Notifications can be managed across ecosystems. A centralized hub can implement a learning engine that uses an algorithm to evaluate incoming notifications that are intended for a user that uses multiple computing devices having different ecosystems. The algorithm can be configured to determine on which of the user's computing devices the notifications should be presented given a particular context. Agents executing on the user's computing devices can monitor how the user interacts with the notifications and provide indications of such interactions to the learning engine. The learning engine can then update its algorithm based on the user's interactions to cause future notifications to be delivered to the user via the ecosystem that is most appropriate for a given context in which each notification is received.