Your AI token allowance

Every question an agent answers costs tokens — roughly, pieces of text going to the model and coming back. Your plan includes a number of them per billing period, and Usage shows how many you have used and how many are left.

Why tokens and not just messages

Both are counted, because they constrain different things. A message limit stops a workspace making thousands of small calls. A token limit stops one making a handful of enormous ones, which is where the cost actually is: an agent reading a large table can spend more in a single answer than a hundred short questions.

What you get

• •Free — 250,000 tokens per period
• •Starter — 5 million
• •Pro — 50 million
• •Business — 250 million

As a rough feel, an ordinary question and answer is a few thousand tokens. An agent working through several tool calls to answer something complicated is more, because the conversation so far is sent again with each step.

When it runs out

Everything that needs a model stops. Agents will not run, chat will not answer, and reports that ask a question will not produce one. The people who own or administer the workspace get one email when it happens — one, not one per blocked request.

Nothing is deleted and nothing else stops. Your agents, connections, history, change log and settings are exactly as you left them. Reports built from saved data still open. It is only the model that is unavailable.

Two ways forward. The allowance resets at the end of the billing period and everything starts working again on its own. Or move to a larger plan, which starts working immediately — see [[changing-plan]] for how the difference is prorated.

Seeing it coming

Usage shows the allowance as a bar, and once 80% is gone the chat screen says so before you type. If you are regularly close to the ceiling, the honest answer is usually a larger plan rather than shorter questions.