Twice a day,
the fleet checks itself.
OpenAI vision compares yesterday’s six photos against today’s. Clean shifts auto-close as reviewed. Suspicious ones go to the top of the queue with the damage already circled.
Pay for deep looks only when needed.
A cheap pre-screen runs over every shift twice a day. Clean shifts auto-close. Only the triggers go to the bigger model. Cost stays low. Verdicts stay sharp.
SCREEN
A lightweight pass over every shift. Three outcomes: OK (auto-close), SUSPICIOUS (escalate), or DAMAGE (escalate fast).
DETAIL
The bigger model takes a careful look at only the flagged shifts. Returns confidence, photo IDs, and a paragraph of reasoning.
“New diagonal scratch on right rear quarter panel, ~14cm, not present in prior reference scan. Verdict: DAMAGE. Confidence: 94%. Flagged photo IDs: P02, P03.”
Sales can enable it. Only you flip it on.
The plan flag ai_photo_analysis_enabled is decoupled from the opt-in flagai_photo_review_opt_in. Your Pro plan unlocks the feature. You decide when to start spending.
- Single toggle in Owner → Settings
- Spend caps configurable per organization
- Per-route or per-vehicle disable available
- Crontab is gated server-side, can be paused in one click