Budget
The Budget page is where you cap employee AI spend and see who is spending it. It has two tabs, Organization and Group, and one of them enforces: a Group budget is the cap, and the organization number is what you plan and report against.
Read that order carefully before you rely on it. Setting an organization figure alone caps nothing, and it is not even stored: the dialog labels it “Not saved yet” and it resets when you reload. The Group amounts are what refuse a request.
Set a Group budget
Open Edit budget and the dialog lists every Group synced from your identity provider with its member count. Each Group takes an amount in USD, as either a dollar figure or a percentage of your organization number, and the two fields stay in step so you can think in whichever one your finance team uses. The minimum is $0.01, and clearing a Group’s budget is not supported yet, so blanking a field is refused rather than removing the cap.
Each Group’s budget is written to that Group’s own Gateway project, which is the thing that enforces it. A Group with no project provisioned yet cannot carry a budget: the amount is accepted in the dialog and that row fails on Confirm with a message saying reconcile has not run for the Group. Other rows still save, so a partial write leaves the dialog open with the failures named.
Budgets run on a monthly window, which lines up with invoicing and is what most organizations want. Daily and weekly windows exist in the API for a pilot or a Group you are still finding a baseline for, where a month is too long to wait for a signal, but the dashboard has no period control: a Group keeps whatever period it already carries and a new one gets monthly.
The window resets on its own cadence, so a daily budget gives the Group a fresh allowance each day, shared across its members.
What happens when one is hit
Workforce budgets are hard caps. Once a Group exceeds its amount, requests from that Group are refused with a 402 until the window resets or you raise the number. This is not configurable from Workforce: Gateway’s own default only warns, and a budget that looks set while quietly letting spend continue is worse than no budget, so Workforce always writes the enforcing mode.
Refusals happen before any vendor is called, so a blocked request costs nothing.
Budget changes can take up to an hour to take effect. A raised limit does not unblock traffic instantly, and a new limit does not protect you instantly.
Because the cap blocks rather than warns, size a new Group’s first budget above what you expect rather than at it, and read a window of real spend on the Group tab before tightening. Blocking an employee mid-task costs more than a small overrun.
Read where the money goes
The Organization tab carries three figures, Total spent, Remaining, and Days to reset, over a pace chart that says whether you land over or under. Total spent is real; Remaining is Total spent against the unsaved organization figure above, so it is only as durable as the number you typed this session. The time period selector at the top is display-only for now, since there is no per-period history to switch to.
Two charts sit underneath: daily cost split by Group, and budget by Group. They are how you learn that one team is most of the bill before the month ends. The per-Group totals are real, but there is no per-day-per-Group data behind them yet, so the daily shape is a distribution of each Group’s month-to-date spend rather than what happened on a given day. Read them for proportions, not for the day a spike landed.
The Group tab is the per-Group read: Group, Trend, % used, Budget, Spent, and Remaining, with Set budget or Edit at the end of each row. The trend sparkline is shaped the same way as the charts above. Member counts are on the Edit budget dialog rather than this table, and they are worth pulling up: a Group of 4 at 80% used and a Group of 200 at 80% used are different situations, and spend per employee is the number worth comparing across teams.
For one person, open them from Provisioning. An employee’s page shows their Spend summed across every key they hold, with no cap of its own: each key bills the project of a different Group, each with its own budget, so no single ceiling applies to one employee. To limit an individual, change the Groups they belong to, or narrow the models their key may reach on Model access.
Spending less without capping
A budget caps spend; these three reduce it. Each is a configuration change, invisible to employees.
The third is only worth it where nobody is waiting on the answer, which is why it belongs to specific jobs rather than a whole Group.
Insights names the candidates for all three each night: the calls running on a model more capable than the job needs, the prefixes being re-sent, and the jobs that could move tier.
Next
Spend here is provider cost per request; for what your organization is actually invoiced, see Billing and usage.