Deterministic policies
Define fail-closed governance rules with condition-based evaluation. Know exactly how AI agents will behave.
Zdravo puts identity, policy, evidence, and authorization on the path to agent execution. For supported protected actions, Zdravo returns an explicit ALLOW, DENY, or ASK_HUMAN decision before the action proceeds.
THE AUTHORIZATION LAYER
Define fail-closed governance rules with condition-based evaluation. Know exactly how AI agents will behave.
Persist policy outcomes and related evidence so teams can inspect what a protected agent action was allowed to do.
Keep governance rules separate from model choice and apply them across the client surfaces currently supported.
Require human approval for high-risk operations. Maintain control over critical decisions.
Core capabilities
Security & compliance
PostgreSQL row-level security and organization-scoped authorization on supported data paths
Persistent decision, observation, and provenance records; hash-chain verification is available in supported flows
Server-side envelope encryption and key-version management where implemented
Owner, admin, and member roles with permission checks on supported organization workflows
Current boundaries
How it works
01
Resolve the organization and governed-agent identity before evaluating the request.
02
Check the requested action against the applicable organization policy and evidence.
03
Return ALLOW, DENY, or ASK_HUMAN, then retain the decision and related evidence.
Proof, not promises
The governance workspace includes an attack lab for exercising protected agent actions against changed tools, missing or stale evidence, unearned authority, prohibited data export, and other adversarial conditions.
Start with the guided setup wizard and have governance policies running in minutes.