Skip to content
Guides: all pages

Guides

Enforcement integration

Choose where to mediate, start in monitor mode, move to enforce mode deliberately, and never describe a posture you cannot evidence.

Release-gated · interface not public

Enforcement points depend on IntentControl, which is release-gated. The design guidance here applies now; nothing on this page is callable.

Accountability Chain:DecisionAction

Choose where to mediate

SurfaceEnforcement point
MCP toolsAn MCP gateway between the host and the server
HTTP APIsAn HTTP gateway in front of the API
Your own codeAn embedded check at the single choke point for consequential calls

Only an inline point, in the action path and able to block, can ever evidence enforcement. An advisory point informs its caller and never does.

Roll out in two modes

  1. Monitor mode. Actions run whatever the outcome. Posture is observed. Use it to learn what your Intents and policies would have decided.
  2. Enforce mode. Outcomes are obeyed. Posture becomes enforced only while an active inline point keeps reporting consistent mediation within 24 hours.

Describe posture honestly

  • Say “enforced” only with the scope note: enforced only for actions routed through the enforcement point that reported mediation.
  • Never say “enforced” for evaluated, advisory, observed or local-only decisions.
  • Never say an allow means an action succeeded.

This page is explanation, not authority. Current state is only ever the live Resolver record. Live facts on this page were measured read-only on 2026-10-10.