The encryption keys are supposed to be generated client side. Obviously you can’t verify this since 1password is closed source, but if your threat model involve

The encryption keys are supposed to be generated client side. Obviously you can’t verify this since 1password is closed source, but if your threat model involves the client misbehaving, then you should be using any proprietary password manager. Cloud vs not Cloud is irrelevant.