AI usage is billed as credits, which raises a reasonable question: how do I keep it predictable? The answer is that the controls are built for exactly that. Here is how to set a budget for your agents and keep spend tracking value, not surprising you.
Spend follows the work
Every agent action draws from a credit balance, so cost rises only when work is actually done. That alignment is the point: a quiet week costs little, a busy one costs more, and you only pay for work that actually happens.
How credits work
Usage is metered and billed as credits from a balance you hold. You can see spend broken down by action and by app, so the bill is never a black box. The richest, cheapest path is bringing your own agent, where your own AI does the reasoning and SOIS spends nothing on your behalf.
Set a cap and auto-recharge
Set the ceiling you are comfortable with. Turn on auto-recharge so the balance tops up by a fixed amount when it runs low, with its own cap, so your agents never stall mid-task and spend never runs away. You decide the numbers; the system stays inside them.
See where it goes
Predictability comes from visibility. Because usage is itemised, you can see which tasks and which apps consume the most, and tune accordingly, the same way you would manage any operating cost. A plan that bundles monthly usage at a better rate smooths things further if your volume is steady.
A sensible starting budget
Start small. Put one painful task on the agent, set a modest cap, and watch a week of real usage. That gives you a grounded number to scale from, rather than guessing up front. You expand the budget as you expand what the agent runs.
Usage-based does not mean unpredictable; it means controllable. See how credits work, check pricing, or launch a workspace and set your cap.