Intertrace Docs
Sign in

Policy setup

Define allowed tools, domains, and approval gates.

Step by step

  1. List tools your agents may call and mark high-risk actions.
  2. Define data domains (PII, secrets, customer data) and blocked prompt patterns.
  3. Set internet usage rules and human approval requirements for sensitive operations.
  4. Set organization rules in Policies (Access, Data & attacks, Approvals), then tune each agent on its own page (Connections, Data, Approvals, Limits). An agent can tighten the organization's rules, never loosen them.
  5. Optional: give an agent a role. A role decides, per consequence tier, whether a call runs, asks in production, always asks, or is blocked. New runtime agents start in a learning period that records what they use; nothing loosens until a person reviews and locks in the proposal.
  6. Optional: validate / diff / apply YAML from Policies → Policy as code or itr policy validate|diff|apply. Our rules: format and Preloop v1.0 policy documents are both accepted (imports are tighten-only, and unmapped parts come back as warnings). Applied YAML compiles into authorize, not a second engine. Control plane.