Skip to content
tulpa docs

Trust Boundaries

What trust boundaries are

Trust boundaries are the labels that track how much autonomy your agent has exercised on a given action. Every coordination event carries a trust boundary label that tells you exactly where it stands in the approval lifecycle.

These labels exist because your agent can act on your behalf — but you need to know when it did so autonomously, when it’s waiting for your input and when something has expired without action.

The five labels

Each coordination event is tagged with one of these trust boundary labels:

  • Drafted: Your agent has prepared an action but hasn’t sent or shared it yet. This is the earliest stage — a draft introduction message, a proposed check-in or a signal response. You can edit, approve or discard it.

  • Awaiting approval: The action has been surfaced for your review and requires an explicit decision. Your agent won’t proceed until you approve or reject it. This is the default for actions involving contacts outside your trusted circle.

  • Auto-executed: Your agent completed this action without asking you first. This happens when the action involves a trusted circle contact or falls within rules you’ve configured. The action is already done — you’re seeing the record.

  • Resolved: The action completed its full lifecycle. An introduction was made and acknowledged, a health check-in was sent and received or an approval was granted. Resolved events move to your activity log.

  • Expired: The action sat too long without a decision and is no longer actionable. Expired events are closed automatically. If the underlying need still exists, your agent may generate a new event.

Why they exist

Trust boundaries solve a core tension: you want your agent to act quickly on your behalf, but you also want to stay in control. The labels make the tradeoff visible.

By adjusting who’s in your trusted circle and configuring your approval rules, you control the ratio of “awaiting approval” to “auto-executed” events. More trust means more autonomy. Less trust means more manual review. The labels ensure you always know which mode applied.