Agentic product configuration
The product layer for your catalogs, configurators, and agents.
ConfigAI turns existing sources into structured product definitions with evidence, rules, 3D, and human review attached.
Source evidence stays attached from draft to published revision.
Source-backed
Every claim keeps a path to its catalog, file, or evidence.
Human reviewed
Agents prepare changes; people choose what becomes product truth.
Versioned by design
Accepted changes publish as exact, traceable product revisions.
From source to product
Give ConfigAI what you already have.
Website, PDF, pCon, OFML, CAD, or a product brief. ConfigAI captures the evidence first, then builds a reviewable product structure instead of flattening everything into a list.
website
source evidence
catalog.pdf
source evidence
pCon
source evidence
product brief
source evidence
Product Type
Modular seating
Import what you have
Keep the original source identity, hierarchy, text, images, CAD, and evidence.
Structure what it means
Draft Product Types, semantic fields, products, relationships, and identity matches.
Keep unknowns explicit
Low-confidence facts stay in evidence or open questions instead of becoming product truth.
Refine in conversation
Change the model by talking about the product.
Ask what a field means, resolve an ambiguous variant, correct a relationship, or request a new revision. ConfigAI shows the semantic impact instead of silently rewriting the schema.
Product Setup Agent
source-aware conversation
Semantic diff
- arm_variant: string
+ arm_type: option[3]
Identity impact
No product identity changes. Existing source bindings are preserved.
Fields
7 → 7
Products
18 unchanged
Rules
1 binding affected
Prove it before Apply
Test the proposal, not just the code.
Source checks, identity matches, semantic-field impact, rule bindings, and optional 3D paths are validated before a proposal reaches a reviewer.
Quality control
proposal validation
Change summary
Safe to review, not yet applied.
People release. Always.
Agents draft. People decide what becomes product truth.
Product Apply, semantic migrations, 3D attach, and rule publication remain explicit review decisions with exact revision history.
ConfigAI agent
drafts the product revision
Validation
checks source and semantic impact
Product owner
approves the proposal
ConfigAI
applies exact revision
Bring your own agent
And your own model.
Codex, Claude, local models, or any MCP client can work through the same governed product tools. The agent changes; the product authority does not.
ConfigAI MCP
review-aware product tools
Codex
scoped tools
Claude
scoped tools
Local model
scoped tools
Any MCP client
scoped tools
Search & inspect
Find products, exact revisions, allowed options, assets, and source evidence.
Start workflows
Launch bounded imports, product setup, modeling, and rule-test tasks.
Review-aware mutations
Agents submit proposals or approved actions instead of bypassing domain rules.
One product layer
Data, 3D, and rules stay on the same product identity.
Once a product revision is reviewed, the same identity can carry its 3D artifact, configuration rules, saved configurations, MCP context, and later commercial or production layers.
Palmas İkili Kollu Modül (Sol)
pCon source → verified product.usdz
Build from what you already know
Bring the catalog.
Build the product system.
Start from your existing sources and turn them into structured, reviewed product definitions that your people, configurators, 3D workflows, rules, and agents can share.