Sanction

One-off human approvals for exact AI agent actions, within the trust and governance layer for autonomous AI agents. Authorizes spend against policy, injects encrypted credentials under short-lived scoped tokens, and enforces clearance-level access — every decision logged and auditable.

Skills

• Request one-off tool approval
  Ask an authorized person to approve or reject an exact tool action (tool, server, and arguments) with require_approval: true. Approval mints an expiring, one-use grant; the caller repeats the identical request with that grant before acting. Hard denials still apply; changed arguments and reused grants are refused. Cooperative: the caller must ask and honor the answer.
  approvalhuman-in-the-looppolicygovernance
• Authorize spend
  Decide whether an agent may make a purchase/subscription/transfer under wallet policy. Returns approved, denied, or escalated with a typed code and remediation hint. Call before any financial action.
  paymentspolicygovernancebudget
• Inject credential
  Exchange a short-lived scoped execution token for a decrypted credential the agent is cleared to use. Every injection is audit-logged; raw values never appear in logs.
  credentialsvaultsecretsclearance
• Track LLM token usage
  Record token counts and cost of an LLM call against the wallet's daily budget for spend tracking and audit.
  budgetobservabilityaudit

How to call

https://getsanction.com/api/v1

Listed in

Directories this entry was found in.