Private AI and offline drafting

What Private Mode does

Pro Private Mode drafts using Apple’s on-device model through the Foundation Models framework. LittleDraft prepares a focused brief from your instructions, the available conversation, your selected writing profile, and relevant authorized references. It does not fine-tune Apple’s model or claim that a locally generated reply is automatically accurate.

The selected email and drafting context stay on your Mac for AI processing. The app does not automatically switch to cloud AI if the local model is unavailable, a request is too long, or generation fails. Changing to Cloud AI is a separate explicit choice.

Requirements

• A compatible Mac with Apple silicon and macOS 26 or later.
• Apple Intelligence enabled in System Settings, with its model fully downloaded and available in your region and system configuration.
• Eligible Pro access and a successful online activation on this Mac.
• A supported drafting language. LittleDraft currently enables English and German for Private Mode when supported by the local model.

The documentation’s display language does not expand the local model’s supported languages. A French or Spanish website page does not mean Private Mode can draft in that language.

Activate and choose the mode

1. Install the current LittleDraft release and sign in to Account.
2. Open Preferences and find AI Processing.
3. Select Private Mode. Read the readiness message and complete activation online.
4. If Apple’s model is not ready, finish its download and enable Apple Intelligence before trying again.
5. Once the app reports readiness, draft a short English or German reply and inspect the result.

Pro Monthly access works offline within the licensed paid period. A renewed period needs an online renewal/refresh of the activation. Pro Buy Once checkout is not currently available; do not infer a purchasable lifetime offer from historical or prepared app text.

Offline example

How LittleDraft helps the local model

The app structures the brief, separates your instructions from source material, includes compact style hints, selects relevant context, and checks the available context budget. It can remove repeated quoted text or ask you to narrow an oversized request. This is application-level preparation and checking, not a promise that every local answer will match a cloud model or handle every long document.

Credits and online features

Private drafting does not use cloud reply credits. Account sign-in, licensing, subscription renewal, and email delivery are separate online services. Little Engine research uses online AI and internet. Version 0.20.57 supports a separately approved online research request while Private Mode remains selected; it never silently uploads a private draft. See the release details below. Existing Mail synchronization, user-chosen exports, and operating-system services are outside a promise about local AI processing.

If it is unavailable

Read the specific readiness message. Check the Mac/OS, Apple Intelligence, downloaded model, current Pro activation, language, and context size. Shorten the message or remove unnecessary references when asked. If you prefer to use Cloud AI, choose it deliberately after reviewing the processing difference.

Apple’s introduction to the Foundation Models framework explains the underlying on-device technology.

What changes in the 0.20.57 release

These improvements are included in the public 0.20.57 release. Read the release notes and check your installed version.

The Cloud and Private AI controls distinguish your selected processing mode, local model availability, and Pro activation. A selected mode alone does not establish readiness: the model and the local license must also be available. After online activation, eligible ordinary drafting can work offline while the local license remains valid, without using cloud reply credits.

Local prompts give your direction priority over the incoming question, preserve English or German conversation language, and separate source text from instructions. German replies can receive a bounded local polishing pass. Additional checks help catch wrong-language output, missing identifiers, repeated customer questions, and editing instructions before insertion. These checks reduce specific failure modes; they do not prove that facts or wording are correct.

Local context remains bounded. Cancellation handling helps prevent stale results from being used, and background work is limited when the Mac is under memory, thermal, or power pressure. Private drafting never silently switches to Cloud AI. A paused or cancelled request can require you to retry after the relevant condition improves.

The release was tested with nine fictional English/German scenarios covering tracking, delivery updates, refunds, invoices, support, and new emails, and the native regression suite passed. This is a description of the release’s testing, not a guarantee for every mailbox or Mac.

Private AI is beta in 0.20.57

Private · Beta: Apple’s on-device model is still being tuned in LittleDraft. Local replies can be shorter and less accurate than Cloud AI. Overview and Preferences show this notice, and private drafts carry a beta label. Check every draft carefully before sending.

Your Professional, Friendly, Formal or Direct tone and Brief, As needed or Detailed length now guide local replies. Detailed replies can include an opening and closing, without permission to invent facts or promises. Greetings use a first name, or an explicitly supplied title and surname; LittleDraft does not infer gender.

Local checks can request a rewrite for invented follow-ups, completed-action claims, missing points or apologies, shipping-versus-delivery mistakes, repeated customer text, and inconsistent German du/Sie. These checks reduce specific errors and do not guarantee accuracy. Ordinary short German messages have improved safety-filter handling; a blocked request explains the reason.