Meltype drops the half-width key and guesses the script for you

Meltype removes the key everyone in Japan hates

The pitch is stated in one line on the project page: the half-width and full-width key no longer has to be pressed. You type romaji, and Meltype decides whether the word is Japanese or English, switching on its own at that point. The rest of the project exists to make that decision well. The README gives the argument in a blunt way, telling users that the two keys for half-width and full-width are no longer necessary, and that you simply type in romaji while Meltype works out whether the text is Japanese or English and switches on the spot. The audience is Japanese developers and writers who mix scripts all day, in chat, in code, in documents, and who currently pay for that mix with a keystroke per switch. The project is written in C#, published as a Windows 10 and 11 application, licensed under GPL-3.0, and its releases moved quickly through October 2026, with v1.1.0 on 2026-10-08, v1.2.1 on 2026-10-11 and a v1.2.2-pre.1 pre-release on 2026-10-13, with the last push to the repository on 2026-10-15.

Two ways to run it, with different trade-offs

Meltype can take over the input system or stay out of it, and the README describes both. In Meltype IME mode the characters go straight into the field you are typing in and the candidate list appears below the insertion point, which is the behaviour a native Windows IME has. Getting into that mode requires registering Meltype as a Windows IME, and the installer asks for administrator permission for that step. If you decline the prompt, the second mode still works. In Meltype keyboard mode a conversion box appears under the cursor instead, and the text is inserted from that box. The difference matters for two reasons. The keyboard mode needs no administrator rights and no registration, so it is the fallback for a locked down machine, and the trade-off is that the text does not appear in the field as you type. Other input methods remain reachable in both cases: in IME mode you switch with Win and Space, and in keyboard mode holding Ctrl with the half-width and full-width key pauses Meltype temporarily.

The detector is wrong sometimes, and F10 teaches it

Automatic detection is only as good as its training, so the project is explicit about the escape hatches. The FAQ covers the two mistakes people hit. If an English word comes out as kana, or a Japanese word comes out in Latin letters, you correct it with F10 for Latin letters or F6 for hiragana, confirm the word, and the correction is remembered, so that word comes out in Latin from then on. The README states the same idea in its own words, noting that a word corrected to Latin with F10 appears in Latin afterwards, and that the tool adapts to the way you use it. There is a second control, a setting in the tray menu for the strength of automatic detection, which is the knob to reach for when the detector is too eager or too reluctant rather than when one word was wrong. Frequent words can be registered directly in a user dictionary from the same tray menu, with the full instructions in docs/USAGE.md. The project also lists emoji conversion, spelling suggestions and predictive conversion among its extra features.

Code editors and terminals keep ASCII, comments do not

The feature that matters most to the developers in the audience is how Meltype behaves in an editor. In VS Code or a terminal, alphanumeric input is treated as English by default, and Japanese only appears inside comments and strings, where it is what you want. On top of that, the commands an AI agent uses are left alone: slash commands, dollar-prefixed skills and at-prefixed file names are not converted. That is a specific design decision rather than a general claim, and it is the difference between an IME that fights a coding workflow and one that fits into it. The same mixed-script logic applies to ordinary prose. The README's example shows a romaji phrase turning into a sentence that mixes Japanese and Latin script, with the English word inside the Japanese phrase left in Latin letters, and a second example where the words that stay English are exactly the ones that should stay English. A separate line in the features list mentions per-application settings, so different applications can be given different behaviour.

Installing on Windows, and checking the file you downloaded

The Windows install is a setup executable from the releases page. You download the file named Meltype-<version>-setup.exe, run it, follow the prompts, and the tray shows あ when it is ready. During the process Windows asks for administrator permission to register Meltype as an IME, and you choose yes; if you choose no, the Meltype keyboard mode still works as described above. The build is not code signed, so Windows may display a message that the PC has been protected and the documented route is to choose more information and then run it. There is a zip edition too, where you unpack Meltype-<version>-windows.zip and double click Install.cmd, and the installer and the zip edition put the program in the same place, %LOCALAPPDATA%\Programs\Meltype. The Windows build needs Windows 10 or 11 on 64-bit and the Microsoft IME, and the .NET runtime is bundled. Mac and Linux editions exist as preview builds, the Linux one working through IBus or fcitx5, with installation notes in the mac and linux directories of the repository. Because the binary is unsigned, antivirus software may flag it, and the project has a page for that case that says to update the definitions, download again from the official release and report the detection name in an issue if it persists, explicitly advising against disabling protection or excluding the folder. The integrity check is a SHA-256 comparison against the release page.

Get-FileHash .\Meltype-<version>-windows.zip

Removal is either from the tray icon, which offers an uninstall entry, or from the Windows settings, and the installer asks at the end whether the settings and learning data should be deleted too, while the zip edition's Uninstall.cmd removes those as well.

Two network connections, and nothing leaves the machine

The privacy section is specific about what crosses the network. Typed text is not sent, and both the detection and the conversion complete inside the PC. There are only two network connections, the check for automatic updates and the report form that you open yourself. Settings, the learning data and the dictionary are stored on the machine as well. That combination is what makes the learning behaviour tolerable in an input method, since the corrections you make with F10 and F6 and the words you add to the user dictionary are exactly the data an input method accumulates about you. The repository supports that claim with the usual files for a project of this kind: SECURITY.md, CONTRIBUTING.md, CODE_OF_CONDUCT.md and THIRD-PARTY-NOTICES.md, along with dictionaries/, src/, native/, packaging/, tools/ and tests/ directories, a Meltype.sln solution, and the PowerShell scripts Build-Package.ps1, Install-Meltype.ps1 and Uninstall-Meltype.ps1 that handle packaging and installation on the build side.

Where Meltype is the wrong choice

The limits are visible in the project's own documentation. The Windows build is the product, and the Mac and Linux editions are described as preview versions, so a team that works primarily on macOS or Linux is running something the author marks as preliminary. The binary is not code signed, which means a SmartScreen warning on a fresh install and a real chance of an antivirus false positive that only the project's troubleshooting page can resolve; on a managed corporate machine with application allow-listing, that alone can block it. The detection model is also a trade-off rather than a strict improvement. A user who wants to force the script on every character has lost a control that the half-width and full-width key gave them, and the project answers this by keeping other input methods one shortcut away. Finally, the pre-release numbering, with v1.2.2-pre.1 as the most recent tag, says plainly that the release line is still settling, which is a fair thing to weigh before switching a machine you depend on.

Editorial conclusion

Adopt Meltype if you type mixed Japanese and English every day and are tired of the half-width and full-width key, and if you are willing to let the detector guess and correct it when it guesses wrong. Do not adopt it if you need a signed installer on a locked down Windows machine, if the Mac or Linux preview is not acceptable for your workflow, or if you want per character control over script switching. Verify first that the tray shows the あ indicator after installation, that you can still reach Microsoft IME with Win and Space, and that the downloaded archive matches the SHA-256 value published on the release page.

Frequently asked questions

Official sources

1. License: GPL-3.0
2. Project website
3. README
4. Releases
5. yksr-melt/Meltype on GitHub