Back to Blog
Product Update2026-08-20

Budget Alerts Are Now Live in AI SpendOps

Budget alerts are live in AI SpendOps for all plans. You can now put a spend budget on anything the platform can attribute cost to: your whole organisation, a single API key, one provider, or one dimension value such as a team, feature, environment or customer. When spend approaches the line, the right people get an email. When it crosses the line, they get another. And at no point does any of it touch your traffic.

Full details are in the budget alerts documentation.


Budgets that match how the money is actually organised

A single "monthly AI budget" number is rarely how a finance team thinks. Budgets are owned: this product line has one, that customer-facing feature has one, the experimentation key the data team uses has a different, smaller one. Budget alerts follow that structure.

Because AI SpendOps already attributes every request to an API key and to the dimensions you pass on it, a budget can be scoped to exactly the slice of spend someone is accountable for:

  • Organisation-wide: the whole AI bill, across all 16 providers, in one number.
  • Per API key: the production key gets a serious budget, the staging key gets a small one.
  • Per provider: keep an eye on Anthropic spend separately from OpenAI spend.
  • Per dimension value: a budget on team=data-science or feature=chat, using the same dimensions that already drive your cost reporting.
46%
Data science team
team=data-science · $2,760 / $6,000
82%
Chat feature
feature=chat · $4,920 / $6,000
104%
OpenAI spend
provider=openai · $8,320 / $8,000
Three budgets, three scopes, on the Budgets page. Illustrative figures.

Periods work the way reporting periods work: calendar month, calendar year, or your fiscal year, using the financial year start already configured on your organisation.


Warnings at the levels you choose

Each budget takes up to five warning thresholds plus a critical level, so you can shape the escalation to fit the audience. A common pattern: a quiet heads-up to the engineering lead at 50%, a wider warning at 80%, and the critical alert at 100% going to finance.

Each level fires once per period and resets when the period does, so nobody's inbox fills up because spend is hovering around a threshold. If a burst of usage crosses several levels between checks, only the highest one sends an email; the lower ones are still recorded in the budget's history, so the audit trail stays complete.


A forecast, not just a rear-view mirror

Knowing you have spent 60% of the budget is only half the answer. The other half is whether that is fine or a problem, and that depends on the date and the trend.

Every budget's detail page charts cumulative spend day by day, then projects it forward at the current burn rate. The headline does the arithmetic for you: on pace to end the period at 84% of budget, or on pace to hit the threshold on the 24th, before the month is out.

All providers · calendar month · $30,000
Approaching · 72%

At ~$1,080/day, on pace to hit $30,000 around 28 Aug. Projected period total ~$33,500 (112%).

$36k$27k$18k$9kBudget1 Aug8 Aug15 Aug22 Aug29 Aug
Spend to dateProjectedBudget
A budget detail page mid-month: spend to date, and the projection crossing the budget line before the period ends. Illustrative figures.

This is also your early warning for runaway usage. A looping agent, a prompt regression that doubles token counts, a workload someone forgot to switch off: each shows up as a jump in daily burn that drags the projection forward, and the warning levels start emailing while it is still a blip rather than a line item. The budgets page shows every budget as a gauge, and the main dashboard carries a "This month's budgets" panel, so the state of play is visible without going looking for it.


Alerts inform. They never block.

This is a deliberate design decision, not a limitation. A budget alert exists to start a conversation with the person who owns the spend. It should never be the reason a customer-facing feature stopped working at 11pm because a threshold ticked over.

So budget alerts in AI SpendOps are purely informational: spend is tracked, forecasted and alerted on, and traffic flows regardless. Hard control lives where it belongs, in the model and provider policies on each API key, which are enforced at the edge on every request. That split gives finance real-time visibility and engineering predictable behaviour, and neither has to compromise.


Available now

Budget alerts are live for all plans. Create your first budget from the Budgets page in the dashboard: pick a scope, set the amount, choose your warning levels, add the recipients.

Read the budget alerts guide, or start with your first 3 months free.