Small tools that make work lighter.

Nothing to install — a lecturer opens one link, on an iPhone or an Android, and the class list appears with everyone marked present by default. Tap the absentees, press send, done. If the signal drops, the phone keeps the marks and sends them by itself later. The office screen shows today's classes, who has not sent yet, and how many times each student has been absent. The data lives in your own school's Google account — it never reaches VALUES, and there is no monthly fee. This does not replace your LMS; it is the running record a department keeps for itself during the term. (Korean interface.)

Pick 1, 2, 4 or 6 photos per page and drag them into the order you want — the two things paying customers of the leading app have been asking for since 2024. Photos are shrunk on the phone, never uploaded: nothing leaves the device, and the app collects nothing at all. Comes out as PDF, and — the part no other app does — as a Hancom (.hwpx) document, which is what Korean sites are asked for. No ads, no subscription, no account.

Drop in the folder and it works out by itself what to collect — cells inside tables and values written in running text alike (“Reason: …”). Preview the fields, pick the ones you want, rename the Excel headers. Files with a different shape or an empty cell are listed separately instead of being silently wrong. Reads old .hwp, new .hwpx and .docx. No internet; originals are only read.

Hancom's own finder only masks HWP files; it finds but does not mask Excel or Word. Business products make you phone for a price. This one takes a file or a whole folder and one button. Your originals are never changed — a new “name(masked).ext” file comes out beside them. Tables, fonts, conditional formatting, data validation and frozen panes all survive, because the file is edited in place rather than opened and re-saved, and the character count does not change (880101-1******), so column widths and line breaks do not shift. After masking it reads the file again by itself and tells you if anything is still there. It only masks when the checksum matches, so student numbers, amounts and dates are not mistaken for ID numbers. Korean resident and foreigner registration numbers, business numbers, bank, card, phone, email and passport — whichever you tick. Old .hwp is not supported; save as .hwpx. No internet, ever.

Not a spell checker — it looks at the things a word processor never checks: empty recipient or subject cells, a missing closing mark, attachment numbers that skip (1, 3), unfilled blanks like “(   )”, mixed date formats, and whether “N people” in the body matches the rows of the attached list. Every check can be switched off, because every institution's rules differ. Files that are not official letters are skipped in silence. Reads .hwp, .hwpx and .docx.

Every easy tool for this uploads your file to somebody's server and charges per page (about $39 a month); the free ones need Python (Camelot) or a template per layout (Tabula). Drop in a whole folder, see a preview of every table found, tick the ones you want, and get them as one sheet per table, all in one sheet, or one sheet per PDF — each marked with the PDF and page it came from. Your originals are only read, never changed, and nothing leaves your machine. A scanned (photo) PDF has no text, so no table can be pulled out: it tells you that plainly instead of handing you an empty sheet.

Split chairs, rooms or staff so several people can be booked at once; find free times instantly; repeat a weekly class in one go; and prepare tomorrow's reminder texts in a single copy. Customer cards show visits, no-shows and who hasn't been in for a while. Everything stays on your computer — you send the texts yourself.

Excel, Word and Hangul files. Your original file is never touched — the masked copy is a new file, and tables, fonts, conditional formatting, data validation and frozen panes all survive because the file is never opened and re-saved. Character counts stay the same (880101-1******), so column widths and line breaks do not shift. After masking it reads the file back itself and tells you if anything is left. Nothing leaves your computer.

At the hour you choose, it reads overnight mail, strips the ads out and picks the ones asking for an answer, and shortens each to a few lines. It pulls out the deadline, time, place, amount and phone number written in the text, and merges those with your Google Calendar into one Coming up list. It writes draft replies and stops there — nothing is ever sent until you press send.

The summarising happens on your own computer — no AI service, no server, nothing uploaded. Your mail password goes into the computer's own credential lock (Windows credential lock, macOS Keychain), not into a settings file.

Please read before buying. (1) To read your mail you must create an app password yourself in Gmail, Naver Mail or Daum Mail — for Nate Mail you simply use your ordinary login password. There is no "sign in with Google" button. It takes about ten minutes; if that sounds like too much, please do not buy it. (2) It connects to Nate Mail, Naver Mail, Gmail, Daum Mail, Google Calendar and Paddle. It cannot yet connect to a company or school intranet.

Windows, macOS and Linux builds are all included, and the whole program runs in English or Korean.

Feed it the statement file exactly as the bank gave it to you and the ledger you keep in Excel. It works out which line pairs with which — including one ledger line paid out in several transfers, several ledger lines paid in one transfer, dates a few days apart, and amounts that do not agree — then shows you what is left over with no partner at all.

It checks the running balance first. Last line's balance minus money out plus money in should equal this line's balance; when it does not, your downloaded statement is missing rows, and reconciling a statement with missing rows makes innocent lines look wrong. Almost no other reconciliation tool does this check.

It does not memorise the government form. It opens the blank form you downloaded, reads its column headings and fills in underneath. When the form changes next year, it still fits.

It reports nine kinds of problem: statement-only, ledger-only, amount disagreement, outside the project period, entered twice, missing receipt, interest not returned, and spending over the granted amount. Rows worth a second look are shaded yellow, with a separate "VALUES check sheet" tab saying why — you upload only the first tab, so nothing gets in the way.

Everything is computed on your own computer. It works with the network unplugged.

Please read before buying. (0) The program's screen is in Korean only — there is no English interface yet. (1) Windows only — there is no macOS or Linux build yet. (2) It does not yet read card statement files (their columns differ from a bank statement). (3) It cannot yet split one line across several budget categories. (4) It leaves the free-form narrative tabs of the settlement report alone, because there is no fixed government template for them.

Give it your invoice list and your bank statement. Customers never pay the way your books expect: they lump several invoices into one transfer, deduct the transfer fee, pay half now and half later, and pay under a different name. It tries all four.

Exact matches are assigned first. Filling the oldest invoice first makes an exactly-paid invoice land on an older one and read as "paid in part" — that really happened, so the order was changed.

When a deposit arrives under a name you never invoiced and exactly one unpaid invoice matches that amount, it assigns it and tells you "this name was that customer"; write the name into your invoice list and it matches automatically from then on.

You get a three-tab spreadsheet: Outstanding (per customer: invoiced, received, fee, remaining, days overdue — anything past 30 days shaded red), How it was matched, and Unidentified deposits. Everything is computed on your own computer; customer names never leave it.

Please read before buying. (0) The program's screen is in Korean only — there is no English interface yet. (1) Windows only. (2) It does not read card settlement files. (3) It does not handle VAT or tax invoices — receivables only. (4) If the sender name differs and the amount does not match, it leaves the deposit unidentified rather than guessing. (5) No drag-and-drop yet — use the Choose button.

It transcribes the recording on your own computer — the audio never leaves it, and it works with the network unplugged. It also works out who said what from a single microphone: on a ten-minute, four-person meeting it got 40 of 40 turns right and worked out "four people" without being told. Measured against a human transcript, the light model (464MB) scores 88.8% of characters and the accurate one (809MB) 94.6%; ten minutes of audio takes about 4–6 minutes.

It gathers sentences that sound like decisions into their own section, and prints a line asking you to check them — a program should not have the last word on what a committee decided.

Please read before buying. (0) The document it writes is a Korean HWPX file (Hangul word processor) — there is no .docx output yet. (1) Windows only. (2) Those accuracy numbers come from clean recordings; overlapping speech, a distant microphone or a noisy room will be worse, and we have not measured those. (3) The model is downloaded once (464MB or 809MB); after that no internet at all.

No internet, nothing sent to any server. It hears when someone starts talking, works out which of the two spoke, and writes both sides down — no button to press. It also turns video files into .srt subtitles. Ten languages you can hold a conversation in (Korean, English, Japanese, Chinese, Russian, Indonesian, Vietnamese, Hindi, Thai, Tagalog) and five more it only transcribes (Uzbek, Nepali, Mongolian, Kazakh, Urdu) — we measured all twenty against human translations before drawing that line. Of those five, Kazakh and Urdu are transcribed very accurately (3% and 9% character error against human transcripts); Uzbek, Nepali and Mongolian sit at 46%, 31% and 38%. Uzbek is now written in the Latin alphabet Uzbekistan actually uses, not Cyrillic — that change alone halved the error.