Local AI and cloud AI

Neither option is automatically right for every person or every message. The practical difference is where the work happens and who receives the information needed to do it.

At a glance

Question	Local AI	Cloud AI
Where does the AI run?	On your own computer	On a provider's computers
Does message text leave your computer?	Normally no	Yes, selected content is sent over the internet
Does it require internet access?	Not after setup for ordinary use	Yes
Is there a usage charge?	Usually no per-message charge	May be billed by usage or subscription
How powerful is it?	Depends on your Mac and installed model	Often provides access to very large models
Who controls model updates?	The user chooses when to install or change a model	The provider may change models or service behavior
What happens during a provider outage?	Local processing may continue	Processing may be unavailable
Is setup required?	Yes; a local model must be installed	Usually an account and API access are required

When local AI may be the better fit

Local AI may be preferable when:

• the mailbox contains sensitive or confidential material;
• an employer, school, client, research agreement, or professional duty limits external processing;
• the user wants processing to continue offline;
• predictable cost matters; or
• the user wants direct control over the installed model.

When cloud AI may be the better fit

Cloud AI may be preferable when:

• the task benefits from the most capable available model;
• the computer does not have enough memory or processing capacity for a suitable local model;
• the user has reviewed and accepts the provider's terms and data controls; or
• convenience and speed matter more than keeping all processing on the device.

The choice can differ by account

A person may reasonably make different choices for different mailboxes.

For example:

• a personal account might be allowed to use a cloud AI provider;
• a university account might be limited to local processing;
• a clinical, legal, or research account might have AI disabled entirely; and
• another account might use a provider approved by the relevant organization.

Spillway's long-term design treats those as separate trust decisions. Combining accounts in one interface does not mean that all account data should be processed in the same place.

What Spillway will show

Before a processing option is enabled, Spillway should make clear:

• the name of the selected AI service;
• whether it runs locally or remotely;
• what kind of message content may be sent;
• which accounts may use it; and
• whether the user may incur provider charges.

The aim is informed choice, not a blanket claim that one technology is always better.