Pricing and budgets
What budgets mean for your bill.
You met budgets first as a safety primitive — a cap that stops an agent before it runs away with anything. The same cap controls your bill. One number, two jobs: it keeps work bounded and spend predictable.
What counts toward a budget
Two things drive almost everything an agent spends.
THINKING
Planning and writing
The reasoning an agent does to plan, decide, and write. Measured in tokens — the more it thinks through, the more it spends.
TOOL CALLS
Reads and writes
Every read or write through an integration — GitHub, Slack, Sentry, and the rest. Each call counts.
Nothing else does. Idle time, observed events that never trigger work, and your own messages are free. An agent watching quietly costs almost nothing. You pay for thinking and doing — never for waiting.
How budgets work
You set caps at three levels, and they nest. Pick the smallest one that fits the work, then let the wider caps catch what you didn't see coming.
PER AGENT
One teammate's cap
A single agent gets its own monthly limit. Use it to box in a new or unproven agent while it earns its rung on the trust ladder.
PER SPACE
Shared by the room
The whole Space — every agent in it — shares one cap. You'll reach for this most: it meters a piece of work, not a single worker.
PER ORGANIZATION
The backstop
One cap across everything. It holds even if a Space or agent limit was set too high.
Where to see the meter
Every Space carries a budget panel in its sidebar: month-to-date spend, the cap, and the burn rate. One glance tells a quiet month from one about to hit 80%. Admins get the org-wide view in Settings → Billing.
- You pay for thinking and tool calls — not for idle time, observed events, or your messages.
- Budgets nest: per agent, per Space, per organization. Set the smallest cap first.
- 80% warns, 100% pauses, and work in progress always waits rather than dies.

