AI & AutomationMCP · governed · audited

AI that sets up your operation — with context, permissions, and control.

We are all-in on AI, but not reckless with it. AI in MX drafts the form, the dashboard, the map layer, the charging rule, and the assignment from a plain-English ask — validated against live data, never overwriting what a person built, never acting until a person approves. It also answers, retrieves from approved documents, and reacts to events — always within the platform's permission, security, and audit model.

Discuss AI in your operation ↗See the MX platform

↳ Operational AI that does the set-up, on real data, under real permissions.

mx.assistant · role: ops-supervisor · zone: ANZ-east

permissions enforced

OPS-SUPERVISOR

Three damage assessments on Route 22 mention the same downed pole. Can we consolidate and dispatch one recovery crew?

↳ tool · mx.work.query(zone=Route 22, keyword=pole) ✓ permitted

↳ tool · mx.assets.lookup(POLE-09812) ✓ permitted

↳ tool · mx.docs.search(corpus=ops-sops · recovery procedures) ✓ permitted

ASSISTANT

Found 3 work orders (WO-48198, WO-48211, WO-48222) referencing POLE-09812. Severity: 1 high, 2 medium. SOP §4.7 allows consolidation when a single asset is the cause. Suggested: consolidate under WO-48198 (highest severity) and dispatch Crew B (closest with credential E-HV).

Grounded in: 3 work orders · SOP §4.7 · audit ref AUD-2026-05-14-09812

Consolidate & dispatchShow diffDismiss

§ 01 · Seven ways AI extends MX

AI that sets things up — and stays governed.

We are not introducing AI to MX for its own sake. It starts with the set-up work — forms, dashboards, map layers, charging rules, assignments — where the result is in front of you and nothing exists until a person approves it. All of it shares the platform's permission, audit, and governance model.

Drafted from a description

Describe the form you need — or upload a PDF, spreadsheet, Word document, CSV or text file, a scanned paper form, or a photo or screenshot of one, new or existing — and MX drafts the template. A drafted form can be tied to an asset and pre-fill itself from that record, with repeating sections and calculated fields built in. Describe the dashboard you want to see and it drafts that too: every widget validated against live data before it is shown, and nothing built by hand ever overwritten. The same goes for map layers, drafted or changed and published for the team that needs them, and for charging rules, drafted with a costed preview before anything is applied. Nothing is created until a person approves it.

EXAMPLES

› Draft a pole inspection form from this photo of our checklist, pre-filled from the pole's record
› Build me a dashboard of open work by zone, oldest first
› Draft a map layer of open inspections for the Sector 7 crew
› Draft a charging rule for generator hire on this project

Recommended, then confirmed

Recommends who should take an unassigned job from qualifications, location, current workload and existing appointments. It does not act on its own: the dispatcher confirms, and the record keeps who confirmed it and when.

EXAMPLES

› Who is best placed to take WO-48211 this afternoon?
› Which crew can reach Sub 14 before 15:00 with an E-HV credential?

Operational assistant

Ask questions about your operation in natural language. Search work, assets, locations, people, forms, and documents. Summarize activity, generate SITREPs, and interact with approved MX functionality — within the user's permission scope.

EXAMPLES

› Summarize overnight activity in Sector 7
› List all work orders blocked over 24h
› Show open inspections on POLE-09812

Help on the screen you are on

A help assistant available from the screen in front of you, answering about that screen — what it is for, what the fields mean, and how to finish the task at hand — so the answer arrives where the question came up rather than in a manual somewhere else.

EXAMPLES

› What does "hold" mean on this load?
› How do I add a photo field to this form?
› Why can't I complete this work order?

Document AI / RAG

Standard Operating Procedures, manuals, asset documents, vendor docs, and policies — queried through controlled retrieval. Answers are grounded in approved corpora with citations. Hallucinated sources rejected.

EXAMPLES

› What is the SOP for high-severity pole replacement?
› Show me the vendor manual section for pump P-918
› When was this asset last certified?

MCP architecture

A modern tool-integration architecture that lets AI agents call approved MX and external functions securely, while preserving role-based access control. Every tool has a scope, a permission boundary, and an audit trail.

EXAMPLES

› mx.work.query
› mx.assets.lookup
› mx.docs.search
› mx.crew.dispatch (gated)

Event-driven agents

Agents that react to changes in MX — a status, a completion, a missed inspection — and propose or perform actions under guardrails. Triage incidents, group related work, prepare reports, notify teams.

EXAMPLES

› Group 3 related damage assessments under one recovery order
› Draft a SITREP at 18:00 daily
› Flag work orders missing required photos

§ 02 · The same approval, every assistant

Shown first, costed first, undoable after.

AI features that simply act ask you to trust them. Every set-up assistant in MX asks for approval the same way: you see exactly what would change, what it would cost where money is involved, and you can take it back afterwards.

Preview

You see exactly what would change — what is created, what is edited — before anything is applied.

Costed dry run

Where money is involved, the change runs as a dry run first, so you see what it would charge before it charges anything.

Undo

Once applied, the change can be undone. An assistant making it does not make it permanent.

§ 03 · MCP architecture

Controlled AI access to your platform.

MX uses the Model Context Protocol to expose approved functions and integrations to AI agents. Each tool has a name, a scope, a permission boundary, and an audit trail. Agents cannot call what they have not been granted.

━ standard tool━ gated · requires human approval

§ 04 · AI usage & spend

What it costs, and who is using it.

AI usage and spend are visible in MX, so the question that comes up in every evaluation has an answer before it is asked: how much is being used, what it is costing, and where.

By feature

Which assistants and agents are being used, and what each one costs.

By model

Usage and spend for each model behind those features.

By team

Who is using AI, so the cost can be traced to the team that incurred it.

Exportable

The figures export, so they can go into a budget review or your own reporting.

§ 05 · Event-driven agents

AI that reacts — within guardrails.

Event-driven agents subscribe to changes in MX and act on them under explicit rules: a trigger, a guard condition, an action, and a permission boundary. Nothing posts silently. Critical actions require explicit confirmation.

TRIGGER

GUARD

ACTION

incident.created

severity == high

Draft summary · notify on-call · prepare initial dispatch list

work_order.completed

photos_missing

Flag for supervisor review · pause completion

damage_assessment.repeated

same asset · 3+ reports

Propose consolidation under recovery order · request approval

scheduled · daily 18:00

project.active

Generate SITREP across all active work · attach exceptions

asset.condition.drop

tier-1 asset

Open inspection work order · attach manual section

finance.line.outside_rule

amount > $500

Flag in Missing Rules · suggest matching rule from history

§ 06 · Governance

Eight controls every AI action passes through.

All-in on AI, but not reckless with it. The platform's permission, security, and audit model applies to every AI action — assistant prompts, RAG queries, MCP tool calls, agent triggers.

Role-based permissions

AI can only invoke MX functions the user already has access to. The same RBAC model as humans. Standing guidance for the AI can be set per project, not only centrally, by someone with the permission to set it.

Tool scopes

Each tool declares scope — corpus, zone, asset class. Out-of-scope calls are refused before execution.

Human-in-the-loop

Critical actions — dispatch, void, publish — require explicit human confirmation. Nothing posts silently.

Audit trail

Every prompt, tool call, parameter, response, and approval is logged with a reference ID and retained.

Grounded answers

Document RAG retrieves from approved corpora only. Hallucinated sources are rejected at generation time.

Rate limits

Per-role, per-tool rate limits prevent runaway agents from saturating downstream systems.

Reversibility

Agent-initiated changes are versioned and reversible. A rollback is the same as for human changes.

No platform bypass

AI does not see, write, or call anything outside the platform's access model. Ever.

§ 07 · Why we are credible on this

All-in on AI. Not reckless with it.

The ProceMX team has been working with Artificial Intelligence since the 1990s. That heritage means we know which AI capabilities translate into real operational value — and which ones create new risk surfaces enterprises cannot accept. MX uses modern AI strategically: to set up forms, dashboards, map layers, charging rules, and assignments from a plain-English ask, to ground answers in real operational data, and to take the routine work off people. It does not replace the audit trail, the permission model, or the human supervisor.

1990s

AI work in our team's history

20+ yrs

In emergency and critical operations

MCP

Standards-aligned tool integration

AI actions bypassing the permission model

AI with operational control

AI extends MX.
It does not bypass it.

Talk to us about AI for your operation — assistant, document RAG, MCP integrations, event-driven automation. We will show you the governance model.

Discuss AI in your operation ↗See the MX platform

AI that sets up your operation — with context, permissions, and control.

AI that sets things up — and stays governed.

Drafted from a description

Recommended, then confirmed

Operational assistant

Help on the screen you are on

Document AI / RAG

MCP architecture

Event-driven agents

Shown first, costed first, undoable after.

Preview

Costed dry run

Undo

Controlled AI access to your platform.

What it costs, and who is using it.

By feature

By model

By team

Exportable

AI that reacts — within guardrails.

Eight controls every AI action passes through.

Role-based permissions

Tool scopes

Human-in-the-loop

Audit trail

Grounded answers

Rate limits

Reversibility

No platform bypass

All-in on AI. Not reckless with it.

AI extends MX. It does not bypass it.

AI extends MX.
It does not bypass it.