Proprietary vocabulary that defines how AI agent spend control works. Every term is a concept sipi.bot enforces natively.
A pre-spend policy layer evaluating every agent transaction against your rules before money moves.
The set of rules governing what an agent is allowed to spend — caps, limits, and allowlists.
Caps transactions per time window. Primary defense against retry loops.
Blocks unknown vendors by default. Defense against prompt-injection-driven spend.
A hard ceiling on any single spend. Catches the catastrophic one-off.
Rolling budget across all transactions in a 24-hour window.
Tamper-evident, append-only log of every spend decision.
Transaction pattern deviating from expected. Caught pre-spend or observed post-hoc.
Unexpected cost spike in agent spending patterns.
Uncontrolled agent spend — typically from retry loops or unattended sessions.
Treating each agent as a trackable cost center with its own budget and audit trail.
Rate at which an agent consumes LLM tokens, measured in tokens or dollars per time unit.
Automatic halt when a spend threshold or anomaly is detected.
Emergency stop on all agent spend when a critical threshold is breached.
Maximum allowed amount for a single transaction.
The total spending authority assigned to a specific agent.