Catalog AI
Agent · Product Enrichment

Turn a SKU and a name into a finished listing.

Enrichment reads live product pages, manufacturer sources, identifier records, and your own supplier documents, then writes the attributes you defined — with the source for each one attached.

Enrich 50 products freeBook a demo

Attribute set · Fasteners

What a finished product carries

DescriptionHTML · long form
SpecsHTML · factual only
Meta titleText · 55–75 chars
Meta descText · 150–160 chars
Thread pitchText · custom
CategoryGoogle taxonomy
The backlog

The listings are written. Just not by anyone yet.

There is a spreadsheet somewhere with the SKUs that never went live. It grows every time a supplier sends a new range and shrinks only when someone is given a quiet week to work through it.

The blocker is never the decision to sell the product. It is that publishing one requires a description, a specification block, a meta title, a category, and six attributes that nobody has time to research — multiplied by four thousand.

Enrichment is that research, run in parallel, with the sources kept so you can check the work.

Output

What comes back

The defaults cover a standard commerce product record. Everything below is editable, and you can add as many fields of your own as the catalog needs.

Description
Long-form product copy, structured with an opening paragraph and labelled sections, written to your prompt.
Specifications
A factual technical specification block, kept separate from marketing language so it can be rendered as a spec table.
Meta title
Search-length title, 55–75 characters, carrying the primary term.
Meta description
150–160 characters, written for the click rather than the crawler.
Custom attributes
Any field you define — voltage, thread pitch, allergen list, compliance mark — each with its own extraction prompt.
Images
Product imagery generated where your catalog has none.
Category
Classification against the Google Product Taxonomy.
Control

You write the brief.

An Attribute Set is the specification for a finished product: the fields, their types, and the prompt that produces each one. Sets attach per category and per brand, so a category of industrial fasteners is not enriched by the same instructions as a category of candles.

Attribute · Technical specifications · prompt

“Present only factual, verifiable technical details in a well-organised bullet-point format. Do not include marketing language, benefits, promotional claims, or descriptive storytelling. Do not repeat content already covered in the product description. Do not include pricing, shipping information, warranty, or return policy.”

That is a shipped default. Rewrite it, and every product enriched under that category follows the new brief.

Sources

Where the values come from

Nothing is generated from the product name alone. Each attribute is read from somewhere, and that somewhere is stored alongside the value.

Live product pages

Retailer and marketplace listings for the same SKU, read at the moment the job runs.

Manufacturer sources

Brand sites and published datasheets, which is usually where the real specification lives.

Identifier records

UPC, MPN, and GTIN lookups, used to confirm you are looking at the right product.

Your own documents

Supplier PDFs and spec sheets you upload, indexed and searched alongside the web.
Process

How a job runs

  1. STEP 1

    Add the products

    CSV import, AI Import from supplier files, the API, or one at a time. A name and a SKU is enough to start.

  2. STEP 2

    Define the fields

    Build an Attribute Set: the fields a finished product must carry and the prompt behind each one.

  3. STEP 3

    Run enrichment

    Select products, choose a model tier, review the cost estimate, and start. Progress updates live.

  4. STEP 4

    Review and export

    Inspect the values and their sources, edit anything you disagree with, then export CSV or read it back through the API.

Model tiers

Spend more where it matters.

The tier is chosen per job, not per account. Backfill the long tail on Budget and put the high-margin range through Ultra.

Budget
Bulk backfills and large low-margin rangesLowest cost per product
Pro
Production jobs. The one most teams run.About 1 credit per product
Ultra
Ambiguous specs, technical categories, high-value SKUsHighest cost per product
Variants

Enrich the family, not every child

Products can be modelled as parent-child families. Enrich the parent and the shared attributes flow down; variant-level attributes like size and colour stay where they belong. One job instead of forty near-identical ones.

Enrichment

Questions worth asking first

How accurate is it?
Accuracy depends on how much public information exists about the product. Where sources are rich, values are reliable and cited. Where they are thin, run Product Validation over the output — it scores each attribute so you can see exactly which ones deserve a human look.
Our category is unusually technical. Will it understand?
You write the prompt. Attribute Sets are defined per category and per brand, so a fastener category can demand thread pitch and tensile class while a food category demands allergens and net weight. Run a batch of 50 on Ultra and judge the output before committing.
Can it write in our tone of voice?
Yes — the prompt behind each attribute is yours to edit. The shipped defaults are written to avoid the usual generated-copy tells, and you can replace them entirely.
What happens to products it cannot find anything about?
They come back with the attributes it could establish and low confidence on the ones it could not. Nothing is invented to fill a gap.
Does it overwrite data we already have?
Enrichment writes into the attributes you select. Full history is kept per product, so you can see what a job changed and when.
How fast is a large job?
Products are processed in parallel with live progress. Large jobs run in the background and fire a webhook when they finish, so you do not have to sit and watch.
The rest of the platform

Six agents, one catalog

Product Validation

Is any of this still true?Score the data you already have, 0–100.

Market Analysis

Who else sells it, and for how much?Competitor price, MSRP, and stock status.

AI Import

How do I get this catalog in?Supplier PDFs and spreadsheets into products.

Brand Enrichment

Who makes it?Brand profiles written at catalog scale.

Category Enrichment

Where does it belong?Category copy, meta content, and slugs.

Run it on 50 of your worst SKUs.

Pick the products you would least like to write by hand and see what comes back. That is the honest test, and it is the one the free credits are for.

Start freeBook a demo

5 credits on signup · no card required