Structured product data · AI classification

AI that fills in your product attributes — and never makes them up.

Upload your catalog. ATTRIBUTOR assigns attributes from your own controlled vocabulary — with confidence scores and evidence for every pick — then exports straight back into Shopify.

CSV in · reviewed metafields out · your API keys, your data
classifying…
Product 0148
series 0148 · 6 variants · 1 classification
✓ VALIDATED AGAINST REGISTRY · FANNED OUT TO 6 VARIANTS
01 · CLOSED VOCABULARY

Only your values exist

The AI picks from the attribute lists you define — nothing else. Anything outside the registry is rejected before it can touch your data, so filters and feeds never get polluted.

02 · EVIDENCE REQUIRED

Every pick shows its work

Each assignment carries a confidence score and the exact phrase or image observation that supports it. Reviewing takes a glance, not a product-page deep dive.

03 · STRICT WHERE IT MATTERS

No guessing on facts

Mark an attribute strict and it's only assigned when the source data explicitly supports it. An empty cell tells you a spec is missing — it never becomes a hallucinated claim.

Five steps from spreadsheet to structured catalog

Built for real product exports — a raw Shopify CSV works as-is.

01

Upload

Drop a CSV with titles, descriptions and image URLs. Columns are auto-detected.

02

Map

Product types point at your attribute groups. Exact matches map themselves.

03

Classify

One AI call per product series — text and vision — never per variant.

04

Review

A confidence-colored grid. Bulk-approve the greens, eyeball the ambers.

05

Export

Metafield-ready CSV that imports straight into Shopify and your feeds.

Built around how catalogs actually behave

Thousands of SKUs are usually a few hundred products wearing different colors. ATTRIBUTOR treats them that way.

SERIES-AWARE

Classify once, fan out

Variants sharing a series are classified a single time and inherit the result — turning a 200,000-row problem into a few thousand calls. Review edits propagate series-wide too.

VISION PASS

Sees what text can't say

Look-and-feel attributes are judged from the product image by a vision model, while factual ones stay on a fast, cheap text model.

TWO TRUST MODES

Infer vs. strict, per attribute

Let AI use judgment on style and use-case. Require explicit source evidence for claims that cause returns or liability. You choose, per attribute.

REVIEW GRID

Human in the loop, fast

Green, amber, red by confidence. Click any cell to see evidence and override with an allowed value. Save a batch mid-review and resume later.

SHOPIFY-NATIVE OUTPUT

Metafields, not spreadsheets

Exports in metafield-import format mapping 1:1 to choice-list definitions — ready for Search & Discovery filters, shopping feeds and AI search surfaces.

YOUR REGISTRY

Fully editable taxonomy

Add product types, attributes and values in settings — no code changes. The AI's world updates the moment you save.

Private by design

Your taxonomy stays yours

The attribute framework a store develops is competitive knowledge — it shapes filters, feeds and how AI shopping surfaces describe your products. ATTRIBUTOR is self-hosted: your registry, your catalog and your API keys live on your own infrastructure and are sent to nobody but the model you choose.

Stop typing attributes. Start approving them.

Currently in private use. Interested in running it on your catalog?

Request access