ExtraltExtralt

Glossary

Capture

A raw extraction output produced by Extract from a source URL. Captures preserve source-page product data in the original language, including images, options, SKUs, offers, identifiers, and capture-time aggregates such as price range and availability.

See Working with Captures.

Credit

The shared unit of usage billing in Extralt. Credits are consumed by product-specific actions: Extract uses 2 credits per successful product-page Capture, Enrich uses 1 credit per Capture processed, and AI assistant messages use 1 credit per started 1,000 model tokens.

See Credits & Billing.

Item

An enriched output produced by Enrich from a Capture. Items carry normalized English product content, taxonomy, attributes, signals, option and size interpretation, selected media, identifiers, transformed offers, normalized review state, and lineage. Items are the normal input to Extend.

Listing

A final, store-specific product record produced by Extend from Items. A Listing carries the source/store context for a product appearance, including store, URL, selected media, identifiers, option values, and the best enriched content available for that listing. The same Variant can appear as many Listings across stores.

See the Product Taxonomy Explorer for the category and attribute structure used during enrichment.

Offer

A price and availability observation attached to a Listing and Variant. Offers are append-only time-series records and include fields such as price, currency, condition, seller type, stock level, size, SKU, and availability. Current price and availability are computed from all offers for a Listing in the latest Extract run_started_at, so disappeared sizes or sellers do not remain current.

Organization

The top-level scope for all Extralt resources. Robots, schedules, runs, captures, credits, and API keys belong to an organization. Each org has its own subscription. Users can belong to multiple organizations.

See Account Setup.

Product

A logical product family that groups sibling Variants. Product is not stored as a separate public entity; it is represented by a shared product_id on final Variant rows. For example, multiple colorways of the same shoe can be separate Variants under one Product.

Review

A latest review aggregate attached to a Listing, Variant, and Store. Final Reviews store compact normalized state: review count plus score on a 0-100 scale. Raw average and scale remain on Captures for audit.

Robot

An AI-generated crawler compiled to Rust for a specific host and country. Robots understand a site's structure and extract structured product data. Reusable across multiple runs.

See Robots.

Robot Build

The process of creating a robot. If the site is already covered by Extralt's crawler library, you get instant access. Otherwise, Extralt's AI analyzes the target site, generates extraction logic, and compiles it to a Rust binary in 3-5 minutes.

See Robots.

Run

An extraction job that uses a robot to crawl start URLs and produce captures. Runs have a lifecycle (pending, running, completed, failed, stopped) and optionally a budget to limit credit usage. Runs can be created manually or by a schedule.

See Running Extractions.

Schedule

A recurring extraction automation. Links a robot to a cadence (e.g. every hour, every day, every week) and automatically creates runs on that schedule. Schedules can be paused and resumed, and auto-pause when credits run out.

See Schedules.

SKU

A source-specific purchasable configuration extracted from a Capture. A SKU is the concrete thing a customer can buy on the source site: it has identifiers, stock, and one or more Offers. SKUs are merchant-specific and are not used as cross-store product identity.

Store

A source/store identity such as a host and country combination. Stores provide source context for Listings, Offers, and Reviews.

Variant

A final identity for a deduplicated physical product configuration across stores. Variants are created by Extend, can have many Listings, and are the cross-store matching unit used for price history, reviews, and market analysis. Internal matching state and embeddings live in the internal Variant Cache, not on org-facing Variants.