A key security feature of 'Dots' is the ability to give agents their own auditable and scopable permissions, rather than using human login credentials