The native engine indexes — it doesn't understand
The one shipped with your CMS reads the catalog, but it ignores typos, your business synonyms, and the order your products should appear in.
The one shipped with your CMS reads the catalog, but it ignores typos, your business synonyms, and the order your products should appear in.
It's in stock, though. It's a lost sale that shows up in none of your analytics. You never see it.
Fixing this doesn't require a new platform, or months of engineering. It takes an engine that reads your references the way your customers think of them: Heurix plugs into your existing site by API, and understands M8x20 from the very first call.
The search bar is the one place where a customer tells you exactly what they want to buy. Most sites answer with "no results." Heurix exists to turn that moment into a sale.
Eleven industries already have a ready-to-use rule pack. You pick yours, your references are understood immediately, with no generic engine to teach from scratch.
Search platforms are built for the enterprise. Heurix is built for high-volume catalogs that have neither the budget nor the need for a dedicated data team.
Every improvement translates into a quantifiable conversion gain, not a quarter of development.
Heurix isn't a layer on top of Algolia or Elasticsearch: it's our own engine, built in France, specialized for technical catalogs. Where generalist engines compare characters, Heurix reads the structure of your references.
The demo at the top of this page runs on this exact engine: the numbers alongside it are measured, not promised.
Heurix recognizes the components of a reference — family, dimension, material, standard — and turns them into usable attributes.
A second layer assembles these attributes into a searchable combination, which the customer can ask for however they phrase it, whatever your sector.
You never write a single regex yourself. Our rule packs — hardware, fashion, industry, automotive, electrical, plumbing, books, electronics, wine & spirits, finance & accounting — are pre-configured and activate when you pick your industry, at indexing time. A term specific to your catalog that we don't cover yet? Add your own Custom Rule from the console, simply describing what it should recognize.
// The cascade, two levels, whatever the sector
"m8" → DIAM_M8 "jeans" → FAM_PANTALON
"stainless" → MAT_INOX "w32 l34" → TAILLE_32_34
DIAM_M8 + MAT_INOX → DIAM_M8_INOX
FAM_PANTALON + TAILLE_32_34 → JEAN_T32_34
A product reference is never a random string of letters: it's several pieces of information compressed into one string, whatever your sector. Most engines compare characters and fail as soon as the format changes.
Heurix works like a prism: the reference goes in as a single beam, comes out decomposed into searchable attributes, and your customer finds it by any fragment, in any order, with or without a typo.
These two diagrams run on the real Heurix engine, not a mockup: the same mechanic, applied to two different sectors. Explore the eleven rule packs or build your own.
The larger a catalog or content base, the bigger the gain from well-tuned search, because native search fails exactly where precision matters most. Eleven industries already have a dedicated rule pack, ready to use; the others are covered with Custom Rules in a few minutes, no regex required.
Compound sizes (W32/L34), colorways, materials, variants, and customers who type fast, with typos.
Tens of thousands of references, standardized product codes (DIN, ISO), the need to filter by dimension, material, compatibility.
Search by manufacturer reference or equivalence, tolerance for multiple code formats depending on the supplier.
ISBN, editions, formats, collections: the same title exists in dozens of versions: the demo on this page proves it.
Precise technical specs (voltage, amperage, standards), catalogs that evolve continuously.
High volumes, professional buyers who search by exact reference more than by descriptive keyword.
Vintages, estates, appellations, formats (375ml, magnum): references where every digit matters.
Articles, FAQs, knowledge bases: the same typo tolerance that finds a product reference finds a poorly-worded question. (Try Ctrl+K on this page.)
If your customers search by reference, dimension, or variant, the engine adapts. Tell us about your catalog.
Full search round trip, network included, measured on 11 September 2026 on the public demo (2,821 and 1,500 products): 120 calls, median 46 ms, from a machine 11 ms from the server. The engine itself accounts for about 30 ms of that at the median, the rest is network. Further from the server, expect more.
"pul rond red neck" still finds red crewneck pullovers. Just one on references: change a single digit, and it's a different product.
14-day free trial on the Search API: up to 2,000 requests and 2,000 products, no credit card. See pricing.