Check local 3D counting availability
The API runs on your own Open Inventory instance. Session-authenticated requests require a session on that instance.
Requires ai.count and item access. Availability requires BULK_3D_ENABLED and a separately provisioned Python/GPU runtime; the standard Docker image does not include it.
bearerAuthAuthorizationBearer <inv_ token>itemId*stringuuidX-Organization-ID?stringOrganization membership selected for this request. User-bound credentials may select any active membership; standalone tokens may only repeat their pinned organization ID.
uuidLocal runtime availability.
application/json- response
available*booleancurl -X GET 'https://inventory.example.com/api/v1/ai/count/3d?itemId=497f6eca-6276-4993-bfeb-53cbbbba6f08'{ "available": true}Count matching physical items in one transient image POST
Requires the ai scope. A selectable server-side Replicate model analyzes the image; text-aware models use itemHint as their query. The result is advisory and does not mutate stock. Supply a stable Idempotency-Key when retrying the same upload so a lost start response cannot create another paid prediction.
Start an experimental 3D bin estimate POST
Requires ai.count and item access. Upload 3–4 JPEG/PNG/WebP/AVIF photos, manual marks and explicit shape/packing assumptions. The first image is the top view. Runs on the configured local GPU; no paid inference API is called. Poll the returned jobId. Reuse the same key for identical retries. Deadline is ten minutes; results remain accessible for 24 hours. No stock is booked.