01
Discover exact pages
Identify brand, collection, series, and product URLs while keeping the original source hierarchy and HTML base resolution intact.
Product imports
Bring the catalog you already have.
ConfigAI treats every import as a source-backed product onboarding job. AI is used where interpretation is valuable; deterministic paths stay deterministic when a supplier already exposes structured product data.
Website
Agent research
Evidence extraction
pCon Catalog
Deterministic
pCon / OFML
Supplier source
Conversation
Agent-assisted
Manual
Direct entry
Evidence layer
identity · text · images · relations · source revisions
Product Setup proposal
18 products
1 product type
7 product fields
4 open questions
source identity matches
Six ways in
Different sources. One product contract.
Every source ends in the same Preview → Review → Apply flow, so adding a website importer or pCon connection does not create a parallel product database.
Agent-assisted
Website
Research public product pages, preserve exact links and captured evidence, and let the agent propose product structure.
Agent-assisted
PDF catalog
Extract text and page evidence, then use the same product-setup workflow rather than flattening the PDF into names only.
Deterministic
pCon Catalog
Read the provider's model list and source hierarchy directly. No AI prerequisite for the deterministic catalog path.
Supplier connection
pCon / OFML
Connect authorised supplier data and preserve source product identities, family relations, configuration context, and assets.
Agent-assisted
Conversation
Let a product owner describe the range, then convert the conversation into explicit fields, products, and open questions.
Direct
Manual
Create or correct products through the same Product Type fields and revision model used by every other import.
Website research
Research should make imports faster, not turn the agent into a browser babysitter.
The website path is designed around exact source capture: start from authorised links, collect product text, JSON-LD and images, preserve low-confidence findings, and reuse captured evidence instead of re-fetching the same pages.
02
Capture evidence
Store the text, structured metadata, images, and source revision the proposal depends on. JEV can accelerate bounded research and replay captured results.
03Evidence first
Propose, do not invent
Map only confident facts to products. Collection-level information remains collection-level unless the source proves a product-level claim.
pCon Catalog
Structured supplier data does not need an AI detour.
The pCon Catalog path is intentionally deterministic: provider product identities and available source relations are imported directly, and available CAD can be converted and attached without making Gatekeeper or an AI agent a prerequisite.
Public catalog
Model list
Provider product identity
Article number
Images
Available family / series relation
Available CAD → GLB
Private / supplier data
When a supplier requires login or Gatekeeper-authorised access, credentials remain part of the connection boundary rather than being copied into product records.
Source authority stays separate from product authority.
ConfigAI records where a product came from; the reviewed ConfigAI revision remains the application-facing product definition.
Import protocol
Collect → normalize → ask → preview → review → apply.
The important part is not the crawler. It is making every import repeatable, idempotent, reviewable, and bound to exact product and source revisions.
01CollectCapture source identity and evidence without creating products yet.
02NormalizeResolve products, families, types, field candidates, images, and relationships into a proposal.
03AskOpen targeted questions only where ambiguity changes the product definition.
04PreviewShow create / reuse / update decisions and semantic field impacts before mutation.
05ReviewLet the human accept, correct, or reject the proposal.
06ApplyWrite durable product/profile revisions with idempotent mutation identities.
ConfigAI workspace
Import the source. Keep the evidence.
The goal is not to scrape a catalog once. It is to create a durable, reviewable product system that can be reopened and updated when the source changes.