Skip to content
How permissions add up
Public Preview

Policy permissions: model, review, and authoring

Describes how Duale AI permissions accumulate from policies and groups, how to review access across sources, and how to author narrow tenant policies.

This page covers how permissions combine, what a permission contains, how to review access, and how to author and bind Tenant policies.

  • Permission accumulation from policies and groups.
  • Permission fields: action and resource_pattern.
  • Access review across every permission source.
  • Authoring and binding a narrow Tenant policy.
  • Actions required for each feature.

Summaries were generated by AI.

Bind a policy to a custom group to give its permissions to every member. Duale AI allows a request only when one of the caller’s permissions matches its action and resource.

How permissions add up

Duale AI adds up the permissions of every policy bound to the identity itself or to a group it belongs to.

Where a permission comes fromA person or an Agent Identity acts. Permissions add up from the supplied policies, the groups you create, and policies bound to the identity itself. Duale AI allows the operation only when a policy permits that action on that resource; otherwise it refuses.

For example, payments-readers holds a policy with action = agent:read and resource_pattern = agent:payments-*. An Agent Identity in that group can read agent:payments-invoice-review, but not agent:support-triage.

Policies only add access: a narrower policy does not remove access that another policy gives, including a supplied policy. To refuse access on one Library path, use a Library Grant. Some operations also need a stronger sign-in whatever the policy grants.

What a permission contains

FieldMeaning
actionOne operation, such as agent:read.
resource_patternThe resource names on which that operation is allowed, such as agent:payments-*.
  • Field
    action
    Meaning
    One operation, such as agent:read.
  • Field
    resource_pattern
    Meaning
    The resource names on which that operation is allowed, such as agent:payments-*.

A policy you author names each action exactly, without a wildcard such as agent:*, so a new action never joins it.

A resource pattern names one resource exactly, or ends with * to match every resource with that prefix, including resources created later. A bare * matches every resource of every type for the action, so prefer the narrowest pattern that covers what you need, such as agent:payments-*.

A Task resource is task:<agent_id>/<task_id>, where <agent_id> is the Agent Identity that owns the Task. Use task:* for every Task in the Tenant or task:<agent_id>/* for every Task of one Agent Identity. A Library resource is library:<path>, named by the Library’s current path; Library access explains what that means for a rename or a shared path.

A permission on task:<agent_id>/* lets its holder read that Agent Identity’s Tasks, for example in Analytics.

In a resource pattern, the placeholder {self} stands for the caller’s own identity: an Agent Identity’s identifier, or a person’s user ID. For the Agent Identity agent:payments-invoice-review, task:{self}/* matches its own Tasks, and library:agent/{self}/* matches Library paths under agent/payments-invoice-review/. To save or bind a {self} pattern, you must hold the action on the whole resource type, for example task:*.

Review access across every source of permission

Before you change access, check each of these for the person or Agent Identity:

  1. Check the supplied policies. Open each supplied group under Groups and check its Members list.
  2. In Groups, check the Members list of each relevant custom group, then review the group’s Policies and each policy’s Permissions.
  3. Open Policies and read the Bound to column. It names each group, person, and Agent Identity that holds a policy. Review the Permissions of every policy bound directly to the identity.
  4. For a named Library, also review its Library Grants.

The Dashboard lists only the memberships and policies you can read. If you cannot read them all, ask a member of tenant-admins to complete the review.

After a change, wait for the time for access changes to apply, then test one action that must now work and one that must still fail.

Author a narrow Tenant policy

To author or change a policy, you need profile:manage_policy on policy:<name>. The supplied Administrator policy includes it.

In the Dashboard:

  1. Open Policies, choose New policy, and start from a supplied policy or one of your own.
  2. Under Where it applies, choose a scope. The copy keeps only the actions of the original that fit this scope.
  3. Review those actions and remove any that the new policy does not need.
  4. Name the policy and confirm its permissions.

The copy is independent: later edits to the original leave it unchanged. You can save only permissions that you hold.

A new policy gives no access until you bind it. To bind it, you need profile:manage_binding on the target group:<name>, and you must hold every permission in the policy. In the Dashboard, open Groups, choose a custom group, then use Policies > Bind policy. To give one person or Agent Identity individual access, bind the policy to a custom group that contains only that identity; Members and groups covers group membership.

Keep permission management with people: an Agent Identity that holds profile:manage_policy or profile:manage_binding can add access that only a person signed in at aal3 can remove.

Actions for each feature

When you author a policy, the Dashboard lists the actions you can grant for the selected scope. These pages list each operation and the action that it needs:

Identity administration checks these actions: profile:manage_member on the Tenant to invite, and on user:<user_id> to suspend or remove a member; profile:manage_group and profile:delete_group on group:<name>; profile:manage_tier on a supplied group; profile:manage_policy and profile:delete_policy on policy:<name>; and profile:manage_binding on the group or identity that gets the policy.