Skip to main content
The approval policy protects Founder control while preserving AI autonomy. Every approval request maps to an ApprovalRequest and surfaces in the Inbox.

Approval categories

AP-01 · Required before action

These actions must always be approved before they execute.
  • Create agent
  • Delete agent
  • Archive an active project
  • Spend above the hard cap
  • Rotate provider credentials
  • Modify company-wide defaults
  • Perform destructive bulk changes

AP-02 · Required unless a policy override exists

Default to approval; can be downgraded via policy.
  • Create a project flagged as high-risk
  • Create recurring automation
  • Assign a critical task to a new agent
  • Call external APIs with side effects
  • Use a premium model above the cost threshold

AP-03 · Clarification instead of approval

These trigger clarification items in the Inbox; the run resumes when the Founder answers.
  • Vague Founder intent
  • Missing deadline
  • Missing budget cap
  • Conflicting project ownership
  • Unclear acceptance criteria

Inbox item template

Approval and clarification items render with a consistent shape:
  • type (approval / clarification / report / alert / info)
  • title
  • why blocked
  • recommendation
  • impact if approved
  • impact if rejected
  • deadline (if any)
  • related entities

Resolution behavior

ResolutionEffect
ApproveContinue the run from its checkpoint.
RejectCancel the run or trigger replanning.
Request changeCreate a follow-up planning step.
Clarification answerInject the answer into run context and resume.

Default budget policy

  • Soft cap: create a warning Inbox item.
  • Hard cap: block execution and require approval.
Budget caps are configured in Settings at the company level and per agent. See Budget for reporting and drill-downs.