1,000 characters · ≈89% of everyday text
Most apps teach you Chinese words. Kezi teaches the characters those words are built from — the 1,000 most frequent, with real depth on each — and then proves you can read, on text you choose rather than text we chose.
刻字 · kè zì — to carve characters
This is the app's Reading check screen, running on the same 1,000-character dataset. There is no separate web demo.
Reading check
A menu you photographed, a message, a headline. Kezi inks what it teaches you, fades what falls outside the 1,000, and gives you the number.
≈0% already readable on the first 100 characters — about a week
Tap any inked character for its pinyin, meaning and Hán-Việt reading.
This is why depth on a small set beats breadth on a long word list: the set stands still while the word count grows on its own.
The compounding
Chinese words are assembled from a small set of reused parts. Learn 电, learn 脑, and 电脑 — computer — costs you nothing extra.
电+脑=电脑
electricbraincomputer
Estimated against Jun Da's modern Chinese corpus, so the numbers always carry a ≈. The app renders them the same way.
Coverage
Character frequency is brutally skewed: a handful of characters carry most of everything ever written. That is why Kezi stops at 1,000 instead of chasing a bigger number for the box.
Each piece is verified mechanically against every frequency band — not written and then assumed to be readable.
Reading you can finish
Not "a paragraph". A noodle shop board, a train timetable, a note taped to a door. 58 real pieces, each written so that the characters you own are enough to read it end to end.
家常面
面点：一天三种
大面加两个面点
果子：都是新到的
加面：可以加两次
人多的日子，得等一会儿
These six lines use nothing outside Kezi's first 200 characters. Translations stay hidden until you ask for them — read first, then check.
The ladder: 9 pieces readable at 100 characters → 18 → 28 → 38 → 49 → all 58 at the full set.
109 confusable groups across 186 characters, each with a note on telling them apart — and a drill aimed at the ones fooling you.
Lookalikes
This is where "I know that one" quietly turns into a wrong answer. Kezi tracks which character you mistake for which, and builds the drill out of your own mix-ups.
Plain, nothing added. White.
One extra line on top. Hundred.
Three lines inside. Self.
Every mnemonic is written for that character, in English and Vietnamese — not generated boilerplate poured into a template.
A page per character
This is one character in Kezi. All 1,000 look like this.
With a stroke-order animation, and write-from-memory practice on this exact grid.
Readings come from Unicode's Unihan database, checked and hand corrected — 15 high-frequency characters were fixed.
If you speak Vietnamese
Around 60% of Vietnamese vocabulary is Sino-Vietnamese. Every character already has a reading you have been saying your whole life, and Kezi puts it on the page. No other major app uses this.
The whole interface, every meaning and mnemonic exists in Vietnamese too — and the two are separate settings, so an English interface can still carry the Hán-Việt layer. Xem bản tiếng Việt.
The pitch only works while it is true. So here is what Kezi does not do, before you find out yourself.
Plainly
Your learning data lives on your device. Cloud backup is optional and off until you turn it on, and you can delete everything from Settings at any time. Privacy policy.
Coming soon
No newsletter, no drip campaign. One message, the day Kezi is on the App Store and Google Play.
iOS & Android · English and Vietnamese · No ads