Use case
How to detect a runaway AI agent before it burns your budget
A strict budget refuses calls at the limit. Detection can flag unusual spend earlier, but it alerts rather than blocks.
Spendline runs detection hourly and on demand from the dashboard: a daily-spend anomaly check against each agent's recent baseline, and runaway-cost rules you write. Alerts go to a Slack or Discord webhook. Detection alerts; it does not block calls.
How it works
Three separate mechanisms. The anomaly check flags a day's spend that is more than twice the recent baseline and at least $0.10. Runaway-cost rules you define fire on an absolute dollar amount per day or a percentage above a seven-day baseline. Budget threshold alerts fire synchronously after billable calls at 50, 80, 90, 100 and 120% of each configured budget, at org, team, agent or customer level. A burn-rate forecast also exists but is switched off unless enabled for the deployment.
{
"name": "research-agent daily spend",
"entity_type": "agent",
"entity_value": "research-agent",
"threshold_type": "percentage",
"threshold_value": 300,
"period": "daily"
}What to know before relying on it
Hourly detection can lag a fast loop by up to an hour, and it never stops a call. To refuse calls, pair it with a strict budget. Delivery is Slack or Discord webhooks only; there is no email alert.
Why it matters
A strict budget is the last line of defence. Detection that runs regularly lets someone look before the limit is reached.
Frequently asked questions
Does detection run automatically, or do I have to trigger it?
Both. It runs hourly for every account, and the Detect button runs it immediately. Each anomaly or rule breach is raised at most once per day.
What channels can an alert go to?
A Slack or Discord incoming webhook. There is no email delivery today.
See what this would look like in your product.
Start with a guided 30-day free pilot on live production traffic. The 30 days start when your traffic flows through Spendline, not at sign-up. We agree the scope, success criteria and paid terms before you start. There is no Spendline subscription fee during the pilot; your model-provider charges still apply.
Prefer email? fida@spendline.ai