What it costs.

From the same table the till uses.

The monthly amount per plan sits in one place in the code, and that is the same place the amount comes from that is handed to our payment provider when a subscription is created. This page copies nothing. Were a difference ever to arise between what is stated here and what is charged, that is a broken test and not an unhappy customer.

The limits come about the same way: how many projects, how many paid seats, how much AI usage per month, and whether signing in through your own identity provider is included. They are read from the same table the API uses to refuse.

Enterprise has no button. It comes with a contract, a data processing agreement, arrangements about hosting and support, and usually an invoice at thirty days. That does not run through a direct debit mandate and certainly not through self-service — so it says “let’s talk”, and that is exactly what it is.

Included up to an amount, and there it stops.

AI work is not a fixed cost. One ticket from analysis to pull request easily costs a few tens of eurocents in model usage, and a team that goes through that a hundred times a month is in a completely different order of magnitude than a team that does it five times. One fixed amount for both cannot be fair: it is too expensive for the one or loss-making on the other.

That is why every plan names an amount of AI usage that is in the price. Reach it and the work stops with a sentence saying so — it does not roll on into an invoice you did not see coming. Nothing is ever charged on top of your subscription. Not as leniency, but because a bill you did not agree to in advance is not a bill you should get.

Usage is visible per run, per ticket and per project, so you can see where it goes — and an agency can pass it on to its own client one to one. If you need more than your allowance, there are two ways on, and both of them really exist: a larger plan, or your own key for the code work. That second one does not count towards the allowance at all.

Bring your own contract. It costs you nothing, and it saves you nothing.

For code work you can supply your own key or your own endpoint — OpenAI, Azure OpenAI, or a model running inside your own walls. That work then runs on your contract: we see not a cent of it, and it counts neither towards the allowance nor towards the ceiling above. If you run into your ceiling, your own key is the way around it.

What it does not do is lower the subscription price. That is deliberate: you pay us for the board, the gates and the agents walking over them — not for reselling tokens. A discount there would mean charging the most to the customers who cost us the least, and that is not a pricing model but a penalty on self-reliance.

Process work — parsing an analysis, classifying comments, deriving test cases — runs on our own European model and is simply part of the plan. Asking for a key for that would cost more administration than the amount it is about.

You pay for who decides, not for who watches.

Viewers and customers do not cost a seat. That is not a commercial gesture but a condition for the product to work: feedback that runs aground on a licence moves to e-mail — and then the truth sits outside the board again.

AI colleagues are not accounts and therefore never cost a seat. What they consume is recorded per run on the work item — which model, how many tokens, what it cost and how long it took. So a team of five people with ten AI colleagues pays for five seats. That usage is gathered per project and per ticket, so an agency can pass it on to its own customer one to one.

The refusal is there, word for word.

The last row of the table above carries, per plan, the sentence you get as soon as that plan is full. That is not a paraphrase: it is the same function the API calls to stop you. If the table ever falls out of step with the limit, a test breaks — and not a customer.

Limits also stop you with a reason attached and not with an error code. That holds everywhere in this product: an AI colleague that runs into its cost ceiling stops and asks a person to continue; a model outside the EU is refused with the note that your organisation only allows EU hosting.

What is not here yet.

The limits on projects and seats are enforced: anyone on the free plan who tries to create a second project gets the answer from the last row of the table above — word for word, from the same function.

The AI allowance is measured and enforced: this month’s usage is added up before every model call, work on your own key does not count towards it, and at the allowance the agent stops with the same sentence the table above lists for your plan.

Invoices are created for every payment, with the VAT treatment that applies to you — reverse charged if you are a business in another EU member state. What is not there yet: we do not send them to you, so you fetch them from your subscription screen, and there is no PDF. And we check the format of a VAT number but do not yet verify it against the European VIES register.

If you want a limit that is not listed above, or an arrangement about hosting and support, that is a conversation and not a button. It says so because that is how it is.