Engineering: warehouse and retail
Cubitrek WMS
Warehouse software you can talk to. Ask for the stock report instead of building it
The problem
Operations software is where people go to do data entry. Every question a manager has, such as what is sitting in the warehouse that has not moved in sixty days, becomes either a saved report someone built once or a request to a developer.
What we built
- An MCP integration, so a person updates records and pulls reports by asking Claude instead of doing it all by hand.
- Point of sale, inventory, warehouse and light accounting in one application, extended in house across 333 tracked schema migrations. Our platform, built on a licensed commercial core we have extended.
- Per-unit serial and IMEI tracking, so a device is followed as an individual item through receiving, grading, sale and return rather than as an anonymous unit of stock.
- A stock-ledger invariant harness that gates deploys. Before a release ships, the suite proves the ledger still balances against physical movements. Inventory bugs are silent by nature, so we made them loud and early.
- Integrations built and shipped: third-party logistics, accounting, carrier and shipping, storefront order and inventory sync, and a grading and inventory partner. Storefront exports run through a queue rather than a synchronous call that can wedge the store.
- Multi-currency with document-currency columns alongside base currency and a rate log, so historical documents do not silently re-price when a rate moves.
- An end-to-end test suite that starts its own server and seeds its own login, so the tests do not depend on a shared environment being healthy.
- Resilience against someone else's outage. Every background workflow is a command-line job, and each is also triggerable over authenticated HTTP. That was built because the host's own scheduler proved unreliable, stalling for hours at a time. The system no longer depends on the weakest component in the stack.
Where the automation sits, and where it does not
Everything that runs on a schedule runs as a command-line job: accounting, storefront sync, currency, exports, carrier alerts. None of it depends on a person remembering. The reason that matters is the last point above: when the host's scheduler stopped firing, the work carried on because something else could trigger it. Automation is only worth having if it still runs on the day the platform underneath it does not.
What we would not claim
The base application is a commercially licensed platform. The integrations, the migration history and the resilience work are ours; the original core is not. We publish no customer names or volumes for this product.
Platform status, September 2026.
Your turn
We set the baseline before the work starts.
That is the part that makes a number mean something a year later.