Define how each agent can spend.
Spending controls decide what an agent may buy, how much it may spend on one call, and how much it may spend in a month. They are set on the key the agent uses, and they are checked before every call reaches a provider. A misbehaving agent cannot burn the line. A well-run one keeps growing it.
The controls available today
- One key per agent
- Create a key for each agent and name it. Every transaction the agent makes is attributed to that key, so limits and history stay separate between agents.
- Category restrictions
- Restrict a key to the categories it needs. The categories are Data, Search, Scraping, Media, Compute, LLM router, and Outbound. A key with no restriction can spend in all of them.
- Monthly limit per key
- Set the most a key may spend in a calendar month. When the limit is reached, further paid calls on that key are refused until the month rolls over.
- Per-call ceiling
- The agent sends a maximum cost with every call. If the quoted price is higher, the call does not run. This is the control the agent applies to itself on each purchase.
- Account ceiling
- The account as a whole has a limit on what it can owe. It is set per account and does not depend on how many keys exist.
- Approval for outbound work
- Emails and messages an agent drafts wait for your approval before they are sent. Automated sending needs an explicit rule, and every rule carries a daily cap.
A policy, written out
This is what a policy for a research agent looks like on the account.
- Key
- research-agent
- Categories
- Search, Data
- Monthly limit
- $25.00
- Per-call ceiling
- Set by the agent on each call
- Not permitted
- Media, Compute, LLM router, Outbound
With this policy, the agent can run searches and enrich contacts up to $25 in a month. It cannot generate media or start a sandbox, and no single call can cost more than the agent said it would pay.
What happens when a limit is hit
- 01The call is refused before it reaches the provider
Nothing is charged. The provider is never contacted.
- 02The refusal is recorded
It appears on the account with the reason, so you can see what the agent tried to do.
- 03The agent gets the reason back
The response says which control refused the call. The agent can carry on with work that is within its policy.
Set up a key for your agent.
Sign up, name the key, choose its categories, and give it a monthly limit.