise-managed, governed alternative built for safety-critical environments.

Every agent action follows the same loop.

A deterministic, auditable sequence. Not a black-box inference chain.

OBSERVE MONITORING

Agent reads from connected data sources on trigger or schedule. Historians, SCADA, CMMS, DCS. Builds a structured operational context window.

PROPOSE GENERATING

Agent generates a candidate action: raise a work order, send a briefing, adjust a setpoint, escalate to an operator. Explicit and auditable before execution.

VALIDATE EVALUATING

The governance layer evaluates the proposal against deontic rules, permission boundaries, spend caps, and network constraints. Validation runs on projected outcomes -- not reasoning paths.

EXECUTE AUTHORISED

Only validated actions reach the execution layer. Write capabilities require explicit role-based authorisation. The proposal, validation decision, and execution record are written to the immutable audit trail.

Start where you are. Progress at your pace.

HAS (Human Agency Scale) is the framework for progressive autonomy -- from monitoring to fully autonomous operations, on one platform.

HAS-01

MONITORING

Agents observe and alert. Humans decide and act. Zero agent-initiated actions.

READ-ONLY
HAS-02

ADVISORY

Agents observe, diagnose, and recommend. Humans approve. Agents explain their reasoning.

RECOMMEND-ONLY
HAS-03

ASSISTED ACTION

Agents initiate routine actions (create work orders, send briefings) within approved boundaries. Humans retain override authority.

BOUNDED-WRITE
HAS-04

SUPERVISED AUTONOMY

Agents handle the full exception workflow. Humans monitor and intervene only on exception. Operations run between shifts.

SUPERVISED
HAS-05

FULL AUTONOMOUS MULTI-AGENT

Agent networks coordinate autonomously across the operation. Humans set objectives and governance parameters. Full audit trail maintained.

FULL-AUTONOMY

Tier 1 oil & gas operator running AI Operated Control Room under safety-critical autonomous operations

Operational Identity Model

Your institutional knowledge becomes the moat.

The Operational Identity Model (OIM) is the knowledge layer that separates IndustrialClaw from generic AI platforms. OIM captures your operational processes, asset relationships, alarm patterns, and institutional decision logic -- and encodes them into agent behaviour that compounds with every deployment.

Generic AI tools are trained on the general internet. IndustrialClaw agents are trained on your plant, your historian, your maintenance history, and your engineering decisions. That knowledge is yours -- not ours. It becomes progressively harder to replicate as the deployment matures.

Processes

Operational procedures, decision sequences, and escalation paths -- encoded as agent behaviour.

Asset relationships

Equipment hierarchies, failure mode dependencies, and maintenance schedules.

Institutional knowledge

The heuristics experienced operators carry. Preserved, not lost at shift handover.

Compounds over time

Every deployment makes the OIM richer. The longer you run, the wider the moat.

Connected to the systems your operation actually runs on.

150+ OT/IT connectors via XMPro DataStreams -- the widest connectivity library in industrial operations.

Connected systems

OSIsoft PIAVEVA PI SystemIgnitionEmerson DeltaVHoneywell ExperionDistributed Control Systems (DCS)Rockwell FactoryTalkSiemens SIMATICSAP PMIBM MaximoInfor EAMServiceNowMQTTOPC-UAModbusEtherNet/IPREST APIsSQL / time-series databasesAzure IoT HubAWS IoT Core

Why connectivity matters

The intelligence of an agent is bounded by the data it can see and the systems it can act on. IndustrialClaw starts with the widest OT connectivity library available, so agents aren't artificially limited by integration gaps.

Connections are bidirectional: agents read from historians, SCADA, and CMMS -- and write back to work order systems, notification channels, and approved control setpoints within declared permission boundaries.

Skills are infrastructure. Not plugins.

In open-source frameworks, any skill can be added and executed. In IndustrialClaw, skills are governed infrastructure:

  • Pre-vetted for OT safety -- No skill in the library can issue unconstrained writes.
  • Version-pinned -- Operators know exactly which skill version is running at every moment.
  • Hash-verified -- Skills cannot be modified post-deployment without breaking the hash.
  • Role-gated -- Skills require explicit authorisation for the asset class they operate on.

Skill manifest -- example

SKILL: alarm-triage-v2.3.1

STATUS: APPROVED

HASH: a8f2c91d...

PERMISSIONS: READ historian, READ alarms

WRITE: DISABLED

LAST_AUDITED: 2026-01-15

Powered by XMPro

Production-proven at scale.

IndustrialClaw nodes are powered by XMPro MAGS - the orchestration platform running governed autonomous agents at Tier 1 operators across mining, oil & gas, and energy. Each node you deploy at the plant is orchestrated by the same platform backing an AI Operated Control Room at a safety-critical petrochemical operation and enterprise-wide process control monitoring across thousands of control loops.

AI Operated Control Room

Tier 1 oil & gas operator -- safety-critical autonomous operations

15+ days continuous

Autonomous operation in a safety-critical petrochemical process

Thousands of loops

Control loops monitored across a Tier 1 mining producer's full operation

Learn more at xmpro.com →
Active Value Optimisation

The governance layer doesn't just prevent bad outcomes.

It actively generates measurable commercial value. Every agent transaction is an opportunity to optimise within governed boundaries -- automatically, at scale, with full audit trail.

Governed Spend

Hard spend caps and circuit-breakers govern every agent action. CFOs get complete visibility into AI operational costs -- per agent, per shift, per asset class.

Staged Investment Model

Start at HAS 1 with a single use case. Each stage generates its own auditable proof before the next investment decision. The governance architecture is in place from day one -- autonomy expands as operational trust is earned.

Compounding Knowledge Moat

Every deployment enriches the Operational Identity Model with your plant's specific processes, failure patterns, and decision logic. That knowledge is yours -- and progressively harder to replicate as the deployment matures.

See the platform in your operational context.

We'll map IndustrialClaw to your current stack, your use cases, and your governance requirements.

Talk to us

Apply for early access — 2026 cohort

Enterprise, heavy asset & mission-critical industries only. Senior decision makers prioritised. Acceptance at XMPro's discretion.

By submitting you agree to receive communications from XMPro. Applications reviewed — acceptance at XMPro's discretion.