We build custom, full-stack manufacturing software — quoting, pricing, order entry, shop floor, and the books — using AI agents directed by engineers with 30 years of ERP experience and real time on the machines. Owned by you. Priced like a project, not a subscription.
Generic enterprise platforms charge you every month, forever, to run a business they've never seen. The workflows don't fit, so you pay integrators to bend them. The data model doesn't fit, so your people live in spreadsheets on the side. And the moment something breaks, the person who understood it is three vendors ago.
Meanwhile the software that actually runs many shops is a desktop app from another era, wired to a database nobody dares touch — one bad day away from stopping production.
It didn't used to be possible to just replace all of this. Custom systems took years and seven figures. AI changed the math: a small senior team directing AI agents can now rebuild the entire pipeline — faithfully, verifiably — in months.
“The first thing he wants to do is rip out Salesforce, Oracle and SAP, for which they're paying a combined $1M per month.” — A YC partner, describing a founder buying factories. We build for that founder.
Not a platform. Not modules. A single system shaped exactly like your factory — every screen, rule, and report built for how your shop actually works.
Order entry that speaks your product's language — options, sizes, materials, customer-specific catalogs.
Your real pricing rules — every upcharge, every exception — reconstructed from wherever they live today, then made testable.
Station-by-station MES tracking with hardware integration on the floor — scanners, stations, machines — plus cut lists, schedules, even STL files, generated straight from orders. No re-keying.
Confirmations, approvals, and status emails sent from the system — branded, logged, automatic.
Clean handoffs to QuickBooks and your accountant. Invoices that match the orders, always.
Every write logged, every price traceable to a rule, every change reviewable. Built in from day one.
Modern stack, zero lock-in. Web-based, works on the shop computer and the phone in the truck. You own the code, the data, and the deployment. If we disappear tomorrow, any competent engineer can pick it up.
A decades-old make-to-order manufacturer was running its entire business on a legacy desktop application and a Microsoft Access database — and it was failing in production.
The business ran on a SQL Server from 2012 — past end-of-life — with a Microsoft Access front-end installed machine-by-machine on each workstation. A previous rewrite attempt had stalled at “90 percent” and never shipped. The original developer was gone, there was zero documentation, and the first database backup we found was corrupt: every byte zero. When the system started failing, there was no vendor to call.
Before writing a line of the new system, we recovered the truth. AI agents restored a years-old database backup in Docker, decompiled the stalled rewrite for its logic, and reconstructed the complete pricing model as an explicit, versioned rules engine — then replayed 268,861 historical product builds through the new engine and diffed the results against the numbers the shop floor actually cut. That replay caught a real formula discrepancy the owner's own corrected worksheet had wrong — resolved by the evidence of 400,000+ real records.
In parallel, agents built a modern web application: keyboard-fast order entry shaped like their product, the reconstructed pricing engine — editable by the shop, not hardcoded — a real-time 3D product designer, production boards with barcode tracking across 11 departments, customer analytics, QuickBooks integration, passkey login, and an audit trail on every single write. Roughly 64,000 lines of code, 51 database tables, and 95 automated test files — where the legacy system had none.
A live, client-accessible deployment existed under 8 weeks from kickoff — reviewed against dated before/after evidence at every step, with 43,679 historical orders and ~2,600 customers migrated in. The legacy system stays up until the new one has earned cutover, because the first rule of an ERP rescue is that the factory never stops shipping.
AI agents write most of the code — that's why this takes months instead of years. But agents are directed, reviewed, and verified by engineers who have shipped ERP for three decades. Here's the loop:
We take your legacy stack apart — old databases, backups, exports, even the application binaries — and recover the rules your business actually runs on. Nothing is assumed; everything is evidence.
The recovered logic becomes an explicit rules engine with tests.
Where the legacy code and the accepted business numbers disagree,
the numbers your customers actually paid win.
Agents build the new system in vertical slices — order entry first, then pricing, then production — each slice deployed to a live URL your team can click the same week.
Dated before/after screenshots, automated test suites, audit logs on every write, and side-by-side reconciliation against the legacy system. You never take our word for it.
The legacy system stays up until the new one has earned trust. After cutover, the same agent workforce keeps evolving the system in fixed-price sprints — no per-seat licenses, no forever-fees.
Every sprint ends deployed. You judge progress by clicking the app — not by reading a status deck.
Every visible change is captured as dated screenshots you can scroll back through — a running record of what changed, when.
Tickets, work logs, and sprint status in one place. What's in flight, what shipped, and what it cost — visible at all times.
Every write in the delivered system is logged and reviewable — enforced by automated tests, not by promises.
New pricing and production engines are replayed against your real historical records before cutover. Discrepancies get surfaced, not smoothed over.
MES station tracking shows every order's real position on the floor. The transparency you get from us during the build is the same transparency your team gets from the system every day.
You commit to outcomes, not open-ended hours — and you can stop after any step with everything you've paid for in hand.
Tell us what you're running. We'll tell you straight whether a rebuild makes sense — and what we'd tackle first.
A bounded working session across your systems and data, ending in an actionable written plan — built to be a great plan of action no matter who executes it. Useful even if you never hire us.
Fixed sprint cycles on a two-week cadence — each one ends in deployed software you can click, not a status report. You judge us by what ships.
Rivlet is led by Daniel Gray — thirty years building ERP systems for real businesses, with time spent working in machine shops, not just writing software about them. Before founding Rivlet, his last full-time role was automating chip-design pipelines — where "the software must be exactly right" isn't a slogan, it's tape-out.
That combination — ERP depth, shop-floor literacy, and modern AI engineering — is the whole point. Agents make the work fast. Experience makes it correct.
Order management, pricing, inventory, scheduling, accounting integrations — designed, built, and lived with in production.
We know what a traveler is, why the router table lies, and what happens to software that ignores how parts actually move.
Dan's most recent full-time role before founding Rivlet: automating semiconductor design flows — the most unforgiving pipeline discipline there is.
A directed workforce of AI agents doing the volume work, with senior engineers accountable for every line that ships.
Tell us what you're running — the ERP you're renting, the desktop app you're afraid of, the spreadsheet holding it together — and we'll come back with what a replacement looks like.
Prefer direct? db@rivlet.io · 1 (510) 900-0374