# What a question costs

> What makes up the price of a question and a research, where to see it, tokens in plain words, the three kinds of work in a plan, and adding people for free.

You pay **for the AI agent’s work**, in roubles, by the amount read and written. The cost of a conversation is shown in its header — the total for all questions and answers in the conversation. Members, spaces and connections cost nothing: **adding people costs 0 ₽**; only the model’s work is paid for.

## Tokens in plain words

The model reads and writes text in pieces — **tokens**; a token is part of a word. Everything the model has read counts as **input**, everything it has written as **output**, and input and output have different prices per token. The organisation’s plan sets these prices; in the model selection window they are shown in roubles per million tokens.

- **Input** — your question, the documents the agent opened for the answer, knowledge base entries, earlier messages in the conversation, attachments.
- **Output** — the answer and the reasoning.

## Where to see it

- **In the header of an “Ask” conversation** — input and output tokens (input counts the cache, output the reasoning) and the cost. The figures update with every message.
- **In the balance card** at the bottom of the sidebar — how much money the organisation has, its plan and the spend over 30 days; the card opens billing.
- **In the billing history** (Settings → Billing) — every operation with its kind of work, plan and amount; the history can be exported to a file. The organisation owner can view it.

## What the price depends on

- **How many documents had to be read.** A question that names the document and section is cheaper than a general one: the agent opens one page instead of going through a folder — see [How to ask a good question](/en/docs/asking-well).
- **The length of the documents.** A page, a spreadsheet or a contract is read in full: nothing is cut off, so that no requirement gets lost. A long document means a long input.
- **Attachments** are read by their content, in full.
- **Knowledge base entries marked “show to the model straight away”** arrive in every question in full; above the list of entries you can see how many tokens they add to each request — [how this reaches the model](/en/docs/knowledge/prompt).
- **Continuing a conversation.** Earlier messages are read again — even so, a follow-up in the conversation is cheaper than a new question from scratch, because the documents have already been found.
- **A research costs more than a question:** two agents re-read the sources several times — [when you need one](/en/docs/ask-vs-research).

## Three kinds of work in a plan

In a [plan](/en/docs/tariffs) work is divided into three kinds — **task development**, **“Ask”** (the assistant in the app counts here too) and **deep research**. On the default plan we pick the models; on the “own choice of model” plan the owner chooses a model separately for each kind and sees its price when choosing. A space without code needs only two of the three kinds.

**The plan applies to the whole organisation, and the owner changes it** — Settings → Billing → “Change plan”. While questions, research or tasks are running, the plan cannot be changed: work started at one price is not finished at another. The new plan applies from the moment of the change, with no retroactive recalculation.

## “Knowledge” passes

Entries that the AI agent makes in the knowledge base during a question are part of that question’s work. What is paid separately is a [knowledge base pass](/en/docs/knowledge/passes) — when the agent re-reads finished work, adds the lessons and tidies up the base: in the billing history this is a separate kind of work, “knowledge”, and the pass card shows what it cost. If you do not need passes, they can be switched off with a platform setting — write to us.

## No figures here

We do not give guidelines like “a question costs this much”: the price depends on the plan’s model and on how much had to be read. Ask your first question — the cost will be in the conversation header. Plan prices are on the [Pricing](/en/pricing) page.