Plans

A plan determines what the AI runs on and what makes up the price. It is not a volume discount: each plan has its own price, and moving between them changes not the size of the bill but the way the computing is obtained.

Current prices are on the Pricing page. This page explains how it works in the app.

General rules

A plan applies to the whole organisation. Balances and charges are kept per organisation, so the plan is shared as well.

The organisation owner can change the plan — in the app: “Settings”, pick the organisation, the “Billing” section, the “Change plan” button.

The plan can be changed when nothing is running. While tasks, “Ask” conversations or research are running, you cannot choose a plan — the switches are disabled, and the window says how much is running at the moment. This is deliberate: work started at one price must not be finished at another.

It does not stop at a counter: the “More” link opens a list of exactly what is in the way — each line names the kind of work and the project, and leads to the work itself. So you can see what you are waiting for, not just that “something is running”.

Knowledge base passes and the model’s entries made during work are paid for on the same plan as the rest of the work: in the billing history they show up as a separate kind of work, “knowledge”, and the pass card shows what it cost. If you do not need them, they can be switched off with a platform setting — write to us.

Buff Code is included in every plan — there is no separate one. Developers’ work in Buff Code on the organisation’s models is charged on the same plan as the platform’s work; in the billing history it is labelled “Buff Code · device name” and can be picked out with the “Source” filter.

A new plan applies immediately, with no retroactive recalculation. What has already been charged stays at the old price: a plan assignment is dated, and each charge is priced on the plan that was in force when that work ran. In the billing history each operation shows its own plan, and the change itself appears in the history as a separate line.

The window shows all offers, not only the ones you can switch to. Those that are connected after a conversation are in the same list, but instead of a choice they have a “Send a request” button — promising a switch where none will happen would be untrue.

A change does not always end with the window closing. The “own choice of model” and “dedicated capacity” plans have a setting without which the plan does not work — the model. So after the switch the window stays open and asks you to choose it right there, while you are thinking about it; the “Done” button closes the window. The other plans close at once: there is nothing to choose.

OpenRouter (balanced models)

The default plan. We pick the models — balanced between quality and price; there is nothing to configure. You pay for tokens.

Code, tasks, questions and the documents read are processed by third-party providers’ models through OpenRouter, outside Russia; we select only providers marked “does not train on requests”.

Suitable if you do not want to deal with choosing a model and it is enough that it “just works”.

OpenRouter (own choice of model)

You choose the model — separately for three kinds of work:

• task development — this also covers the project audit, repository preparation and developers’ work in Buff Code;
• “Ask” — conversations about code and connected sources, as well as the assistant in the app;
• deep research.

The choice is stored for the organisation and applied when work starts. As on the default plan, the work is processed by third-party providers’ models through OpenRouter, outside Russia.

You are offered to set up the models right after switching to the plan — the change window does not close but shows the choice. You can come back to it later with the “Choose models” button on the current plan’s card.

The price of this plan cannot be named in advance — it depends on which model you choose and is shown right as you choose, in roubles. That is why the pricing page shows no figure for it: naming the price of a model nobody has chosen would almost certainly mean naming the wrong one.

What the chosen model costs is shown in two places. In the “Models” window the “Now” line carries not only the name but also the price — input and output, in roubles per million tokens. The same is repeated on the “Current plan” card in billing: under the plan’s name is the active model and its price. If different models are chosen for different kinds of work, the card has a line for each kind.

If the chosen model turns out to be unavailable, the work does not stop — the run goes ahead on the default model.

Dedicated capacity outside Russia

A separate machine for your organisation. The price has two parts: tokens and machine time.

Nothing is sent to third-party model providers: the model runs on this machine.

The hourly price depends on which model you run — it determines which machine is brought up. Model classes and the hourly price for each are on the pricing page.

Everything else about this plan — modes, preparation, shutdown, balance — is on a separate page: Dedicated capacity.

Dedicated capacity in Russia

The same, but in the Russian jurisdiction: nothing is sent to third-party model providers, and the materials do not leave the Russian Federation. Two important differences:

• it is connected on request — not by switching. The request is sent right in the app: the plan is in the same change window, but with a “Send a request” button instead of a choice. The form on the pricing page is the same thing through a different door;
• it is a permanent deployment: its operation cannot be paused from the interface. The environment is kept up for you, so it has no “switch off” button. This is a property of the plan, not an omission.

The cost of machine time and the terms of connection are discussed individually.

Custom setup

A separate environment for your company, with access to the app through a deployed VPN. Also a permanent deployment that cannot be paused from the interface. Nothing is sent to third-party model providers.

This is still our service, which we deploy and operate for you — not a boxed product and not an installation on your servers. The cost of deployment and the terms are discussed according to your requirements.

It is not connected by switching: in the plan change window “Custom setup” stands apart, marked “on request” — and the request is sent from there too, from the app.

Bring Your Own AI

Work on your own model or subscription. We are preparing the release — for now you can send a request for early testing; the plan cannot yet be connected as an available option.

That is why it is not in the list of plans: in the change window it stands alongside, marked “coming soon”, with the same request button.

Requests

Plans that are connected after a conversation, and early testing, are requested the same way — with one and the same form. It has two doors: the Pricing page and the plan change window in the app, where each such offer has a “Send a request” button. A sent request is labelled just that — “Request sent”, not “Connected”.

The form asks for an email, a company and the team size; a comment is optional. We reply by email only — we do not take other contact methods, so we do not ask for them.

A request does not assign a plan. It opens a conversation: we reply, discuss the terms, and only then is the plan assigned in billing. This is so that a submitted form does not change the organisation’s price by itself.