Limits and budgets
Admin → Budget, Spending, Requests
AI usage is metered. Without limits, one enthusiastic user or a runaway agent can spend a month’s budget in a day. Lucy gives you three layers of protection, all in euros.
1. The organisation budget
Section titled “1. The organisation budget”A monthly ceiling for the whole organisation. You set it, up to the maximum in your contract. When it is reached, new answers do not start.
2. Per-person limits
Section titled “2. Per-person limits”Limits per person over three windows. Set any combination.
| Window | Behaviour | Good for |
|---|---|---|
| Five hours | Rolling, starts with the first answer | Smoothing bursts, discouraging marathon sessions |
| Week | Resets Monday 00:00 UTC | Fair share across a team |
| Month | Resets on the 1st, UTC | Hard cost cap |
A person’s limit is, per window: their own override, else the most generous override among their groups, else the organisation default per member, else none.
Overrides let you give analysts more headroom than the default, or cap a trial user. Use groups for roles, individuals for exceptions.
3. Budget alerts
Section titled “3. Budget alerts”Alerts notify you before the limit bites. Choose a scope, a window (week or month), a threshold, and optional extra recipients such as finance. Admins are always notified. Each alert records when it last fired, and firing is written to the audit log.
Requests: the release valve
Section titled “Requests: the release valve”People who hit a limit can request more. You see pending requests in Admin → Requests (a count shows on your account), approve or decline with a note, and approving applies the change immediately: a new override for that window, or a model grant.
This keeps limits firm without making your admins a bottleneck: people ask with a reason, you answer in one click.
What people see
Section titled “What people see”Members never see amounts. They see how much of their allowance is used and when it resets, which keeps costs private and conversations about it simple.
Checks and charging
Section titled “Checks and charging”Limits are checked before an answer starts and charged when it finishes. One answer can slightly exceed a limit; the next will not start. Each answer is charged at the price of the model that wrote it.
Spending and chargeback
Section titled “Spending and chargeback”Spending and Usage show cost and consumption by day, person or group, which supports internal chargeback and invoice reconciliation.
Examples
Section titled “Examples”A month with a hard ceiling
Section titled “A month with a hard ceiling”Situation: The budget for the pilot is fixed. Setup: Organisation monthly budget at the agreed figure, a budget alert at 70% of the month with finance as an extra recipient. Result: Finance hears at 70%, not after the invoice. If the budget is reached, new answers wait until the month resets.
Fair share, with room for the heavy users
Section titled “Fair share, with room for the heavy users”Situation: Most people use a little. The reconciliation team uses a lot. Setup: A default per-member limit for the week and the month, plus an override for the Operations group. Result: Everyone has an allowance, and the team that needs more has it, without raising limits for all.
The request that saves an escalation
Section titled “The request that saves an escalation”Situation: A risk analyst hits the five-hour window during a deadline. Setup: They choose Ask for more with a reason. You see it in Requests. Result: You approve a higher limit for that window in one click, with a note. No ticket, no meeting.