Coming soon

TypeBlink 0.5: a new name, display settings and smarter spelling fixes

What's new
Guide

Private, offline AI autocomplete explained

Why running autocomplete on your own computer keeps your writing private, what a local model needs, and what to check before trusting a writing tool.

Updated 31 October 2026 · 6 minute read

Autocomplete sees almost everything you write: emails to clients, messages to friends, notes you never meant to share. Where that text goes is the most important question to ask of any writing tool. With TypeBlink, the answer is simple: nowhere. The model runs on your own computer.

Cloud versus local, in one paragraph

Many AI writing tools send what you type to a server, where a large model writes the suggestion and sends it back. That can work well, but your text crosses the internet on every pause, it may be logged, and the tool stops when you are offline. A local tool runs a smaller model on your own processor and memory. The text never leaves your machine, there is no per-request cost, and it keeps working on a plane.

Why local works for autocomplete

Autocomplete does not need an encyclopaedia; it needs to guess the next few words of a sentence you have already started. Modern compact models are good at exactly that. A model of a few billion parameters, such as Gemma 4 E2B, fits in about 3 GB, runs comfortably on a laptop, and answers in a fraction of a second, which is what makes suggestions feel instant.

What stays on your computer with TypeBlink

  • What you type. The words near your cursor go only to the model running on your computer.
  • What it learns. If you let TypeBlink learn from your writing, those examples are stored on your computer, encrypted with a key in your system's secure credential store (Keychain, Credential Manager or GNOME Keyring). You can delete them any time.
  • Your usage. The app has no analytics or telemetry. The word count in Settings is calculated and kept locally.

TypeBlink goes online only to download a model you choose and to check for new versions; neither request contains anything you write.

Offline, for real

Once the model is downloaded, unplug the network and nothing changes: suggestions, typo fixes, per-app rules and settings all keep working. Updates simply wait until you are back online.

Safety rails that matter

  • Password fields are never read.
  • Clipboard context is off unless you turn it on.
  • You decide which apps get suggestions, and you can pause it in any app with one shortcut.
  • Terminals and code editors start switched off.

A checklist for any writing tool

Whatever you end up using, these questions are worth asking:

  1. Does the text I type leave my device? If so, where is it stored, and for how long?
  2. Does it work offline?
  3. Does it read password fields or everything on screen?
  4. Is there analytics or telemetry, and can I turn it off?
  5. Can I choose which apps it works in?
  6. Does it need an account?

For TypeBlink the answers are: no, yes, no, none, yes and no. Details are on our privacy page.

What a local model needs

On a Mac, Apple silicon with 8 GB of memory runs the standard model well; 16 GB leaves room for larger ones. On Windows and Ubuntu, 8 GB of memory is recommended. Either way, plan for about 3 GB of disk space. Read the Mac guide or the Windows and Ubuntu guide to get started.

Type less. Say more.

TypeBlink is free on Mac, Windows and Ubuntu. Set up takes about two minutes, most of it the model download.

Download freeSee how it works