Is the latency claim about the model or about the page?

About 100 ms of model time is the README's figure, for eight typed questions answered in parallel per call. The page itself is TanStack Start on Cloudflare Workers hosted by Whop. One call is triggered per pause in typing rather than per keystroke, so both cost and latency depend on typing rhythm, and no end-to-end measurement taken from a browser is reported.

Why send eight questions when all I need is an emoji?

Because the option lists differ in size and only the largest one is the visible keyboard. The emoji choice runs over 64, 128 or 254 options, switchable on the page, and returns a probability for every key. Alongside it sit a choice over eight primary emotions shown as a distribution, three scores for mood, urgency and energy, and three booleans for sarcasm, joke and whether the writer wants a reply. The vocabulary and those questions live in `src/emoji.ts`.

Which credential do I need, and do I have to use the Whop CLI?

Either one, and `src/jev.ts` chooses between them: it calls TypeSafe directly if `TYPESAFE_AI_API_KEY` is set, otherwise it goes through the Vercel AI Gateway with `AI_GATEWAY_API_KEY`. The README's own path uses the Whop CLI with `whop login` and `whop apps init` from the published template, then `whop apps dev` and `whop apps deploy`. From a clone, `pnpm install` plus a gitignored `.dev.vars` and `pnpm dev` is the alternative.

What does this actually cost to run at volume?

About $0.0002 per keystroke is the stated figure, on the basis that Jev bills input tokens only, and that is the README's number for the shipped interaction rather than a measurement you can audit. Call count scales with pauses in typing rather than with characters, so a long message costs more than a short one. If you want your own number, the call site is a single file and the questions are a literal list you can edit.

Can I change the emoji set or the questions?

Yes. src/emoji.ts holds the emoji vocabulary and the three questions, and the choice runs over 64, 128, or 254 options switched on the page, so a fork can edit the literal list at the call site.

What stack and dependencies does the page use?

TanStack Start on Cloudflare Workers hosted by Whop, with the AI SDK's experimental_evaluate and the @ai-sdk/typesafe-ai provider. There is no UI framework and no Tailwind, just one small stylesheet and the system font.

Where is the API endpoint defined?

src/routes/api/react.ts serves POST /api/react, the call itself lives in src/jev.ts, and the page is src/routes/index.tsx, so the whole integration is four small files.

CommunityJev Tools & CLIs

Emoji Jev

Emoji autocomplete where every pause in typing triggers one Jev call answering eight typed questions in parallel: an emoji choice, an emotion choice, three scores and three booleans, in about 100 ms of model time.

See highlights

Open github.com

Stars: 3
Language: TypeScript
Published by: Community
Author: colinmcdermott
Last verified: Oct 2026

colinmcdermott/emoji-jev

User repository on GitHub

View on GitHub

Emoji autocomplete at the speed of typing. TypeSafe AI Jev on a Whop-hosted TanStack Start app.

Forks: 0
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

Emoji autocomplete at the speed of typing. Live at emoji-jev.whop.site. Type anything and a keyboard of up to 254 emojis lights up as you go. Every pause in typing is one call to Jev, which answers eight typed questions in parallel in about 100 ms of model time: a Choice over 64, 128, or 254 emojis (switchable on the page), with a probability for every key. a Choice over eight primary emotions, shown as a distribution. three Scores: mood, urgency, and energy.

Quoted from the project's README. Read the full README.

Tagscommunity jevbest catalog-import awesomejev

Works forJev for Classification

Used inJev for Software & DevTools

Languages: TypeScript

Highlights

The emoji choice runs over 64, 128 or 254 options, switchable on the page, with a probability returned for every key.
Because Jev bills input tokens only, the README puts a keystroke at roughly $0.0002.
Credentials are interchangeable: `src/jev.ts` calls TypeSafe directly when `TYPESAFE_AI_API_KEY` is set, otherwise the Vercel AI Gateway.
No UI framework and no Tailwind, just TanStack Start on Cloudflare Workers, one small stylesheet and the system font.

Watch out

The cost and latency figures are the README's own for the shipped page: about 100 ms of model time per call, with eight questions answered in parallel, at roughly $0.0002 per keystroke on the basis that Jev bills input tokens only. Call count is driven by how often the visitor pauses typing, so cost scales with typing rhythm rather than with characters typed. The 64, 128 and 254 option counts are switched on the page rather than negotiated per request. Running it yourself needs the Whop CLI, a Whop login and a key from either the Vercel AI Gateway dashboard or the TypeSafe console; `pnpm dev` alone also needs `.dev.vars`, which is gitignored. The app is hosted by Whop, and it depends on the AI SDK's `experimental_evaluate` provider. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

jev-recall

Python memory filter that judges every stored memory against the current request with one calibrated yes/no per memory in a single Jev request, keeping everything above a threshold.