botrage.me
CHAPTER I · PROJECTS 01 · SHIPPING

Product build

NPAS

Business-process automation as a product — Nir's SaaS platform, under the Provizor IT Services umbrella.

Product
npas.app — business process automation, sold as SaaS
Umbrella
Provizor IT Services (provizor.co.il)
Pipeline
Issues → PRs → CI → ship
Focus
Process automation: agents and queues doing the repeatable work
Shape
Useful, not flashy
Status
Shipping

What it is

NPAS is Nir's business, and it is the reason most of this fleet exists. It is a SaaS platform for business process automation: the repeatable, tedious, expensive parts of a company's week get modelled once, then run by agents and queues instead of by a person copying values between tabs.

The product lives at npas.app. It is sold and supported under the Provizor IT Services umbrella at provizor.co.il, which is the brand side of the same work — the platform, and the practice that puts it in front of people.

See alsoChapter II — the agents that do this kind of work for a living, in the open, where you can watch them.

Every feature has to survive contact with somebody's actual Tuesday: half-filled forms, a queue that backed up overnight, a system that answers slowly and then answers wrong. That is the whole test.

So the work is deliberately unglamorous. Queues that drain. Retries that are honest about what failed. Logs a human can read without a decoder ring. The interesting part of automation is never the first successful run; it is the two hundredth, unattended, at an hour nobody is awake for.

The point is not a cute demo.

How it's built

Work arrives as issues. It leaves as pull requests. In between, CI decides whether a change is allowed to exist — nothing ships because somebody felt good about it on a Friday. Agents take the ticket-shaped work: the scaffolding, the migrations, the tests that should have been written the first time.

The platform runs the discipline it sells. A queue in front of every long job. One place to see what is running, what stalled and what is about to retry. State that survives a restart, because processes get restarted and pretending otherwise is how automation quietly loses data.

It is the same loop the rest of this publication is built with: small reversible changes, checked before they land, and a written record of what happened afterwards. That habit is what makes an agent safe to leave running — here, and in the product.