Things we built that run every day.

The problem: enterprise warehouse software costs a fortune, takes a year to install, and still makes you call sales to find out the price. Small and mid-size shippers were stuck between a spreadsheet and a mortgage.

What we built: a cloud warehouse management system for brands that ship their own orders and for 3PLs that ship everyone else's. Real-time inventory across multiple warehouses, barcode picking and packing, batch pick tours, purchase orders, demand forecasting on 30-day velocity, and carrier rate shopping across UPS, FedEx, USPS and DHL. Orders flow in from Shopify, WooCommerce, Amazon, eBay, BigCommerce and Miva. Books flow out to NetSuite, SAP, Sage, QuickBooks and Zoho. Full REST API with webhooks for everything else.

What happened: one customer, a linen retailer, went from 100 warehouse staff to 35 with the same output, and kept about $180,000 a year. Bigger clients push 1,500 orders a day through it. First million packages in 2020. Past 31 million now.

Our favorite feature: the price is on the website. In warehouse software, that's practically a scandal.

The problem: manufacturers who sell through distributors and dealers run the whole channel on email, PDFs and a pricing spreadsheet only one person understands. Every quote is a negotiation. Every order is a re-type. Nobody knows which dealer said what.

What we built: one ledger for the whole network. Catalog with dealer-scoped assortments. A pricebook with dealer tiers, volume breaks and promo overrides, computed at quote time and locked the moment a quote becomes an order. Orders become invoices. Communication is threaded by subject, so replies land on the right conversation without anyone filing them. Add 3D product tours with a hotspot editor that runs in the browser, and a public service-center locator you can embed on any product page with a single line.

How it's built: multi-tenant, one isolated workspace per distributor, six roles from root to public. It runs beside a Shopify storefront rather than replacing it, and shares the same SKUs, so the product page, the dealer quote and the service locator all agree.

The problem: "AI app builders" hand you a demo that lives on their servers, in their framework, under their subscription. The moment you stop paying, your app stops existing. Cute demo. Terrible landlord.

What we built: a platform where a planner, a set of builders and a reviewer (all AI, on different models so they can't share the same blind spots) produce a real, conventional PHP application with a database, auth and an admin panel. Each project gets its own OS user, its own process and its own database. Push it to your GitHub, host it anywhere, walk away anytime.

What's underneath: a durable workflow engine we wrote from scratch, with lease-locked stateful objects so a crashed worker can't deadlock a job. Thirteen step types. Human approval gates as a first-class primitive: any step can pause, wait for a person, and resume exactly where it stopped. Models are routed by cost, so the frontier model plans and cheaper models execute, escalating a tier only on failure. A headless test agent drives the finished app with Playwright at every permission level before anyone calls it done.

Stripe billing. Subscriptions, metered usage, per-warehouse and per-user pricing, multi-client invoicing for 3PLs who bill their own customers, failed-payment retries, and invoices that reconcile to the accounting system. We built it for ShipCannon and Tiknix, and we bill on it ourselves. When your own rent depends on the billing code, you write it carefully.

Shopify. Custom apps and private integrations for merchants who need the store talking to something else. OAuth, catalog browsing and sync, order intake, fulfillment and tracking pushed back to the store, hand-offs to the ERP and the warehouse. CannonWMS ships with a native Shopify adapter we wrote, and DealerYes shares SKU data with the storefront it sits beside.

Everything else. Salesforce and Marketing Cloud unified with SAP/HANA across four countries for a global tools brand, so one customer record existed instead of four. Real-time middleware between an ecommerce platform, NetSuite and a Deposco warehouse for a luxury resale company, after two prior attempts had stalled. FedEx and Ryder APIs behind a "where is my order" tool that cut order-status tickets by a quarter. Boomi, when a client already owns it.

FastMCPHP. A framework for writing Model Context Protocol servers in PHP. MCP is how an AI model reaches your tools and data safely. A big chunk of the web still runs on PHP, so we made the two shake hands.

MariaDB-to-LLM bridge. Ask your database a question in plain English. The question becomes SQL on your server, runs on your server, and the answer comes back. No data leaves the building. Built for clients whose lawyers say "absolutely not" to cloud AI, which is a lot of lawyers.

MyCTOBot tooling. The open pieces behind an on-demand architecture and strategy service for early-stage companies who can't yet afford a CTO. Answers the questions a founder is embarrassed to ask out loud.

Mid-size makers who want to sell direct online without torching their dealer relationships. Shopify for the storefront, DealerYes for the channel, CannonWMS for the warehouse, and AI keeping an eye on demand and quality. Wired together by the people who wrote two of the four pieces, which shortens the finger-pointing considerably.