How teams replace spreadsheets with billing that runs itself.

Customers here prepay for compute credits that burn down mid batch-job. A dry wallet can mean a stalled pipeline and a support ticket at the worst possible moment — batch jobs don't pause politely when a balance hits zero, they fail wherever they happen to be in a multi-hour run.

There was a balance check before each job kicked off, but nothing watching mid-run. A job that started with plenty of headroom could still burn through its wallet hours in, at 2am, with nobody watching a dashboard.

Credit wallets are a native object in Kribana, not something bolted onto a usage ledger after the fact — top-ups, low-balance alerts and burn-down tracking are first-class, so the balance updates with every usage event instead of a periodic reconciliation. The agent watches every account's balance continuously and fires a one-click top-up link the moment a balance crosses a configurable threshold, well before a job would actually stall.

Because the threshold and the top-up flow are configured once per account, the same protection applies whether a job runs for ten minutes or ten hours — the agent doesn't need to know what's running, only that the wallet is heading toward zero.

Pipeline stoppages from an empty wallet dropped to zero since launch. The support tickets that used to open at odd hours because a batch job died mid-run stopped showing up, because the wallet gets topped up automatically before the balance actually reaches zero rather than after a job has already failed. For a company selling prepaid compute credits, that's the difference between prepaid billing being a source of churn risk and being invisible to the customer entirely.

It also changed what a low balance means internally. Before, a low-balance alert was a signal someone had to manually chase — a Slack message, a support ticket, a customer waiting on a top-up link someone had to generate by hand. Now it's a workflow that runs itself end to end, and the team's attention goes to the accounts that actually need a judgment call, not the ones that just need a top-up.

The alert-only mode turned out to matter for one segment of accounts in particular: large customers whose finance teams wanted visibility into every top-up before it happened, not automatic action. Being able to run the same detection logic in a mode that surfaces a recommendation instead of executing it meant those accounts didn't need a different system — just a different setting.

“A wallet dropped to 3% mid-batch on a Saturday morning. It was topped up in six minutes. Nobody on either team was awake for it.”