Word guessing, heads up.
Six guesses per word, and the next word is dealt the moment a round ends. Your streak and guess chart survive a full restart. Hold the touch bar, say a word or spell it, let go, and pick what was heard. Recognition runs offline on the phone.
01how it plays
Marks by shapeSolid tile: right spot. Circle: wrong spot. Strike: not in the word. Readable at any brightness.
Keyboard on the glassesSwipe to walk the keys, tap to type. The keys carry the same marks as the board.
Hold to talkSay the word, or spell it with letters or alpha, bravo, charlie. Near misses and homophones are offered too.
Private by designThe mic is on only while you hold. Speech is decoded on the phone and never recorded or sent.
Stats that stickStreak, best, win rate and a guess-distribution chart, kept across restarts.
Fast on the glassesKey focus and the voice card are firmware text, so each step is a ~60 ms update, not an image send.
02run it
npm install npm run voice-model # once: the 41 MB offline speech model npm run dev # Vite on :5173 npm run qr # QR code to sideload onto your phone npm run pack # production build โ wordlens.ehpk
Needs Node, the Even App 2.2.10 or newer and G2 glasses. npm run simulate runs it in the Even Hub simulator instead.
03spec
| platform | Even Realities G2 via the Even App (2.2.10+, SDK 0.0.15+) |
| stack | TypeScript + Vite, vosk-browser for speech |
| words | 2,069 answers, 8,679 allowed guesses (ENABLE + SCOWL, no NYT lists) |
| permissions | Glasses microphone, only while you hold to talk |
| network | None. No camera, no location |
| cost | Free |