Platform
Three layers. Take one, or take the stack.
Closest to your user, the components they look at. Underneath, 24 inference endpoints that read the vehicle. Underneath that, the 7 data domains a model cannot see. Nothing here requires adopting the rest.
The stack
One key, one schema, one invoice.
The layers are numbered by distance from your customer, not by importance.
Layer 1
Components
The front end your customer sees
Four SDKs — player, capture, review, report — shipped for Web, iOS and Android, so nobody on your team writes a WebGL viewer or a guided-capture camera flow.
Open Layer 1 →
Layer 2
Models
24 inference endpoints
Each endpoint is priced and rate-limited on its own. Call one in isolation, or chain the set — nothing assumes you adopted the others.
Open Layer 2 →
Layer 3
Data
7 domains, one schema
Sourced from best-in-class providers per country, then normalised into one schema behind the same key. You integrate once instead of once per country.
Open Layer 3 →The bundle
Or one call through all three.
POST /v1/inspections chains 8 stages across the layers — walkaround video in, a priced and annotated vehicle out, hotspots anchored for the viewer.
- 01Onboard walkaround
video-onboarding - 02Isolate the vehicle
car-segmentation - 03Decompose parts
parts-decomposition - 04Detect damage
damage-detection - 05Grade severity
severity-estimation - 06Price the repair
reparation-pricing - 07Render for listing
image-backgrounding - 08Anchor hotspots
viewer-hotspoter
Pick the layer you need first.
Most teams start at Layer 2 with a single endpoint, then move outward. The key is the same either way.