Pricing

Two sites free, then USD 100 per site.

You are charged the first time a site is computed, not when you create it. There is no subscription, no seat count, no annual commitment and nothing to cancel.

Per site

USD 100 / site

Charged once, on the first computation of that site. Recomputing the same site after changing a curve number, a design storm or a practice costs nothing further, however many times you do it.

The first two sites on an account are free and are consumed the same way: on first computation, not on creation.

Free, with no account

USD 0

The stateless quick check: enter segments by hand, get the runoff and the reduction from the same engine. Nothing is stored, so there is no site record, no run record and no audit trail behind it, and that missing record is the part a reviewer relies on.

The reference tables and the defaults registry are published free and are meant to be citable whether or not you ever pay for anything.

What a site is

A site is one boundary, inside a project, with two conditions hanging off it: existing and proposed. Under those sit its segments, whatever green infrastructure practice you model, every run you compute, and the append-only record of every parameter and every departure from a default. That whole object is what the hundred dollars buys, and it stays yours to recompute and re-report.

Two things deliberately do not bill:

  • Creating a site and deriving the public data. You can draw a boundary, pull the soils, the land cover, the slope and the rainfall, read what the survey actually publishes about the ground, discover that a third of it is unrated, and walk away without spending anything. A derivation retrieves published data and computes no runoff, so charging for it would be a surprise on the invoice.
  • Recomputing a site you have already paid for. The charge is a property of the site, not of the run. This matters because the honest workflow involves running a site again after you have answered a question the engine asked, and a product that charged for that would be teaching you not to answer.

What happens when you run out

Nothing silent. The computation is refused before anything is computed or charged, and the message says exactly where you stand:

app/src/cnapp/billing.py, verbatim

This site costs USD 100. Your balance is USD 0.00 and your 2 free sites are used. Add credit to continue. Nothing has been computed or charged.

Credit sits on the account as a balance and is spent one site at a time. Every charge, every use of the free allowance and every credit added is written as its own usage event, so the account page is a ledger rather than a number.

The mechanics of that ledger were written with more care than a pre-launch product usually gives them: a site is claimed before anything is spent, and every write that spends money is conditional on the state it read, so two computations of the same site racing each other cannot take two hundred dollars or burn two free sites. The reasoning and the measurement behind it are in the source.

How you pay today

There is no checkout. Payment is being wired now, and until it is finished this is what happens instead: write to support@curvenumber.com with your account email and how many sites you want, and you will get an invoice. Credit is added to your account by hand once it is paid, and it then shows as a balance on your account page in the tool.

That is a manual process and it is slower than clicking a button. It is described here rather than hidden behind a "contact sales" link because pretending a checkout exists would be the first thing this product told you that was not true.

No refund policy has been written yet. If that matters to you, and on a hundred dollar purchase it reasonably might, ask before you spend anything.

What is not charged for separately

ReportsHTML and JSON
Included with the site, and regenerable from the stored run. The JSON companion is built from the same call as the HTML, so the two cannot disagree on a figure.
The audit trailDepartures and exports
Included. Every override, with who, when, the prior value and the reason, exportable.
Public data retrievalSoils, land cover, terrain, rainfall
Included, and cached, so a second derivation of the same site touches the network not at all.
The defaults registryEvery entry
Published free and meant to be cited, with what each default is worth in the answer and what evidence would displace it.
The proactive warningsengine, not assistant
Included, and they are not a model feature: they are detection over your site's record, they work with the assistant switched off entirely, and no allowance state can suppress, degrade or defer one.

The assistant, where it is switched on

The tool has an optional assistant that answers questions about a site you have computed. It is off by default and this product is complete without it. Everything above is what it costs with the assistant off, which is what you get unless somebody has deliberately turned it on, and nothing on this page changes if you never use it.

Where it is on, it is metered in messages rather than in anything a vendor would recognise as a unit. You will never see a token count in this product.

Messagesper month, by plan
Fifteen a month on the free plan, four hundred on the professional plan, with the other plans in between and above. There is no way to buy a plan today and nothing on this page is a subscription: every account is on the free plan unless whoever runs the deployment has set it otherwise, which is the same manual process the rest of this page describes. A message is one question and its answer. The count is shown in the tool before you spend it, and a question is charged when it is asked rather than when it succeeds, so a question whose answer was blocked by the software's own checks still spends one. That is deliberate: the alternative lets a loop of blocked attempts run for free against our costs, and we would rather apologise for a message you did not get an answer to than not notice.
Deep analysisUSD 100, and it asks first
Running a new scenario through the engine from the assistant is a deep analysis. No plan includes any today, so each one costs one site credit, USD 100, taken from your balance. The tool makes you start it deliberately, tells you the price before it runs, and refuses before anything is analysed or charged if the balance does not cover it. Nothing else the assistant does costs money beyond a message.

The credit is taken when the analysis is delivered, which is the opposite of the rule for messages above, and the difference is the price. If the model vendor fails, or the software's own checks withhold the response, nothing is taken from your balance and the message you get says so. A refused response opens by telling you that nothing in it is a statement about your site, and charging a hundred dollars for that document is not something we could defend, least of all in a product with no refund route to put it right afterwards. It is the same rule as the site charge at the top of this page: billed on delivery, not on the attempt.
What it sendsto a model vendor
A question you ask the assistant sends your site's record and your question to a commercial language model vendor. The privacy page lists exactly what is sent and what is held back, which is worth reading before you use it on a confidential project.