Jev for Google Sheets

It is a Google Sheets add-on that classifies, tags and scores text by asking plain-language questions in a formula. Answers come from TypeSafe's Jev model in about a tenth of a second per row as real values: TRUE/FALSE, one of your categories, or a number.

JEV_IF returns TRUE or FALSE from a yes/no question using a threshold that defaults to 0.5; JEV_PROB returns the probability of yes; JEV_CHOICE picks one of 2-255 options; JEV_SCORE maps text to 1 through N across 2-10 ordered levels.

Three ways: the Google Workspace Marketplace listing, which is pending review; downloading release files (jev.js, functions.js, Sidebar.html, appsscript.json) and pasting them into Extensions, Apps Script, which needs no Node.js; or building from source with npm install and npm run build, optionally uploading with clasp.

You need your own TypeSafe API key from console.typesafe.ai/keys. TypeSafe charges per input token ($0.042 per million at launch), so a short cell costs a fraction of a thousandth of a cent, and identical questions on identical text are cached for six hours so recalculation does not re-bill.

Yes. A range like JEV_IF(A2:A500, ...) returns one answer per cell, running requests in parallel batches with automatic retry when rate-limited. A multi-column range like JEV_IF(A2:C500, ...) judges each row as a single item, reading all its cells together.

JEV_CHOICE and JEV_SCORE return UNSURE when confidence is below min_confidence, so uncertain rows say so instead of guessing. Errors appear in the cell as #JEV, and a bad or missing API key makes the whole formula error out.

Each formula handles up to 1,000 cells because Sheets gives custom functions 30 seconds, options are limited to 2-255 and score levels to 2-10, and the add-on requires your own TypeSafe key. The project site is jevsheets.link; the Marketplace listing was still pending at the time of the README.