Scenario: Multi-Agent Compounding
A coordinator spawns 50 subagents. One API they all depend on fails, and every subagent retries — 50 loops, compounding.
What happens without a firewall
Each retry bills the same operation. 50 agents × N retries = a large bill in minutes, with no single transaction looking wrong.
What happens with sipi.bot
The shared daily ceiling catches the aggregate: once the fleet's spend crosses the cap, further attempts are BLOCKED.
The velocity limit catches it sooner: too many transactions per window stops the loops at the source.
The rule set
Shared daily ceiling across the coordinator's agents + velocity limit (e.g. max 10 transactions/60s per agent) + category rule.
The decision
{
"decision": "BLOCKED",
"reason": "Velocity limit exceeded: 60 transactions in 60s",
"rule_id": "rul_velocity_02",
"transaction_id": "txn_9f2c11"
}The fleet can't compound what the cap won't allow.
FAQ
Does one cap cover all subagents?
Yes — rules can be scoped to the coordinator's agents, so the fleet shares one ceiling.
What about legitimate fan-out?
Set the ceiling at a level that fits legitimate work; the cap only bites when the fleet exceeds it.
Related
Stop the next $12,400 night.
One API call (or MCP tool) in front of every agent transaction — APPROVED, BLOCKED, or FLAGGED, deterministic, ~5 ms, fully logged.
See plans — from $99/mo Try a live check