cloudrop.it

Manifesto

Not a dashboard. An instrument.

A dashboard reports what already happened. An instrument tells you what to do about it, and then, if you let it, does it. Cloudrop is the second thing.

Autonomy is worthless without a brake.
We shipped the brake first.
01 Intake

Everything arrives as noise.

Eight point four million events a day. Order placed, order paid, order disputed. Stock moved, stock lost, stock counted twice. Price changed by a competitor at 04:12 on a Sunday. A courier scan in a depot nobody has heard of. A buyer asking whether the cable is two metres or three.

None of it is information yet. It is raw signal: continuous, arriving whether or not anyone is watching. The instinct is to hire people to watch it. That instinct scales linearly and the input volume does not.

Nobody reads 8.4 million events. Nobody should have to. The only honest question is what the operator sees instead.

02 Compression

Signal is a structural discipline.

The industry answer is twelve subscriptions and an automation bill. A listing tool that does not know what the warehouse holds. A warehouse tool that does not know what the accountant booked. Between every pair of them, a webhook nobody owns and a spreadsheet somebody maintains at 23:00.

We refused the integration tax. Thirty-two modules share one PostgreSQL database and one process boundary. Not because monoliths are fashionable, but because a decision that spans pricing, stock and cash cannot be made across four vendors’ eventual consistency. It has to be made inside one transaction, or it is not a decision. It is a hope.

Every integration you add is a promise that two companies will keep agreeing about the meaning of a field.

Compression is a structural change, not a summary. The output is not a smaller pile of events. It is a different artifact: an offer that should move, a shipment that will be late, a margin that stopped being one. Eighteen lanes resolve into one state, and the state is what the operator actually reads.

03 Decision

The drop is the decision.

Her name is Adaline. She reads the compressed state, forms a view, and proposes. Reprice this offer. Bundle these two. Pause that listing before it eats the month's margin. She writes down why, in a log the operator can read in ten seconds and reject in one.

She does not have a budget she can talk her way into. Spend ceilings are compiled into the core, not configured in a settings page. Moving one is a code change, a review and a deploy, which is exactly the amount of friction a spending limit deserves. The fence binds to a named set of offers, so nothing outside that set can be armed. Cross-lane moves are surfacing-only. The autonomy ramp exists, is written, is tested, and is dormant behind flags nobody has flipped.

Every model call she makes leaves through one gateway, with a per-feature budget attached at the door. There is no second path, which is why the number in the deck is the number on the invoice.

This is not caution as marketing. It is the shape of the product. An instrument that cannot be overridden is not an instrument. It is a passenger seat.

The fence

Hard spend caps live in the binary. Human veto on every proposal. Autonomy ramp dormant behind flags. Three fences, none of which the AI can move.
04 Constraints

What we will not trade away.

A manifesto that only lists ambitions is a brochure. Here is the other half: the decisions that are already load-bearing and that we do not intend to revisit.

  • Customer data stays on a fabric we own. Four regions of private compute, encrypted replication to three geographic vaults, and no analytics vendor anywhere in the request path.
  • This website watches nobody. No pixels, no session recording, no consent banner, because there is nothing to consent to.
  • Autonomous spend stops at a compiled ceiling. If a limit can be raised from a settings page, it is not a limit. It is a suggestion.
  • Every migration ships with a rehearsed way back. Forward-only schema, restores drilled on a schedule, and a written path back from every flip.
  • Every capability is a flag before it is a habit. One commit switches any of it off, including the parts we are proud of.
05 Return

A small crew, running commerce at planetary scale.

The company is a small crew. That constraint is not a hardship story; it is the specification. Everything here had to be built by people who would also be woken up by it. Software written under that condition acquires a particular quality: nothing survives that is merely impressive.

Leverage is the whole thesis. Eighteen lanes, thirty-two modules and four regions do not need a floor of headcount. They need one state, one chain of command, and a machine that proposes without arming itself.

Cloudrop is the instrument we could not buy, running the business that pays for it. The landing you are reading sits above the noise. The deck sits inside it.

A small crew. This is what it runs.

Marketplace lanes
18

Across Europe and Asia.

Modules
32

One deck. One database.

Regions
4

Warsaw · Frankfurt · Singapore · Tokyo.

Availability
99.99%

Measured, not promised.