Twelve games. Ten are built on English words, one on numbers and one on history, and the interesting part is how differently they are built. Five — Proximity, Traverse, Quartet, Outlier and Huddle — share a single map of how words relate in meaning, and Ratio does arithmetic on word vectors. Lexis runs a dictionary backwards. Pollen and Perimeter work from curated word lists, Rebus from hand-written clues, Tally from an exhaustive arithmetic solver, and Epoch from fifteen hundred hand-picked moments in history. All twelve are free, work offline, and never ask you to create an account.
Five of the games treat a word as a point in space rather than a string of letters. That single decision does a surprising amount of work. It lets Proximity tell you that tea is the third-closest word to the answer without caring that the two share no letters, lets Traverse accept snow after winter and reject bicycle with a reason, and lets a Quartet grid be tested — before anyone plays it — for any word that secretly belongs to two groups. The map is a word-embedding model shipped inside each app, so nothing you type leaves your phone. Three blog articles go into how it works: how one word map became five games, what “close in meaning” measures, and why every Quartet and Outlier puzzle has one answer.
Ratio takes word vectors somewhere the other five do not: arithmetic. In a good word map the step from man to woman points the same way as the step from king to queen, so an analogy can be solved by taking that step from a new starting word. Its 38,000 puzzles come from hand-written relation templates, and every one was checked against the same vectors that judge your answer — which is also why a word that fits the pattern as well as the intended one is accepted.
Three more word games needed something the map cannot give. Lexis needs precise definitions, so it is built on WordNet, the lexical database Princeton has maintained since the 1980s. Pollen needs a dictionary you will not argue with, so its word list is trimmed hard to words in common use. And Rebus needs clues that a person enjoys solving, which no program can judge — so its three hundred clues were written by hand.
Perimeter is the newest of them. Twelve letters sit on the four sides of a square, every word has to cross the square between letters, and each word starts where the last one ended. Its dictionary of more than 22,000 words is built from ENABLE, a public-domain word list, and each of its 500 puzzles was verified solvable in two to four words before it shipped — the par you are chasing is a real minimum, not a guess.
Tally is not about words at all. Six numbers, a target, and the four basic operations. What it shares with the others is the insistence on fairness: every puzzle is run through an exhaustive solver before it ships, so each one is guaranteed solvable and its difficulty is a measurement rather than a guess.
Epoch leaves words behind too, for history. Each round is five events — inventions, battles, treaties, films, space launches — to put in order on a timeline. Its 1,500-plus events were picked by hand, with data from Wikidata, and difficulty comes from spacing rather than obscurity: easy rounds spread events across centuries, hard ones pack five into a few years.
No accounts, no sign-up, no sync. Your statistics live on your device and nowhere else. No timer in the games that are about thinking, and no streak that punishes you for a busy Tuesday. Advertising keeps them free, and it stays out of the way: never on a board while a clock is running, never on a screen being passed around a table, never in the middle of a puzzle. Where an ad unlocks something, the unlock is permanent and the choice is yours.
EN
FR
ES
DE