Velocity Caps for AI Agent Spending
A velocity cap limits how many payments an agent can make in a given time window. Unlike a budget ceiling (total spend), a velocity cap limits the rate of spending.
Velocity caps prevent the 'death by a thousand cuts' scenario where an agent makes thousands of micro-payments below the single-transaction limit. Default: 10 payments per minute per agent.
Sipi Bot applies velocity caps at the agent, endpoint, and tenant level. An agent hitting the velocity cap gets automatic exponential backoff — retry intervals double with each attempt.
What's the right velocity cap for my agent?
Start at 10 calls/minute per endpoint. Monitor for 1-2 weeks. Adjust based on actual usage patterns. The goal is to allow normal operations while preventing runaway loops.
Do velocity caps affect user experience?
For well-behaved agents, no. Velocity caps only trigger on anomalous patterns. Sipi Bot's backoff is transparent to the agent — it just gets slower responses.