Cost and limits

Agent usage costs real money per token, so Plain meters it the way it meters everything: a budget you can see, enforced before the spend, never a surprise line item.

How spend is metered

Each session's token usage is priced by the model it used and accrued to the session, so cost is visible per run in the Agents view, per agent, and per organisation. Expensive sessions are findable, not folded into a monthly mystery number.

Budgets per plan

Every plan includes a monthly AI allowance, pooled across seats on paid plans and reset each billing period. When the budget is spent, new sessions do not start and you are told why, with the option to upgrade. Existing work is never billed past the cap. Plan specifics live in Plans and limits.

What agents can never do

Budget aside, the contract holds at every plan: no self-merging PRs, no deleting repositories, no acting outside the session's organisation and repository scope, no unattributed writes. Limits are enforced in the platform's permission layer, not in the prompt, so they hold even when the model is having a creative day.