USE CASES where governance earns its keep
A single agent conversation can trigger hundreds of sub-cent purchases — API calls, data pulls, model inference, other agents' tools. That's exactly where an unbounded wallet turns a bug into an invoice. Here's who's spending, what goes wrong without a governor, and the SpendVeto controls that stop it.
An inventory agent predicts a shortage and places supplier orders on its own, settling in stablecoins. A bad forecast or a retry loop can order ten times what's needed before anyone looks.
A compliance bot pings a paid sanctions database and buys supplier-risk scores per check. Charged per query, a mis-scoped batch job can rack up thousands of paid lookups overnight.
A triage agent resolves common tickets and escalates hard ones, paying per model call and per tool. A prompt-injected ticket can steer it into paying an attacker or looping a premium tool.
A copilot runs data pulls and summaries against metered datasets, billed by the query. "Analyze everything" becomes an unbounded fan-out of paid queries — precise consumption, no ceiling.
Agents pay other agents for data and API access in real time, at machine velocity. A chain of micro-payments across untrusted counterparties is impossible to reconcile after the fact — or to stop mid-chain.
Claude Code / LangChain / CrewAI fleets call paid models, review tools, and sandboxes across many parallel sub-agents. One runaway parent can fan a budget out across dozens of children before a human notices.
The playground runs SpendVeto's real policy logic in your browser: set a budget, fire agent spend at it, and watch each attempt pass, pause for approval, or get blocked with a reason.