The Problem
Industrial catalogs don't agree on a vocabulary. AI agents can't parse around that.
A person browsing a manufacturer's site can figure out that "Ie" and "operational current at AC-3" mean the same thing. An agent calling a tool can't — it needs one schema it can write a condition against, like rated_current_ac3_a >= 40. Every catalog is its own inconsistent, often JavaScript-rendered site, with its own vocabulary, its own units, its own way of hiding a discontinued part. That inconsistency is a minor annoyance for a human. For an AI agent sourcing a part number, it's the difference between a real answer and a hallucinated one.
The Method
How the translation actually happens
No per-product LLM guessing. The vocabulary translation is designed once per source and reviewed by a person — then applied the same way every time.
Collect from the source
Data is collected directly from each manufacturer's own site or catalog. The raw value is kept exactly as published — nothing is normalized away at this stage.
Translate by reviewed rule, not guesswork
A mapping dictionary — drafted with LLM assistance, reviewed line-by-line by a person — translates each source's labels into canonical keys. It's applied deterministically, so results are reproducible and auditable, not re-guessed per product.
Serve with proof attached
Every fact carries its source key, a confidence score, a source type and a verification status; each product links back to the manufacturer, labeled with what kind of page it is. Versions are kept, so a spec change or discontinuation is a verifiable diff, not a guess.
Three Audiences
Built for three audiences
Same underlying data, three different ways in.
For Developers & AI Teams
Give your procurement copilot, BOM tool, or engineering assistant a source of truth it can cite — cross-brand equivalents, structured specs, and datasheets, queryable through MCP.
See the tools →For Distributors
Cross-reference a real substitute in seconds when a line is discontinued or out of stock, fill catalog gaps, and get ahead of spec changes before a customer finds them first.
See what's available →For Manufacturers
See exactly how your parts get cross-referenced against competitors, make sure your own specs are the verified source, and get ahead of EOL communication.
See what's available →Trust
Trust is the product, not a footnote.
Every number on this page has a definition, and every fact in the database carries two labels: where it came from, and whether it has been checked.
- 94% of products were collected from a manufacturer source — its own website, catalog or API. The other 6% came through distributor catalogs and are labeled that way.
- 74% of facts come from the manufacturer itself — stated in its catalog or API, or parsed from its own descriptions. About 16% are derived from the model number or the product family, and 10% come from distributors or an undetermined source — each marked as such.
- 75% of facts are verified — confirmed against the manufacturer's page or document, or agreed by a second source. The rest are labeled unverified: not yet confirmed, not known to be wrong.