For the Privacy-Conscious

The model runs on your PC.
Your keystrokes stay there too.

Most AI writing tools work by sending what you type to a server. Typeahead was built the opposite way: the language model runs locally through llama.cpp, suggestions are generated and discarded in memory, and there's no server that ever sees your text.

 Get it on Microsoft Store See full feature list
Qwen 3 + llama.cpp
Runs entirely on-device
No account
No sign-in, ever
Zero text telemetry
Nothing you type is logged
Local-first, not local-optional.

How Cloud Writing Assistants Work

The typical browser-extension writing assistant intercepts what you type, sends it to a remote server for processing, and streams a suggestion back. That round-trip means your draft, your emails, and your messages pass through infrastructure you don't control.

  • Text is transmitted off-device for every suggestion generated
  • An account is usually required, tying your writing to an identity
  • Vendor privacy policies vary, and can change after you've started using the tool
  • No connection means no suggestions — the tool stops working offline

How Typeahead Works

Typeahead downloads a compact language model once, then runs it locally through llama.cpp every time it generates a suggestion. There is no server in the suggestion loop at all.

  • The only network calls Typeahead makes are downloading the model and checking for app updates
  • No keystroke logging — suggestions are computed and discarded in memory, not stored or transmitted
  • No account, no sign-in, no analytics tied to what you write
  • Works with Wi-Fi off, once the model is on disk
Protections
What's protected by default

Nothing here needs to be configured to be safe — these protections are on out of the box.

Password fields are always ignored

Typeahead never generates or shows suggestions in a password input, and there's no setting that turns this protection off.

Password managers & finance apps protected

These app categories are excluded by default, so Typeahead doesn't read or suggest inside them without you explicitly changing that.

No server in the suggestion path

The model runs through llama.cpp on your own hardware. Generating a suggestion never involves a network request.

No text telemetry, no account

There's nothing to sign into and nothing tracking what you type. The app doesn't know who you are beyond a Microsoft Store license check.

Comparison
Typeahead vs. cloud writing assistants

The category is full of browser extensions that route your keystrokes through a server. Typeahead doesn't.

  Typeahead Cloud writing assistants
Where text is processedOn your deviceSent to a server
Account requiredNoUsually yes
Works offlineYesNo
Password field protectionAlways on, no toggleVaries by vendor
Typing telemetryNoneVaries by vendor
Pricing$9.99 onceMonthly subscription
Privacy questions
Does any of my typed text leave my computer?
No. Suggestions are generated by a model running locally through llama.cpp. The only network traffic Typeahead makes is downloading the model itself and checking for app updates — neither involves the text you type.
Does Typeahead log my keystrokes?
No. Suggestions are computed and discarded in memory. There is no persistent log of what you've typed.
Do I need an account to use it?
No. There's no sign-in and no account tied to your writing. Licensing is handled through your Microsoft Store purchase.
Is it safe to use around passwords and financial apps?
Yes. Password fields are always ignored, and password manager and finance apps are protected by default, with no setting that disables this protection.
What model does it run, and where does it run?
Typeahead runs Qwen 3 through llama.cpp directly on your PC. You choose a fast (~400 MB), balanced (~1.1 GB), or maximum quality (~2.5 GB) model based on your hardware.

An autocomplete tool with no server to trust

One-time purchase, no subscription. Windows 10 / Windows 11.