Running an AI model directly on your own computer rather than sending data to a remote server, keeping data private and reducing latency.