Sellpy

Search or browse one Sellpy market. Every garment is unique, so a row is a specific item and not a product line: it carries its own size, its own condition grade, the centimetres Sellpy measured on it and the defects Sellpy wrote down. 29 filters are offered and every single one was measured biting against the same unfiltered control in the same run (brand Gucci, 2,139 rows, exact count) — the four source attributes that look like filters and silently match nothing are NOT offered. The storefront's own availability gate is always applied, so a row is never returned with a missing price for your market; on brand Gucci that gate removed 21% of the indexed rows for EU and 26% for RO. Paging stops at the source's 2,400-result ceiling, which the engine clamps and reports rather than letting it quietly empty.

The complete record of one garment, from BOTH of the source's own back ends in one call: Sellpy's item database and its search index. That means the item's metadata translated into TEN languages, every photograph with the photo TYPE Sellpy shot it as (the robot turntable frames, the brand tag, the sole, the seam, the fabric close-up, the size label and one frame per DEFECT), the full measured defect list, the centimetre measurements, weight in kilos, Sellpy's own sellability score for the item, the intake / photographed / listed timestamps, the warehouse and shelf it sits on — and the price in ALL TWELVE markets at once, which is the only way to see that the seven euro storefronts share one price while SEK, DKK, PLN, CZK and RON are separate rounded lists (one measured item: EUR 48.50 / SEK 580 / DKK 390 / PLN 230 / CZK 1,250 / RON 280). Sellpy employee ids attached to the photos and the warehouse steps are stripped. An unknown id is NOT_FOUND. `language` changes the item's own words here (type, condition, colour, pattern, defect names) because they come from the record's ten translations; the AI keywords and concepts come from the English index, which is the only one carrying every market's price.

Sellpy's own vector model, the one that powers 'you might also like' and its semantic search — reached directly. Give it item ids and it returns the nearest garments with their scores; give it free-text phrases and it finds items by MEANING, which is a different thing from the keyword search in `search` ('black leather biker jacket' works without those words appearing anywhere). The ids come back resolved into full rows, priced for your market. Two measured honesty notes: the service IGNORES its own region argument (`SE`, `EU` and a nonsense `XX` returned identical lists) so the engine applies the market gate itself and tells you how many neighbours survived it, and an unknown item id gets HTTP 200 with an empty list, which is answered here as NOT_FOUND.

Sellpy's whole category tree with a LIVE item count on every node, in the market's own language — four levels, measured at 5 / 25 / 178 / 325 nodes on the EU surface, in one request. The counts come from the same gated search the `search` action runs, so they describe what is actually buyable in that country today (and the top-level words change with the market: 'Women' on EU, 'Kvinna' on SE). Counts on the broad nodes are the source's own estimates and every node says so.

The shape of any slice of the catalogue: pick the facets you want and get their values with counts, plus min/max/average for the numeric ones, for exactly the same search `search` would run. This is how you discover the source's own vocabulary before you filter on it — the brand tokens, the size tokens, the defect names, the style tags — and how you measure a market (one request returned the live price band EUR 2.99-11,906.00, mean 23.73, and waists 50-200 cm, mean 77.2). 34 facets are offered; `model` is deliberately absent because the source filters on it but never returns its values.

Resolve brand names the way the source spells them, with both counts that matter. This action is not a convenience, it is a correctness tool: filtering on `Levi's` returns zero items because the real tokens are `Levi Strauss & Co` and `Levi's Premium`. Each row gives the exact string, whether it is a sub-brand, Sellpy's LIFETIME item counter and — in one extra request for the whole page — how many are actually on sale in your market now. Those two numbers are not interchangeable: Gucci reads 43,466 lifetime and 2,139 live. `similar_to` additionally asks Sellpy's brand-similarity model for neighbours of a brand.

Autocomplete straight out of the storefront's own suggestion indexes, including the one that is really a SEARCH LOG: `queries` returns the phrases Sellpy's shoppers actually typed, each with its popularity and an item-count estimate, per language ('jeans' scored 1,023 on the English surface and 27,187 on the Swedish one, where 'jeansjacka' and 'jeansskjorta' also surface). `entities` returns brand and category entities with the storefront url that opens them, `brands` the brand names with lifetime counts, and `types` the garment-type autocomplete — which is SWEDISH ONLY whatever market you ask for, because the source keeps it on a shared language-less application, and this engine says so rather than pretending it localises.