Docs center

Self-hosted docs for Guardrails for AI.

Technical guidance for observability, guardrails, permissioning, and automation in one first-party documentation surface.

Self-hosted docs

Permissioning Boundaries

Role-scoped permissions keep sensitive tools and actions inside explicit boundaries.

docs.cyiro.comproduction-ready guidance

Role-aware tool access

Define which roles can invoke which tool classes, and enforce those checks inline during agent execution.

  • Example: Developer role can only access sandbox tools and routes
  • Example: Team role can manage production routes with approval workflows
  • Example: Enterprise role has full access with audit logging requirements
  • Request: POST /v1/permissions/check {role: "developer", tool: "export_data", route: "prod-payment"}
  • Response: {allowed: false, reason: "role_restriction", required_role: "team"}
  • API Reference: See [Permission Check](/docs/api#permission-check) for full contract details

Escalation and review

Denied actions can trigger structured human review workflows with complete incident evidence attached.

  • Example: Failed tool invocation creates incident with full context
  • Example: Review workflow includes runtime traces, policy verdicts, provider status
  • Example: Approval decisions logged with timestamps and reviewer identity
  • Request: POST /v1/reviews/create {incident_id: "inc-123", reviewer: "alice@company.com", decision: "approve"}
  • Response: {review_id: "rev-456", status: "approved", timestamp: "2024-03-15T14:30:00Z"}
  • API Reference: See [Review Creation](/docs/api#review-creation) for full contract details

Permission inheritance

Hierarchical permission models with workspace and team-level overrides.

  • Example: Workspace-level permissions override global defaults
  • Example: Team-specific tool access with audit trails
  • Example: Temporary permission escalation with time limits
  • Request: POST /v1/permissions/workspace {workspace_id: "prod-ai", role: "developer", tool_access: ["sandbox_only"]}
  • Response: {permission_id: "perm-789", workspace_id: "prod-ai", effective_permissions: ["read_only", "sandbox_execution"]}