Everything TypeSeer does, in detail

The core

Ghost-text autocomplete, everywhere you type

TypeSeer reads the text field you’re focused on and predicts how your sentence continues. The suggestion appears as grey ghost text floating exactly at your cursor — your document is never modified until you accept.

• Works system-wide: Mail, Notes, Slack, browsers, and most other Mac apps
• Press Tab to accept the next word
• Each Tab accepts another word; optionally make a quick double-Tab accept the whole suggestion in General settings
• Type something different and it disappears — it never gets in your way
• Type what it suggests and it stays put, shortening as you go

Self-learning

It learns how you write — on your Mac

Personalization is optional and can be enabled during setup or later in Settings. TypeSeer then builds a private memory of your writing from the sentences you finish. It picks up your vocabulary, the names and terms you use, and your phrasing, then draws on them when they fit what you’re typing now.

How it stays honest: it learns only from completed sentences, never the one in progress or a suggestion you ignored. Text you accept can become part of a learned sentence once you finish it. Learned names and phrases are offered only when the current text relates to them. It is prompt-based retrieval on your device: no fine-tuning, no cloud, no account.

• Learns your names, terms and turns of phrase from what you’ve already written
• Review Data shows every learned sentence with understandable evidence such as language, application and source
• Export a readable copy, delete one learned sentence, or clear the active collection — Settings → Context awareness → Personalization
• The default base model can imitate style examples; a compatible instruction-tuned custom model can instead follow written role, audience or tone guidance

Spelling fixes

One-Tab spelling fixes, in any app

Finish typing a misspelled word and TypeSeer shows a subtle orange hint with the correction — press Tab to apply it. Nothing is ever changed for you: if you ignore the hint, your typo stays exactly as you typed it.

• One Tab to fix the word you just typed
• Checked against the languages you write in, so a Polish word is never “fixed” into an English one
• Can be toggled on or off in Settings → General

Word completion

Finish long words in one keystroke

As you type a word, the local model uses the sentence around it to decide whether the letters at your cursor are already complete or should grow into a longer word. Type “tomo”, press Tab, get “tomorrow”. Your Mac’s dictionaries still help TypeSeer validate and clean up the result.

Honest limits: completion is contextual rather than a dictionary shortcut, so it can still choose the wrong word or stay quiet. It is always a suggestion, never an autocorrect.

Your context, your rules

Suggestions that fit what you’re replying to

Turn on Context awareness and TypeSeer reads the message you’re replying to and the thread around it, so its suggestions match the conversation instead of guessing. In our testing it noticeably improved suggestions in Mail, Outlook, Safari and Teams. It can also draw on what you last copied, for when you’re writing about something from another app.

All of it happens on your Mac: TypeSeer reads only the window you’re typing in, never password fields, and nothing ever leaves your machine. Both parts are off until you turn them on — during setup or in Settings → Context awareness.

Website affinity is separate and opt-in. In supported browsers, TypeSeer can prefer writing learned on the same website, then fall back to browser-level and general writing. It stores only the normalized hostname — never the page path, search, title, port or account — and shows that evidence in Review Data and exports.

• Reads the on-screen conversation to fit your reply — entirely on your Mac, never from password fields
• Clipboard awareness is optional too, for writing about something you copied from another app
• Disable TypeSeer completely in individual apps from the menu bar
• Use global or per-app style examples with the default base model; compatible instruction-tuned models use written guidance instead
• A snippet containing {clipboard} reads the clipboard when you deliberately expand that snippet, independently of the switches above

Privacy

Private by architecture, not by promise

Privacy isn’t a policy checkbox in TypeSeer — it’s how the app is built. There is no server that could see your writing, because prediction happens on your Mac.

• Everything you type stays on your device — inference is 100% local
• No suggestions are ever generated in password or secure fields
• Network is used for three things, none of which carries your writing: downloading the AI model you choose, a once-a-day check for a new version of TypeSeer, and verifying your license key
• The update check sends only the version you’re on and a random install ID — not derived from your Mac, your name, or anything else about you. It can be switched off; the Privacy page says how
• After a successful licence check, TypeSeer currently keeps working offline for up to seven days; it validates a stored licence at launch and when you open Settings
• Review Data exposes learned sentences and their understandable evidence; optional website evidence contains only a hostname