> For the complete documentation index, see [llms.txt](https://docs.brain.fi/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.brain.fi/build/give-an-agent-a-spending-limit.md).

# Give an Agent a Spending Limit

Define and sign a structured policy that Brain evaluates for every action.

Goal: author a structured policy, collect the required signatures, then evaluate actions against the active policy.

### Compose a Policy

Policy authoring uses the structured JSON DSL. Brain does not currently compile plain English into a policy.

```typescript
const policy = await brain.policy.compose("tnt_acme", {
  version: 1,
  lists: {
    trusted_vendors: ["cp_acme_legal"],
  },
  rules: [
    {
      id: "small-trusted-payments",
      applies_to: ["outbound_payment"],
      when: {
        counterparty_in: "trusted_vendors",
        amount_lte: { currency: "USD", value: 5000 },
      },
      execute: "auto",
    },
    {
      id: "other-payments-require-review",
      applies_to: ["outbound_payment"],
      when: {},
      require: ["signer"],
      execute: "confirm",
    },
  ],
});

console.log(policy.policyId, policy.state, policy.signingPayload);
```

### Submit Signatures

Sign the returned payload with the authorized tenant member keys, then submit the resulting signatures.

```typescript
const result = await brain.policy.sign("tnt_acme", {
  policyId: policy.policyId,
  signatures: [
    { address: signerAddress, signature: signerSignature },
  ],
});

console.log(result.activated, result.policy.version);
```

`compose` returns `policyId`, `state`, and `signingPayload`. Policy activation happens as a side effect of `sign` once the required signatures are present.

### Test an Action

```typescript
const decision = await brain.policy.evaluate("tnt_acme", {
  kind: "outbound_payment",
  amount: { currency: "USD", value: "7800" },
  counterparty_id: "cp_acme_legal",
  agent_role: "payment",
});

console.log(decision.outcome);
console.log(decision.matched_rule_id);
console.log(decision.required_approvers);
```

`outcome` is exactly `allow`, `confirm`, or `reject`. It is not an SDK alias layer. This request records a `policy.evaluate` audit event, but it does not create a payment intent.

### Approval Requirements

Policy `require` contains the roles whose approvals the rule needs.

```typescript
const dualApprovalRule = {
  id: "large-payment",
  applies_to: ["outbound_payment"],
  when: {
    amount_gt: { currency: "USD", value: 50000 },
  },
  require: ["cfo", "ceo"],
  execute: "confirm",
};
```

### Update a Policy

Policies are versioned. Compose a new document with the next version, then submit signatures for that new `policyId`.

```typescript
const next = await brain.policy.compose("tnt_acme", {
  version: 2,
  lists: {},
  rules: [dualApprovalRule],
});

await brain.policy.sign("tnt_acme", {
  policyId: next.policyId,
  signatures: [{ address: signerAddress, signature: signerSignature }],
});
```

### What's Next

* [Pay an Invoice Safely](/build/pay-an-invoice-safely.md)
* [Audit Every Action](/build/audit-every-action.md)


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:

```
GET https://docs.brain.fi/build/give-an-agent-a-spending-limit.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.