Skip to main content
Smart Maintenance is a community Plugin App for AI-assisted maintenance intake. It turns natural-language repair reports into reviewable maintenance work orders, supports candidate service-data import, and keeps business-critical processing actions inside Workbench for human review.

​
When To Use It

  • Production equipment, facility, after-sales, or field-service issues need to be registered quickly.
  • Users describe a fault in natural language, and an Assistant should extract equipment, location, fault type, urgency, and initial diagnosis.
  • Candidate data such as device types, fault categories, departments, roles, parts, and locations needs to be imported for normalized Assistant use.
  • Maintenance supervisors need to supplement, confirm, reject, or close work orders from a review desk.
The first version focuses on an Xpert-side demo loop. It does not directly implement CMMS, FSM, ERP, dispatching, inventory, notifications, attachments, or SLA integration.

​
Plugin URL

Marketplace: Smart Maintenance

​
What The App Adds

​
System Architecture

Smart Maintenance separates natural-language reports, candidate service data, AI-generated work orders, and human review actions. Assistant tools save draft data and query information, while Workbench view actions own all governed human state transitions.

​
Processing Flow

The first version centers on an “AI-generated draft work order + human governance loop”. Imported service data improves field normalization. If fields are incomplete, the Assistant records completeness tips or supplement drafts instead of inventing missing facts.

​
1. Import candidate service data

Upload a service-data file in Workbench, prepare a draft payload, then let the Assistant call smart_maintenance_import_service_data to persist it. The catalog helps normalize device types, locations, departments, roles, parts, and fault categories. If no service data has been imported, the App falls back to a built-in mock catalog for demos and early validation.

​
2. Generate a reviewable work order

Open Smart Maintenance Assistant and describe the issue, for example:
The Assistant extracts structured fields and calls smart_maintenance_save_generated_work_order to save a draft work order. When fields are incomplete, it should record completeness tips instead of inventing device numbers, contacts, occurrence time, or location.

​
3. Review in Workbench

Open Smart Maintenance Workbench to review work-order details. Users can edit title, equipment, location, fault category, urgency, description, AI diagnosis, and suggested actions. Confirm processing, mark processed, and reject/close actions are Workbench-only actions, not Agent tools. This prevents the Assistant from claiming that a job has been dispatched, repaired, or closed.

​
4. Supplement and query work orders

When users provide additional information, the Assistant can call smart_maintenance_prepare_supplement_draft to save a supplement draft. Reviewers can then apply and save it manually in Workbench. For list and detail queries, the Assistant can use smart_maintenance_search_work_orders and smart_maintenance_get_work_order_detail.

​
Tool Boundaries

​
Data Quality

  • AI diagnosis is preliminary and should keep a field-verification tone.
  • Missing fields should be recorded in completenessTips, not guessed.
  • One report should usually create one work order; multi-device or multi-fault descriptions should be marked as multiple issues.
  • Confirmation, closure, and processing results require human action in Workbench.
  • Operation logs should preserve AI creation, manual edits, supplements, confirmation, completion, and rejection events.

​
Troubleshooting

​
Why can’t the Assistant close a work order directly?

Closure is a governed human action. The App keeps confirm, process, and reject/close actions in Workbench view actions so the Agent cannot overstep business state transitions.

​
Candidate devices or departments are inaccurate

Update and re-import the service-data file. The Assistant reads the current catalog; without imported data it only uses the demo mock catalog.

​
Can this connect to a real maintenance system?

This community App is a demo-ready loop. Production CMMS, FSM, ERP, inventory, notification, or SLA integration requires additional configuration, permissions, external APIs, state mapping, and audit design.