Build on ShelfPro
ShelfPro turns a shelf photo into structured retail data. These docs cover how a capture moves through the platform, and the REST contract your systems code against.
Quickstart
Authenticate, request an analysis, and read the result back. About ten minutes.
Core concepts
The image lifecycle, the hybrid write path, reconciliation precedence, and why coordinates are stored twice.
API reference
All 28 operations and 105 schemas, generated from the contract.
How a capture flows
Capture
The device writes an image document and uploads the bytes to Cloud Storage, stamping the object with a trace id. Offline captures queue locally and drain when connectivity returns.
Analyze
A storage trigger calls POST …:analyze. The call returns 202 with a job handle; recognition runs asynchronously.
Write
The vision service submits its result through PUT …/analysis — exactly the same endpoint an on-device model uses. One canonical write path, regardless of where inference ran.
Reconcile
If the image already has an analysis, precedence decides which becomes canonical. The loser is retained for audit and training.
Correct
An operator reviewing the result submits a delta. That pins the analysis to human and feeds the next training round.
Base URLs
https://api.theshelfpro.comhttp://localhost:8080