The pricing page lists seven plans in two groups: five billed per account and, below a divider, two counted on a different basis — Team, charged for every person on it, and Enterprise, agreed by contract. There is no currency selector here. The currency follows the language the page was opened in, and a "currency" parameter in the address takes precedence over that; the Polish zloty, the euro and the US dollar are supported. Team begins at three people, and that number is set inside the payment form, anywhere between three and one hundred. Two cards carry no purchase button: the free plan leads to creating an account, or to a new project for someone already signed in, while Enterprise opens a message to the contact address. Where no price has been configured with the payment provider for the chosen plan, currency and period, the purchase button gives way to that same contact link.
Two independent counters, both written on the server, record consumption: the number of AI runs and the number of tokens. Both limits are stored on the plan and are checked separately before a request reaches a model. A run is not only a message in a conversation. The same entry is written when an image is generated, when several models are compared on one question, when a second opinion is requested, when the facts in an answer are checked and when individual claims are matched to sources — and equally when any of that is done from the phone application. Tokens are not taken from the model provider; they are estimated from the length of the instruction and the reply, which makes the figure an approximation rather than an invoice. Work resolved without a model, such as reading a fact straight out of an attached file, records no tokens at all. The usage panel inside the account shows where both counters stand and the date they start again.
Plans differ in more than the size of the quota; they also differ in access to the deeper working profiles. A single field on the plan record opens all of them: extended analysis, the research, coding and agent profiles, the profile with the widest context, and the mode in which several model roles work on one result. That field has two distinct consequences. Under automatic selection the DI Core engine steps down to the standard profile and the work continues; when a profile outside the plan is named explicitly, the request is refused and the message states which plan covers it. Pro, Power, Ultimate and Team carry the entitlement. Supporting actions — translating an answer, restating it in simpler terms, pointing out its weak spots, scoring confidence, follow-up questions, web search, the automatic title, the summary and tags of a conversation, and writing to memory — need a plan whose limit is not yet spent and each has its own frequency cap, but none of them moves the run counter. Nothing in this pricing list offers a public programming interface, because DI Cloud publishes none; the mention of your own keys under Enterprise concerns a customer's keys to model providers, within a scope agreed individually.
The payment form collects a billing address and a tax identifier, and it has a field for a promotional code. The counter window is the calendar month whatever day the purchase falls on, so counting starts again on the first of the month rather than on the anniversary of the payment. Past four fifths of the quota the panel marks the account with a warning, one message goes to the account address within that period and a notification appears in the application; once the limit is spent, the refusal says whether it was runs or tokens that ran out. A failed instalment does not cut access the same day: seven days run from the moment the payment provider reports the arrears. When a paid period ends the account is not locked, it is measured against free-plan limits instead. A subscription bought in the iPhone or Android application is verified with the store and recorded separately from card payments; the stronger of the two plans then applies, so purchases made through two channels do not add up.