Hola, España! PolterType is an automatic keyboard layout switcher: it notices when you type in the wrong keyboard layout, switches the layout, and retypes the word. Hola, España! ES

Because the keyboard layout your operating system has active does not match the language you meant to type. The keys send the same scancodes either way, so the OS happily prints “Hola, Espa;a” instead of “Hola, España” — the text only looks like gibberish; it is your words, mapped through the wrong layout. Nothing is broken, and the word can be recovered exactly.

Normally you delete it, switch the layout by hand, and type it again. PolterType removes those three steps: it watches the word you are typing, notices the moment it only makes sense under a different layout, switches the layout for you and retypes the word. You keep typing and the correction happens behind you.

Yes, with honest differences. Windows 10 and 11 works end to end. On Linux, X11 works with zero setup, while Wayland is best-effort: layout switching is supported on Hyprland, KDE Plasma, Cinnamon, the GNOME family, IBus and Fcitx5, and it needs a one-time setup script for evdev permissions. macOS 11+ needs both Accessibility and Input Monitoring permission, and is now tuned on real hardware too, though only on Intel so far. Intel Macs need v0.14.4 or newer: earlier builds shipped an unsigned Intel slice, and macOS grants Accessibility only to signed code, so the app ran and corrected nothing.

Yes. PolterType is free and open source under the MIT licence, with no paid tier, no account and no subscription. Builds are published on the GitHub releases page, and you can compile it from source yourself.

No. It never logs typed text. The word being examined lives in memory only and is discarded as soon as it is done with, and there is no telemetry of any kind. PolterType makes exactly one network call, and it is not about you: once a day it asks GitHub whether a newer version has been released, and downloads it if so. That request carries no account, no identifier and no data — GitHub sees an IP address, the same as any website you visit. You can turn it off with a single checkbox. The optional AI subsystem is off by default, and sending anything to a remote model takes a second, separate opt-in.

Automatically, but never while you are typing. PolterType checks GitHub for new releases once a day, downloads the installer for your platform in the background and verifies its checksum — then waits. The new version is installed when you quit the app, or when you click "Restart to update" in the tray menu. If you would rather do it by hand, one checkbox on the General settings pane turns the whole thing off.

PolterType ships with dictionaries for fifteen languages: English, Ukrainian, Russian, German, Spanish, French, Polish, Czech, Greek, Hebrew, Turkish, Bulgarian, Italian and Portuguese in both the European and Brazilian orthographies. Only the layouts your OS actually has enabled are loaded, so bundling fifteen costs a two-keyboard user nothing at runtime. Layouts live in data rather than in code, so any other language can be added with a single TOML file and a wordlist — no rebuild of the app required.

Only if you set that up yourself, and nothing is configured out of the box. PolterType ships the interface for a language model but no model, no vendor client and no default address — so with the default settings there is no AI in it at all and nothing to send anywhere. If you want one, you point it at a model you run on your own machine or an API you hold the key to. A model running locally needs no network permission, because nothing leaves your computer; anything else needs a separate switch turned on deliberately, and PolterType then sends only the word in question — not the sentence, not the document, and not which languages you have installed.

No, that case is designed for — and the protection does not depend on PolterType recognising your editor. The engine refuses to auto-switch on anything that looks like an identifier: snake_case, camelCase, or a token mixing letters and digits. A plausibility check then leaves real-but-rare words alone, so kubectl and terraform survive. Both run in every application, on every operating system. If you would still rather PolterType kept out of a particular app entirely, name it in the exception list — that list starts empty and holds only what you put in it. One caveat: the exception list has to know which app is in front. That is exact on Windows, Hyprland and X11. On GNOME and KDE PolterType asks the accessibility bus instead, which answers for most graphical applications but not for terminals, and on macOS it cannot ask at all. The identifier guard and the plausibility check protect your code either way.

Yes. Ctrl+Shift+Space pauses and resumes auto-switching, and Ctrl+Shift+Backspace force-switches the most recent word, ignoring every filter. Two defaults differ by platform. On macOS the pause default is Ctrl+Shift+P, because Ctrl+Space and Ctrl+Shift+Space are the system input-source switching shortcuts there, and claiming them would take your layout switching away. On Wayland the force-switch default is Ctrl+Shift+F9, because there the key combination also reaches the app you are typing in, and Ctrl+Backspace would delete the very word you asked to fix. Both hotkeys are rebindable in Settings.

Because the installers are not code-signed yet. Windows SmartScreen and macOS Gatekeeper warn about any unsigned app, regardless of what it does. On Windows choose “More info” and then “Run anyway”; on macOS right-click the app and choose “Open”. Code signing is planned for a later release.