nothing leaves your mac
Private, on-device AI autocomplete for Mac
Every suggestion is written by a local model on your own Apple Silicon. No cloud, no accounts wired into your keystrokes, no text uploaded anywhere. You type, it finishes the sentence — entirely offline.
Hi Sarah, thanks for the quick turnaround.
The MLX model runs on your Mac, not a server
MacAutocomplete ships a small language model that runs through Apple's MLX framework, directly on your Mac's Neural Engine and GPU. When you pause mid-sentence, the text around your cursor is read, fed to the model locally, and a completion comes back in milliseconds — like drafting the rest of this thought for you. Nothing is batched, queued, or shipped to a datacenter. The model weights live on disk; inference happens in your own process.
Zero outbound network — verify it with Little Snitch
We don't ask you to take "privacy" on faith. While completions are being generated, MacAutocomplete makes zero outbound network requests. Point a network monitor like Little Snitch at the app, unplug your ethernet, turn off Wi-Fi — it keeps working exactly the same, because there was never a round trip to begin with. The only times the app ever talks to a server are optional and obvious: checking for updates, and validating your license key once when you activate a Mac.
Password and secure fields are never read
The app reads the text around your cursor only to predict the next words — and it deliberately ignores anything macOS marks as a secure input. Password boxes, secure entry fields, and the system's Secure Input mode are skipped entirely; no ghost suggestion appears and nothing is captured. You accept a suggestion with ⇥ Tab or →, ignore it by simply continuing to type. That works the same in every app — Mail, Slack, Notes, Messages, your browser.
Why local beats cloud writing tools for privacy
Most AI writing assistants send what you type to a remote API to generate text. That means your half-finished emails, DMs, and notes leave your machine and sit in someone else's logs. A local model removes that risk by design — there's no endpoint to breach, no retention policy to trust, no network to intercept.
| On-device (MacAutocomplete) | Cloud writing tools |
|---|---|
| Text stays on your Mac | Text sent to a remote API |
| Works fully offline | Needs an internet connection |
| Nothing to leak or retain | Logs, retention, breach surface |
| Verifiable with Little Snitch | Trust the privacy policy |
On-device autocomplete isn't a niche tradeoff anymore. See how it stacks up against Cotypist, or read the longer argument in local vs. cloud AI writing.
Frequently asked questions
Is MacAutocomplete really offline?
Yes. The MLX model lives on disk and runs in the app's own process on your Apple Silicon. Pull your ethernet cable, switch off Wi-Fi, or put the Mac in Airplane Mode — completions keep appearing at the caret exactly as before, because generating them never needed a network connection.
Does MacAutocomplete send my text to the cloud?
No. The text around your cursor is read locally to predict the next words and is never uploaded. There's no remote API in the completion path, so your half-finished emails, DMs, and notes never leave your Mac. The only optional network calls are checking for updates and validating your license key once when you activate a Mac.
Can I verify that nothing leaves my Mac?
Yes — don't take it on faith. Point a network monitor like Little Snitch at the app and watch while you type: you'll see zero outbound requests during completion. You can also disconnect from the internet entirely and confirm it still works.
Is my data used for training?
No. Because everything runs on-device and nothing is transmitted, there is no data to collect, log, retain, or train on. Password boxes and macOS Secure Input fields are skipped entirely, so sensitive entries are never even read.
get it
$125 — one-time, lifetime.
Pre-order — ships in ~1 week · lifetime · 3 Macs · runs on your Mac.
Lifetime · 3 devices · 14-day refund (95%).