Changelog
Every TypeSeer release, newest first
31 releases so far. These are the same notes the app shows under “What’s New”, written for the people who use TypeSeer rather than for a commit log. The current version is 1.0.1, released 2 September 2026.
1.0.1
- The words TypeSeer has learned now finish themselves. Type the first letters of a name or term it has picked up and the rest appears, ready for Tab. These are the words your Mac's dictionary has never heard of, so they were the ones you always had to type out in full. Where two learned words start alike, you get as far as they agree and no further.
- And TypeSeer takes the hint when you finish a word it wanted to make longer. Some short words really are the start of a longer one, so a name like "Sam" kept being offered "Samuel" however often you typed it plain, and dismissing a spelling correction could never fix that because the word was never spelled wrong. Finish the same word yourself a few times while TypeSeer is offering to extend it, and it stops offering. Everything it has learned is listed under Settings → Suggestions, where you can put any word back under correction.
1.0
- TypeSeer is 1.0. Twenty-nine releases in twelve weeks got it here: whole phrases finished by a model running on your own Mac, spelling corrections that learn the words you actually use, snippets, grammar checking (on IT models), per-app and per-website control, and a setup that walks you through all of it.
- TypeSeer now tells you when it cannot read what you are typing in. A few editors, Google Docs above all, paint your text into a picture: there is no text behind it to read and no cursor to sit beside, so nothing can be suggested there and nothing ever could be. TypeSeer used to go silent in those, which looks exactly like an app that has stopped working. Now the menu says "Can't read this editor", the "TypeSeer Enabled" row is marked "can't read here", and the menu-bar icon carries the same bar it shows when you have switched suggestions off, so you can tell at a glance that the page is the reason rather than TypeSeer.
- A new install now starts on the settings TypeSeer is built around, instead of the cautious ones it used to open with: long suggestions, prepared the moment you pause rather than after a gap, with on-device learning and on-screen context already working. Setup still asks before TypeSeer reads or remembers anything beyond the field you are typing in - the boxes are ticked where they used to be empty, and you can untick them there.
- Your own settings are left exactly as they are, including the ones you never touched, so nothing about your typing changes on update. If you would like the new starting point: set "Suggestion length" to Long in Settings → General, "Suggestion speed" to Immediate in Settings → Performance, and turn on "Personalization", "Personalize by website in browsers" and "Surrounding context" in Settings → Context awareness.
0.99.16
- TypeSeer now learns the words it should stop correcting. Your surname, a product name, a piece of jargon: your Mac's dictionary calls all of them misspellings and used to offer a fix every single time. Dismiss that fix with Esc twice, or just keep typing past it a few times, and TypeSeer leaves the word alone from then on. Accepting a correction, or going back to fix the word yourself, tells it the opposite. Settings → Suggestions lists what it has learned and lets you put any word back under correction.
- And it finishes those words for you. Once TypeSeer has learned a word, typing the first few letters offers the rest: "Naka" completes to "Nakamura", "Kubern" to "Kubernetes". These are the words your Mac's dictionary can never help with, so they used to be the ones you always typed out in full. Where two learned words start alike, you get as far as they agree and no further.
0.99.15
- Fixed the grey preview being drawn below a message box that grows as you type, in apps like Claude and Slack. A suggestion too long for the rest of the line carried on underneath the box rather than inside it, because the box only makes room once the text is really typed. The preview now stays on the line you're writing, shortened to fit, and Tab still accepts the whole suggestion. It still wraps onto the next line in boxes with real room below your cursor, like a mail body or a document.
0.99.14
- Turn TypeSeer off on individual websites — your bank, an admin console — from the menu bar or Settings → Apps. One list shared by every browser, and a domain covers everything under it. TypeSeer reads which site a browser is showing only while that list has something in it.
- A bar through the menu-bar icon means TypeSeer is switched off where you are: everywhere, in this app, on this site, or because the trial ran out. The menu's top line says which. A faded icon still means the other thing, that the model is warming up.
- The menu's master switch is now "TypeSeer Enabled" rather than "Suggestions Enabled", and is marked "off on this site" while a rule is holding it back.
0.99.13
- Nimbus is now TypeSeer. Same app, same everything inside it — the old name turned out to be shared with a crowd of unrelated AI products, which made it needlessly hard to find us or tell us apart. The website has moved to typeseer.com and you can reach us at [email protected]. The app in your menu bar, the About panel and every link in the app now use the new name.
- Everything you've built up comes across untouched: your downloaded model, the writing personalization has learned, your snippets, every setting, and your licence. You don't need to re-download the model, grant Accessibility again, or enter your licence a second time.
0.99.12
- Added an Apps section in Settings for giving individual apps their own custom instructions — formal in Mail, casual in Messages — or turning TypeSeer off in them entirely. Apps are managed in a list with the familiar "+" and "−" buttons, where "+" lets you pick any app on your Mac. Your everywhere instructions now live at the top of the same section.
- The menu bar now names this feature: a "custom instructions for <app>" item shows the app you're typing in and whether it already has its own instructions, and opens straight to it. Per-app instructions were previously only explained in TypeSeer Help and easy to miss from the menu.
- Added "Calibrate Suggestion Position" to the menu, for the rare app where the grey preview sits slightly above or below the line you're typing. It lets you nudge the preview into place with the keyboard and send the measurements back through "Send Position Report…", so the alignment can be corrected for that app in a later update.
- TypeSeer no longer offers suggestions inside its own Settings window, and no longer lists itself as an app you can configure.
- Fixed suggestions in Mail appearing too small and in the wrong font. Mail's newer compose area tells TypeSeer nothing about the font it's using, so the grey preview was drawn tiny and a little wider than your own text — which then shifted as you accepted it. It now matches the size and font of what you're writing.
- Fixed the suggestion being drawn over the text below your cursor — the message you're replying to in Mail, or the rest of a paragraph when you're editing partway through one. The grey preview now stays on the line you're typing rather than covering your existing text; press Tab to accept the whole suggestion as before.
0.99.11
- Fixed a burst of dropped keystrokes right after switching to another app, where getting ready to suggest in the newly focused field could briefly monopolize the processor and swallow the first few characters you typed. That preparation now steps aside for your typing.
- Fixed TypeSeer offering far fewer suggestions when you write in a language other than English. On the default model, non-English writing was held to a much stricter confidence bar than English, so many good suggestions never appeared — noticeably in Polish and German. Every language now uses the same sensitivity as English and the confidence level you set, so suggestions come just as readily whatever language you write in.
- Fixed the suggestion sitting slightly too high in Safari on the second line of a message and every line after it, so the grey preview text now lines up with what you've typed on every line.
0.99.10
- Fixed suggestions not appearing in Electron-based apps after pasting a large block of text. TypeSeer was misreading the cursor as sitting just before your most recent typing, and now tracks its true position.
- Fixed suggestions stalling while you type in an Electron-based app, where the next character would show nothing until you deleted one. They now keep up with continuous typing.
- Fixed ghost suggestions sitting slightly too high above the line in Slack.
0.99.9
- Added optional website-aware personalization for browsers. When enabled, TypeSeer prefers writing learned on the current website, then falls back to browser and general writing. It stores only the website hostname — never page paths, searches, account details or page titles — and shows that evidence in Review Data….
0.99.8
- Personalization now learns only from confirmed sentence completions, preventing labels, status values and other app text you never wrote from entering your learned writing. Standalone dates, links, paths and other clearly non-writing values are filtered out too.
- TypeSeer now considers recurrence, recency, the current app and whether text was typed or accepted when using your learned writing. Suggestion feedback is stored privately without retaining rejected text, and saving happens away from the keyboard path.
- Existing personalization data is cleaned up automatically while preserving recoverable backups and retaining ambiguous entries rather than risking legitimate writing.
- Added Review Data… under Settings → Context awareness → Personalization, where you can inspect, export or individually delete learned sentences. Personalization remains local and readable, and turning it off now stops all related collection.
0.99.7
- Added a “Custom instructions” section in Settings, so you can tell TypeSeer how you want it to write. Paste a few sentences in the voice you'd like it to match and it carries that style into its suggestions. You can set a different instruction for any individual app from the “Per-app override” picker, which now lists every app TypeSeer has seen — so you no longer have to type in an app first before you can configure it.
- Inline suggestions now line up with your own text in more apps. In Outlook the suggestion showed in the wrong typeface and too large; in Word and Safari it could sit slightly too big or a few pixels above or below the line. It now matches the font, size and position of what you're typing in Outlook, Word and Safari.
- Fixed suggestions repeating part of a hyphenated word. Typing “re-intr” offered “-introduce”, which went in as “re-intr-introduce”, and “state-of-the-ar” did the same. TypeSeer now finishes the word you're typing after a hyphen — “re-introduce”, “state-of-the-art”, “self-hosted”.
0.99.6
- The default model is now a higher-precision build of the same Gemma base. It's noticeably more accurate at the start of a word — where suggestions were most likely to be wrong — while mid-word completions are unchanged. The download is a little larger (about 3.5 GB) as a result.
- Fixed suggestions flickering in browser-based and Electron apps, where a correct suggestion blinked out and came back on almost every character. They now stay put as you type through them.
- TypeSeer no longer suggests inside a web address, an email address or a file path. Typing “typeseer.com” could draw a repeat of what you'd just typed (“.app”, then “.apple”), and an address halfway through an email could pick up a second “@”. Suggestions — and spelling corrections, which could quietly rewrite part of a link — come back on the space that ends it.
0.99.5
- Fixed the default model failing to download during first-run setup and re-downloads, after its original host removed the file. TypeSeer now fetches the same model from a new location, with no change to how suggestions look or behave.
0.99.4
- Long suggestions no longer get cut off with “…” when you reach the end of a line in a multi-line message or note — they wrap onto the next line so you can read the whole thing before accepting.
- Spelling corrections now underline the misspelled word in orange, so there's always a clear sign a fix is waiting for Tab, and show the corrected word inline on top when it fits on the line.
- Fixed suggestions sometimes disappearing on their own in browser-based and Electron apps, where you had to delete and retype a character to bring them back.
0.99.3
- TypeSeer now uses a base model by default. It completes your text directly instead of following a chat template, so you'll see fewer suggestions but more of them will be what you meant. The download is smaller (about 3 GB), and suggestions appear readily out of the box — tune how eager they are for each length in Settings.
- A base model reads no written style or tone instructions and can't do grammar checking, so the Instructions section is hidden and Grammar checking is greyed out while one is selected. Instruction-tuned models still do both — add one as a custom model; they're no longer in the built-in list.
- Simpler setup: it recommends the model on any Mac, drops the old “how often should TypeSeer suggest?” step, and adds an Advanced option to download a custom model of your choice during setup.
- TypeSeer now always lets the model decide where a word ends, so the two experimental word-boundary options have been removed from Settings.
- Press Esc to dismiss a spelling correction and it stays gone for that word — so when you're deliberately typing something your dictionary doesn't know, the fix won't keep popping back and normal suggestions continue.
0.99.2
- Spelling corrections now show the corrected word inline in orange, reading as part of your text, instead of the old “typo → fix” arrow hint. Press Tab to apply as before.
- Confidence thresholds accept any value: type an exact number such as 0.067 into the field beside each slider and the slider moves to match. The sliders now span the full 0–1 range.
0.99.1
- Double-Tab no longer accepts the whole suggestion by default: each Tab takes one word. If you prefer the old shortcut, turn on “Double-Tab accepts the whole suggestion” in Settings → General.
- Suggestions now return automatically after you stop deleting. TypeSeer stays quiet during a burst of Backspaces, then offers a fresh suggestion after a short pause without waiting for you to type.
- Fixed spelling corrections occasionally leaving the typo's first letter behind in web-based apps, producing results such as “hhouse” or “ddog”.
- Suggestions now work in Zoom Workplace even when its chat composer is hidden from macOS Accessibility, stay aligned across wrapped lines and accepted words, and recover after the composer is cleared. Browser zoom, web editors in dark mode, and Discord styling are more consistent too.
0.99
- Ghost suggestions blend into whatever you're writing: they match the field's font and size, and now show up reliably in light and web-based editors — including Gmail, where they could previously be invisible.
- Fewer misplaced or broken suggestions. In Mail and other web editors TypeSeer only suggests at the end of a line, so ghost text no longer covers text you're editing in the middle of a message; and it no longer inserts broken words like "corr rection" instead of "correction", finishing the word or showing nothing instead.
- More control over what you accept: press Esc to dismiss a suggestion (which also frees Tab to move to the next field), a sentence's closing full stop is offered as its own step (Tab takes the word, one more Tab the stop), and accepting a new word no longer adds a stray extra space.
- Custom base (non-instruction-tuned) models now work properly. They run in continuation mode with confidence settings fitted to them, offer mid-word completions more readily where they're strongest, keep up better as you type, and hide grammar checking, which needs an instruction-tuned model. Add one as a custom model — they aren't in the built-in list.
- Base models now use surrounding context more effectively and finish uncommon names and compound words more reliably.
- Smoother typing: your keystrokes no longer briefly stall while a suggestion is generated, most noticeable with a base model or a non-English dictionary.
0.98
- Settings now has separate Context awareness and Performance tabs, making personalization and context controls easier to find while keeping suggestion speed, streaming and word-completion behaviour together.
- Fixed the experimental model-boundary mode occasionally restarting from the sentence's first word — for example, suggesting “See” after “See if you can complete”.
- Missing "Let the model decide word boundaries" button added back to the menu.
- Fixed a brief pause after switching windows or apps when Immediate and Surrounding context were enabled. Context preparation now runs without blocking the keyboard, so early keystrokes register normally while TypeSeer still prepares the first suggestion in the background.
- Deleting and retyping the same text now restores the suggestion TypeSeer just produced instead of generating it again, making repeated prompts appear immediately.
- Reduced long waits during continuous typing. TypeSeer now groups rapid keystrokes into fewer requests and prioritises the newest text over superseded suggestions and background context preparation.
0.97
- TypeSeer no longer stops suggesting just because the letters at the cursor also happen to be a complete dictionary word. It keeps reading the sentence, so a partial word can grow into the intended longer word or a compatible phrase without waiting for another key.
- Added an experimental option under Advanced: “Let the model decide word boundaries”. Leave it off for the default dictionary-assisted behaviour, or turn it on to skip dictionary autocomplete and let the local model decide from context whether the current word is complete.
- Fixed a capital letter at the start of a name being mistaken for a finished one-letter word — for example, “saw A” no longer produces a suggestion beginning with a second “adam”.
- Fixed streamed suggestions occasionally appearing for a fraction of a second and then disappearing even though no key was pressed.
- Dictionary word completion now uses the preceding sentence to resolve genuinely ambiguous partial words when macOS promotes a contextually better candidate.
- Added per-language “Plain-keyboard spelling” options. For supported languages, you can choose to treat missing diacritics as intentional in spelling corrections and grammar checking while continuing to report other mistakes.
- Balanced now waits for a short pause in continuous typing, and Immediate leaves a tiny window between keys, preventing Accessibility and dictionary work from competing with keystrokes. The Settings window also avoids unnecessary refreshes, reducing background CPU use and typing interference while it is open.
0.96
- Suggestions start sooner as you type. On Balanced — the default — TypeSeer now predicts from your first keystroke instead of waiting for a brief pause, so completions keep up with you without the small delay they had before.
- The "Instant" speed setting is now "Immediate", and does more: with Surrounding context on, it prepares the on-screen conversation the moment you click into a field, so the first suggestion in a reply arrives as quickly as the rest — no more waiting through the first few words.
- Battery saver is now the one speed setting that waits for a pause before suggesting, for the fewest suggestions and the least power.
0.95
- Settings now groups Surrounding context and Clipboard awareness together under a new "Context awareness" heading, so it's clear the two do related things.
- Setup offers to turn on on-screen context as one of its first-run choices — still off unless you pick it, but no longer something you have to discover later.
- Fewer off-task suggestions when both context options are on: with clipboard awareness and on-screen context enabled together, TypeSeer no longer occasionally slips into remarking on what you're replying to instead of finishing your sentence.
- Tidied the menu bar: Check for Updates and What's New now sit together, and Run Setup Again has moved down beside About.
- Fixed the Settings window steadily using CPU — and making typing feel laggy in other apps — when it was left open in the background. It now sits idle unless something actually changes.
0.94
- Fixed the new on-screen context feature occasionally producing a suggestion that talked about the conversation instead of continuing your sentence — for example, remarking on the message you were replying to rather than finishing "Hi John,". Suggestions now stay on task, and the rare off-task one is caught before it reaches you.
0.93
- New: TypeSeer can now use the conversation on screen to make its suggestions fit what you're replying to — the email or chat message you're responding to, and the thread around it. In early testing it noticeably improved suggestions in Mail, Outlook, Safari and Teams.
- It's off by default: TypeSeer offers to turn it on once, and you can change it anytime in Settings. TypeSeer reads only the window you're typing in, never password fields, and nothing ever leaves your Mac.
0.92
- Extended last update's word-completion improvement to every language TypeSeer is tuned for: it no longer offers to finish very common words you've already typed in full — now also in French, Portuguese, Dutch, Swedish, Danish, Czech, Russian, Hungarian, Turkish and Vietnamese.
0.91
- Fewer wrong suggestions when you write in Italian, Spanish, German or another non-English language: TypeSeer is more selective about showing a completion in a language it's less sure of, and no longer fills in the rest of a half-typed word from the dictionary in those languages, where that guess was usually wrong.
- Inline word completion no longer offers to finish very common words you've already typed in full — like "per" or "una" — in Italian, Spanish and German, the way it already did in English.
- Settings now tells you when a language you write in has no spelling dictionary installed on this Mac (so its suggestions and corrections are off), and points you to where to add it.
0.90
- Fixed TypeSeer occasionally "correcting" a correctly spelled word into a look-alike from another language — most often when writing in Italian, Spanish, or German.
- Fixed TypeSeer sometimes suggesting a repeat of the sentence you just finished instead of carrying your idea forward.
0.89
- New guided setup on first launch: TypeSeer recommends a model for your Mac, downloads it once you confirm, walks you through granting Accessibility, and gets you ready for your first suggestion — nothing starts until you say so.
- During setup you choose how often TypeSeer suggests, and whether it personalises from what you write.
- Added "Run Setup Again…" to the menu, to revisit setup any time — it pre-fills your current choices and changes nothing unless you confirm.
- Better completions when you write in Italian, Spanish, German — and other non-English languages: TypeSeer now shows the model examples in the language you're writing in, instead of English ones, so suggestions stay in your language and read more naturally.
- Fixed Spanish opening "¿" and "¡" being dropped from longer suggestions.
0.88
- Improved performance of Spelling Correction for non-English languages
- Added a What's New item to the menu that shows the release notes for each update.
0.87
- Added changelog.
- Added Eager, Balanced and Precise confidence presets so you can trade off how eagerly TypeSeer suggests completions.
- Grammar suggestions now read the rendered text in web fields for more accurate corrections.
Every release is free to try
14 days, every feature, nothing leaves your Mac.