Agent Spend Risk Calculator
Estimate your worst-case loss exposure and get a risk score from 1–10. No account needed — the math runs in your browser.
Count every tool/function that can trigger a payment, an API call, a cloud provision, a database write on a live system, a crypto transfer, or any spend-adjacent action.
If your most dangerous tool were called by accident, how much exposure would one transaction create? Pick the closest bracket.
Overnight, while you sleep — with no human watching.
If a tool call fails, does the agent keep trying?
What you budget for all agent spend combined — API credits, cloud costs, tool calls.
—
Risk score
—
Worst-case 24h
—
Annual exposure
How these numbers work
The risk model is calibrated from real incidents in the AI Agent Incident Database. Unattended agents multiply blast radius by 3x. Retry-on-failure multiplies by 10x (the median runaway loop in the database exceeds 30 iterations before human intervention). Annual exposure is 12 × worst-case-24h. The risk score uses a log₂ of the budget-to-exposure ratio — a 1 means you're covered, a 10 means your budget wouldn't survive one bad weekend.
Frequently asked
Is my data sent anywhere?
No. The calculator runs entirely in your browser using JavaScript. No input is transmitted to a server. The PostHog analytics capture only the score and configuration for product improvement, not any identifying data.
What's the most common risk pattern?
Retry-on-failure + unattended with >3 spending tools. The median overnight loss in our incident database is ~$47K. Without a velocity limit, a single retry loop on a $100 tool call becomes $3,000 in under a minute.
How do I reduce my score?
Add a spend firewall with three rules: per-transaction cap (limit single-call blast radius), velocity limit (stop retry loops), and daily ceiling (hard cap regardless of loop depth). These three rules drop the worst-case 24h from tools × risk × multiplier to capped × ceiling.