Skip to main content
Pre-launch · in testing — WorkSmartOS MaintenanceOS is not commercially generally available yet.

Pre-launch · Preview

The safety rules, written down.

AI assists with intake. Fixed server rules control the lifecycle. Authorised people clear dangerous repairs.

If you smell gas right now or suspect carbon monoxide

Get into fresh air immediately. Do not touch electrical switches or ignite flames.

Carbon monoxide: know the signs. Headaches, nausea, dizziness, breathlessness, collapse. If you suspect it: turn off the appliance if it's safe to, open doors and windows, go outside, and call the National Gas Emergency Service on 0800 111 999. Don't go back inside until you've been told it's safe. Call NHS 111 for medical advice, or 999 if someone is finding it hard to breathe, has suddenly become confused or has lost consciousness.

Safe automation

AI assists. Server rules control. Authorised people approve.

MaintenanceOS collects and prepares a report. It does not give the model authority over safety, money, or a dispatch. The product is still pre-launch.

  1. AI assists

    Capture and structure

    AI can collect caller details, structure maintenance requests, classify routine issues, and prepare information for follow-up. Voice intake remains in commissioning preview.

  2. Server rules control

    Enforce the boundaries

    Fixed rules on our server own safety latches, the progress of a job, approvals, escalation, and anything to do with money. The model cannot override them.

  3. Authorised people approve

    Clear consequential actions

    Critical safety requires authorised human clearance. Human-review queues and explicit transfer keep consequential judgement with people rather than the model.

AI can automate

  • Collect caller and maintenance details into structured intake
  • Classify routine maintenance information in UK terminology
  • Summarise and prepare records for human-controlled follow-up

Rules or people must clear

  • Critical-safety clearance
  • Lifecycle and approval transitions
  • Escalation for a person to review reports flagged as dangerous
  • Any payment, refund, compensation, fee-waiver, or client-money action

The AI must never

  • Invent a booking, dispatch, engineer arrival, refund, or account change
  • Override a safety, lifecycle, approval, or money prohibition
  • Give safety assurances beyond approved guidance

Deep details & technical boundaries

Operational rules, statutory records, and continuous verification.

How the critical safety latch works

When an intake report indicates hazards to life or property—such as gas, electrical sparking, or severe leaks—the server applies an immutable latch.

This latch halts automated transitions and requires an authorised human supervisor to review and explicitly clear the ticket before contractor dispatch.

Auditable evidence and job records

Every caller interaction, transcription extract, hazard detection score, and staff clearance decision is permanently logged with timestamps.

Records provide complete traceability for statutory compliance and Housing Ombudsman evidence without tampering.

Tenant isolation and multi-organisation data controls

Data is strictly partitioned by organisation ID at the database layer. No cross-organisation data sharing or cross-tenant leakage is permitted.

Automated continuous integration regressions enforce database partitioning across all release builds.

Checkpoint test verification (196 automated test suite)

Every deployment must pass 196 automated safety and claims verification tests before shipping.

Tests verify that safety latches fail closed, payment operations remain prohibited, and emergency routing cannot be bypassed.