Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.altheia.xyz/llms.txt

Use this file to discover all available pages before exploring further.

check(action: ActionDescriptor): Promise<Decision>
Sometimes you want a policy decision without executing the action — to decide whether to even propose it, to surface a UI hint, or to test what your policy allows. check() calls /sdk/agent_check with the same payload guard() would, but never runs anything afterwards and never reports an outcome.

​
Returns

type Decision = {
  allowed: boolean;
  reason?: string;          // human readable, e.g. "exceeds 50 per-tx USDC"
  reason_code?: string;     // machine readable, e.g. "over_per_tx_cap"
  audit_event_id: string;   // UUID, points at the audit row
  cached: boolean;          // true if served from local cacheTTL
  evaluated_at: string;     // ISO timestamp
};
A check() call still writes an audit row (so you can see in the chronicle that the decision was requested) but no tx_signature is attached.

​
Failure-mode behavior

If failureMode: "open" and the backend is unreachable, check() resolves with: 
{
  allowed: true,
  reason: "sdk_unreachable_failopen",
  reason_code: "sdk_failopen",
  // ...
}
If failureMode: "closed", check() throws AltheiaConnectionError.

​
Example

const decision = await altheia.check({
  type: "transfer",
  asset: "USDC",
  amount: 1200,
  target: recipient,
});

if (!decision.allowed) {
  notifyOperator(`would deny: ${decision.reason}`);
}

​
When to prefer check() over guard()

  • You want to show the operator “would this be allowed?” before they confirm.
  • You’re testing a new policy without risk of accidentally running anything.
  • Your action doesn’t sign on-chain (e.g. you’re only inferring intent).
In every other case, guard() is the right call — it gives you the audit row with the signature attached.