greyparrot

Point any camera at your stock. Get facts your ERP understands.

Photos, scans and camera feeds become counts, conditions and locations, in your own item codes — with confidence and photo evidence on every fact. One vision layer for warehouses, stores, hubs, kitchens and plantations, designed to connect to your ERP and WMS. Your ERP keeps its rules; we supply the truth from the floor.

Get early accessSee the API
Camera 3Aisle 2, shelf B. Photo taken on a phone.
SKU-10233 ×4SKU-10233 ×2SKU-10412 ×3needs review
Facts sent to your ERP4 facts, 1 for review
  • SKU-10233Cola 330 ml ×4exp 12 Dec
  • SKU-10233Cola 330 ml ×2exp in 3 days
  • SKU-10412Milk 1 L ×3exp 30 Sep
  • ?Unclear ×143% sure · sent for review

Every shift starts with the same five questions

From dark stores to distribution centres to field collection points, most operational work comes down to five questions — answered today by eye, on paper, and typed in again later. High turnover means the knowledge walks out the door. The ERP drifts from reality, and every downstream decision inherits the error.

What is this?

An unlabelled product, a torn barcode, a damaged label, a fruit bunch's ripeness grade.

Where is it?

Which bin, which shelf, which tote slot, which cage, which field block — in your location codes.

How many?

Cycle counts, shelf audits, receiving checks, tote counts, parcel tallies, bunch counts at collection.

Is it OK?

Expiry and best-before dates, damage, freshness, temperature abuse signs, grading and sort quality.

Where should it go?

Put-away to the right bin, replenishment to the right shelf, parcels to the right cage or route bay.

The vision layer answers the first four with facts your ERP can consume — counts, conditions and locations in your item codes — so your people and your ERP rules can get on with the fifth.

How it works

  1. Sync your item list

    Your ERP or WMS sends its item master — codes, barcodes, GTINs, names, locations. Answers come back in that language from day one. Odoo connector coming first; API and GS1 EPCIS 2.0 export work today.

  2. Compose a formula

    A formula is a versioned recipe: detect, identify, read, classify, count. Build it in the studio or in YAML. Swap vision models without rewriting the recipe.

  3. Test before you trust

    Run every formula on sample photos in the test bench. Compare against expected counts, score precision and recall, and publish only when it passes. Corrections become test cases.

  4. Point your devices at it

    Phones, fixed cameras and barcode scanners today. Every source produces the same observation record. Built so RFID, drones and robots can plug in later without starting over.

  5. Facts flow back

    Counts, conditions and locations flow through the API or as GS1 EPCIS 2.0 events, each with a confidence score and the photo as evidence. Low-confidence items go for review, not into stock.

  6. See it live, run it continuously

    Every site appears in 3D or 2D, coloured by live status: fine, expiring, unsure, empty. Schedule scans per area, track world state per location, and let the ERP apply its reorder, write-off and replenishment rules.

See the whole site, not a spreadsheet

Operations leaders see status at a glance; auditors and engineers drill to the evidence. Every shelf, rack, cage and collection point is coloured by what is happening now. Layouts come from 58 ready-made parts across 7 packs, client-defined custom parts, tracing over floor-plan PDFs and photos, or plain spoken description — with drafts, checks and versions.

  • 3D for everyone, 2D for precise work
  • 58 ready-made parts across 7 packs, plus your own custom parts
  • Trace over a PDF or photo of the floor plan, with drafts and version history
  • Describe a layout by voice or text, with any OpenAI-compatible model
  • Live status per location: fine, expiring, unsure, empty
  • Every count links back to its photo evidence
“A 40 by 30 metre dark store. Six aisles of tote racks, a cold room at the back left, and the handover counter by the entrance.”
Cold roomHandover counter

One engine for every kind of site

The engine is the same everywhere: formulas, item codes, observations, evidence, standards. What changes per industry is the item list, the formulas and the floor parts. Run a dark store and a plantation on the same platform, and add your own parts and formulas when the standard packs are not enough.

All use cases

Image in, facts out

A clean, API-first platform with versioned formulas, a test bench with precision and recall scoring, knowledge packs synced from your item master, and output in your item codes with GS1 identifiers. Bring any vision model behind one interface; the studio itself runs on the same endpoints you use.

  • Any vision model, including your own — models are interchangeable parts
  • Formulas are versioned, validated and testable before publishing
  • GS1 EPCIS 2.0 export with GTIN, GLN and SSCC identifiers
  • Confidence score and photo evidence on every fact
  • Same observation shape from cameras, scanners — and later RFID and robots
  • Corrections feed back as test cases, so quality compounds
Read the developer overview

Response from POST /run/shelf-count

{
  "observation": {
    "location": "store-12/aisle-2/shelf-B",
    "formula": { "id": "shelf-count", "version": 3 },
    "detections": [
      { "item_code": "SKU-10233", "gtin": "09501101530003",
        "qty": 4, "attrs": { "expiry": "2026-12-01" }, "confidence": 0.95 },
      { "item_code": "SKU-10233", "gtin": "09501101530003",
        "qty": 2, "attrs": { "expiry": "2026-10-01" }, "confidence": 0.95 }
    ]
  },
  "trace": [ { "use": "detect", "detections": 10 }, { "use": "count", "detections": 2 } ]
}

We're building with our first design partners

If you run sites where stock is counted by hand, or you implement ERP and WMS systems for companies that do — warehouses, stores, dark stores, hubs, kitchens, plantations — we would like to build this with you. One site, one process, direct line to the team. No cost during the design-partner phase.

Get early access