Permissions

AgentPermissions is the model that constrains what the agent is allowed to do in your app. It's a simple, explicit set of switches and lists — no implicit inference — so you always know exactly what surface area the agent can touch.

The permissions object

interface AgentPermissions {
  canNavigate: boolean;
  canFillForms: boolean;
  canInteractUI: boolean;
  canSubmitForms: boolean;
  allowedScreens?: string[];
  blockedScreens?: string[];
  allowedActions?: AgentActionType[];
}

Field	Meaning
canNavigate	Whether the agent may emit navigate actions.
canFillForms	Whether the agent may emit form_fill actions (filling fields, without necessarily submitting).
canInteractUI	Whether the agent may emit ui_interaction actions (press, scroll, swipe, focus, set value).
canSubmitForms	Whether a form_fill action is allowed to set submitAfterFill: true.
allowedScreens	If set, restricts the agent to only this list of screen names — everything else is off-limits.
blockedScreens	Screen names the agent may never navigate to or act on, even if otherwise allowed.
allowedActions	If set, restricts which AgentActionType values the agent may use at all.

Allow/block-list semantics

allowedScreens and blockedScreens work together as a coarse allowlist/denylist over screen names:

• If allowedScreens is set, only screens in that list are in scope — anything not listed is treated as off-limits, regardless of the boolean flags above.
• blockedScreens always wins: a screen listed there is excluded even if it also appears in allowedScreens.
• If neither list is set, all screens are in scope, and the four boolean flags are what constrain the agent's behavior.

allowedActions works the same way at the action-type level: if set, it's the exhaustive list of AgentActionType values the agent may use (navigate, form_fill, ui_interaction, scroll_list, confirm, custom); anything not listed is disallowed.

Default behavior

If you don't configure permissions at all, the SDK defaults to fully open: all four booleans are true and no allow/block lists are set. Permissions are opt-in to restrict, not opt-in to enable — so if you want the agent scoped down, you need to configure it explicitly.

Where permissions are configured

Permissions can be set in more than one place, and they layer together:

• SDK config — passed as AppilotsConfig.permissions when initializing the SDK (see SDK: AppilotsProvider).
• .appilotsrc — the same shape, set once in your project's .appilotsrc file so it applies without touching app code.
• The dashboard — configured per project at app.appilots.com, letting you change the agent's allowed surface without a new app build. See Dashboard: permissions, budget, personalization for how this is managed.

For how permissions are actually applied at runtime — including the SDK's local defense-in-depth checks alongside server-side enforcement — see SDK: permissions.

Related pages

• Agent actions — the action types these permissions gate
• SDK: permissions — applying AgentPermissions in your app
• Dashboard: permissions, budget, personalization — configuring permissions from the dashboard