Five enterprise modules.
Coupa parity, closed.
We mapped the gaps between this kernel and Coupa, then closed five of them end to end — schema, agents, APIs, and UI — each one walked through live on real data.
Migrations 648–652 · 6 pure libraries · ~76 new tests · 21 live modules on one kernel
Guided Buying
One smart front door for any purchasing need
Entry points were fragmented — a requisition form, a Telegram bot, a webshop. Employees had to know which process to use before they could ask.
Live: “5 boxes of A4 paper and two ergonomic chairs” → classified Catalog requisition at 90% confidence, matched to the Office Depot catalog, lines pre-filled and charged to Finance.
Spend Analysis
A real UNSPSC spend cube, not a flat KPI page
The spend page showed totals and a trend line. There was no commodity taxonomy, no sliceable cube, and no way to see leakage.
Live: €7,388 of committed spend, 100% classified, €3,390 (45.9%) flagged maverick — spend that skipped a preferred or contracted supplier.
Services Procurement
Statements of Work with service entry sheets
Goods ran PO → goods receipt → invoice. Services spend — often half the total — had only timesheets, no SOW lifecycle or delivery proof.
Live: a €20,000 SOW at €5,000 delivered (25%). Approving a service entry sheet moved billable value to €17,000 automatically via the value-rollup trigger.
cXML PunchOut
Shop supplier-hosted catalogs over standards cXML
There was a mock REST connector and a SAP IDoc generator, but no standards-based PunchOut — the catalog integration every enterprise buyer expects.
Live: a full roundtrip through the built-in mock supplier — punched out, built a cart, and it returned as a real cXML PunchOutOrderMessage (UNSPSC codes intact, €1,405) onto the review screen.
Compliant e-Invoicing
Suppliers submit — PO-flip or UBL/PEPPOL
Scan-to-Pay was the buyer re-keying a PDF. Suppliers had no way to submit an invoice themselves, and nothing validated structured e-invoices.
Live: the demo vendor flipped a PO into invoice INV-DEMO-2026-001 — tagged PO-FLIP, compliance surfaced as a warning, and the 3-way matcher picked it up (status MATCHING).
Same kernel. New surface area.
Every module above reuses the existing approval engine, treasury loop, and agent fleet — no new infrastructure. That is the whole point of a kernel.