Schemas and fields

Describe a product family with a schema of typed fields, and know which fields the AI matches an enquiry on.

Updated 23 September 2026

On this page

Why: A product the AI cannot read field by field is a product it cannot match or filter.
How: Take one from the library, read one from your file or build your own on the Schemas view of /app/catalog.

Core fields

Every product carries these, whatever its schema. Price and cost are stored in integer minor units (pence or cents).

Key	Label	Needed
`sku`	SKU	Required
`title`	Title	Required
`category`	Category	Optional
`uom`	Unit	Optional
`price`	Price	Optional
`cost`	Cost	Optional
`currency`	Currency	Optional
`stock_qty`	Stock	Optional
`lead_time_days`	Lead time (days)	Optional

Three ways to start a schema

Choose	What happens
Browse the library	Adds an editable copy of a ready made schema. The library offers what your presets carry, or every schema when no preset is on.
Build from a file	Reads the columns of a price list or stock export into a schema for you to check.
Build your own	Starts empty, with a live preview of the product form and the filters.

Fields

A field has a Label, a Key (snake_case) such as diameter_mm and a Type: number, text, select, boolean or dimension. A number or dimension carries a Unit; a select carries its Options (comma separated).

Synonyms (comma separated) are the other names a column goes by (dia, d1). The import wizard maps headings with them.
Show in facet rail puts the field in the grid's filters.
Required marks the field. Only SKU and Title are enforced: a product without a required schema field is still saved.

Match rules

A match rule tells the AI how a field decides a match. Only library schemas carry them (the field builder has no match controls); a schema card reads matched by the AI when it has them and no match rules yet when it does not.

Role	Meaning
hard	The product must satisfy the requested value, or it is not a candidate.
soft	Counts towards the fit score by its weight, but never excludes.
ignore	Not used in matching.

A numeric rule can carry a tolerance (absolute or a percentage) and a direction, such as meet or exceed. Equivalent values (TiAlN and AlTiN) come from Kabaido's knowledge packs or the field's own option aliases, never from a guess, and show as an alternative rather than an exact match.

A schema read from a file never has match rules written for it: a rule decides what the AI will quote, so it is never inferred. Choose Request a schema in the library to have one built with its rules.

From a real test file

Schema changes are additive. Removing a field archives it and its values stay on the product; archiving a schema deletes nothing.

PreviousDocuments NextImporting products