Maintenance Agent

The Maintenance Agent predicts service needs before failures happen — daily fleet-wide checks, per-unit predictions on demand, and automatic routing of damaged equipment to maintenance.

How It Works

1. Daily batch check — every morning, the agent reviews your whole fleet: current hours, condition, service history, and open maintenance records
2. Per-unit predictions — ask for a prediction on any unit (from the equipment page or GET /api/v1/equipment/{id}/maintenance-prediction): what's likely to need service, and how soon
3. Fleet alerts — units approaching or past service thresholds appear in your fleet alerts, alongside overdue returns and low-utilization flags

Data Sources

• Telematics: with Trackunit connected, engine hours and fault codes sync every 15 minutes — predictions run on real usage
• Manual hours: without telematics, keep current_hours updated (returns are a natural checkpoint)
• Service history: completed maintenance records inform what's due next
• Inspections: checkout/return inspection findings feed condition assessment

Automatic Damage Routing

When a return inspection finds damage, the damage assessment:

1. Creates a maintenance record for the repair automatically
2. Moves the equipment to maintenance status so it can't be rebooked while broken
3. Records the damage in the insurance intelligence learning loop, which adjusts future insurance premiums for that category

Acting on Predictions

1. Review the prediction and the fleet alerts
2. Create a maintenance record (preventive type) with the scheduled date
3. When the work is done, complete the record with labor hours and costs — this feeds the next prediction cycle

Schedule

Task	When
Daily fleet maintenance check (all orgs)	Every morning
Maintenance-due threshold check	Daily
Per-unit prediction	On demand

Coming Soon

• Configurable alert thresholds ("warn me 14 days out")
• Recurring maintenance schedules per equipment type
• Work-order assignment and vendor notifications

Troubleshooting

Predictions seem off?

• Predictions are only as good as the hours data — connect telematics or keep hours current manually
• New units with no history get conservative predictions until data accumulates

Alert for something you already fixed?

• Log the completed maintenance record — the next daily check clears the alert

No predictions at all?

• The agent runs with graceful fallback: if AI is temporarily unavailable, threshold-based checks still run, but predictions may be simpler

See Maintenance for record management.