User repository on GitHub
Browser-only playground for Jev (TypeSafe AI's decision model) via Vercel AI Gateway
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
CommunityJev Demos & Experiments
Jevplayground
Single-file browser testbed that sends state plus boolean, choice and score questions to Jev through the Vercel AI Gateway and shows answers, probabilities, latency, token counts and estimated cost for every run.
- Stars
- 1
- Language
- HTML
- Published by
- Community
- Author
- terryds
- Last verified
- Oct 2026
About this project
A browser-only testbed for Jev, TypeSafe AI's decision model, running through the Vercel AI Gateway. Jev is not a chat model. You give it some state (a ticket, a transcript, a JSON record) and a set of typed questions; it answers all of them in one parallel pass with probabilities attached: Paste your AI Gateway API key, pick a preset (work or fun), press Evaluate. Build boolean / choice / score questions against plain-text or JSON state. See verdicts, probability bars, latency, token usage, and estimated cost.
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Software & DevTools
Highlights
- Three typed question kinds — boolean, choice and ordered score — answered in one parallel pass, with a probability attached to each.
- Reports latency, token usage and estimated cost per run, and flags which verdicts would need a human at your own accept threshold.
- Inspects the raw request and response, or copies the equivalent AI SDK / curl call.
- No backend: the page calls the gateway endpoint directly, and the key stays in memory unless you tick Remember.
Watch out
You paste a Vercel AI Gateway key into a page and it calls the gateway from the browser, so the key is reachable by anything running in that tab; it lives in memory unless Remember moves it to localStorage, and it is never placed in shared images or links. The request omits the SDK's evaluation-model specification-version header because the gateway's CORS preflight rejects it from a browser, so the call is close to the SDK's but not identical to it. Cost figures and the accept threshold are computed in the page, not reported by the model. Locally it is one static file with no build step; the deployment target named in the README is Cloudflare Workers static assets. 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.
There is no server component. The page posts the key straight to the gateway endpoint from the tab, holds it in memory unless you tick Remember, and omits it from shared images and links. That keeps the key off the author's infrastructure, but it also means the key exists inside a page you loaded from someone else's deployment, so treat it as disposable or serve the single static file yourself.
Close, not identical. The page reproduces the HTTP call the gateway provider makes, including the protocol version and auth-method headers, but leaves out the evaluation-model specification-version header because the gateway's CORS preflight will not allow it from a browser and the endpoint works without it. So a snippet copied out of here is a good reference, not a drop-in replacement for the SDK call.
One pass. You supply the state and a set of typed questions, and every question is answered against that same state with probabilities attached, so a boolean and a score about one record cost a single request rather than one each. What the README does not show is how that holds up as the question count climbs; only the per-question cost and latency in the panel speak to it.
No. It is a number you set in the page, and any answer whose probability falls below it is marked as needing human review. Jev returns the probabilities and the threshold is applied locally, so moving it changes which verdicts get flagged without changing a single answer the model gave.
Paste an AI Gateway key, pick a work or fun preset and press Evaluate. You can build boolean, choice and score questions against plain-text or JSON state, see verdicts, probability bars, latency, token usage and estimated cost, set an auto-accept threshold, inspect the raw request and response, copy an AI SDK or curl snippet, and share a result as an image or a link.
boolean takes instructions and optional true and false criteria and returns a probability. choice takes named options with descriptions and returns the choice plus a probability per option. score takes ordered levels from lowest to highest and returns an interpolated score plus a probability per level.
It is a single static file with no build step, so serve the public directory with npx serve public or python3 -m http.server -d public 5173. It deploys to Cloudflare Workers static assets with npx wrangler deploy, using the included wrangler.jsonc.
More Jev resources like this
40GitHub stars
Python decision client exposing Choice, Noul, and Score primitives with isolated order-invariant scoring, a deliberative flywheel that crystallizes hard cases into fast paths, and local or hybrid deployment.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
zhangcy122/OpenJevPython
zhangcy122/OpenJevPython2GitHub stars
A teleprompter of talking points rather than a script: every open point is a Jev Score question, all scored in parallel against what you have said, and a point checks off once it passes a threshold.
Jev Demos & Experiments#community#awesomejev#catalog-import
Community
finetuningsingh/intelliprompterHTML
finetuningsingh/intelliprompterHTML2GitHub stars
A single-file browser client that fills Japanese PDF forms: it sends only OCR label fragments and your item names to Jev, gets a probability back for every field, and acts on the probability by filling, asking, or leaving it blank.
Jev Demos & Experiments#community#awesomejev#catalog-import
Community
takumi-golf/jev-fill-pdfHTML
takumi-golf/jev-fill-pdfHTML0GitHub stars
A benchmark for browser-using agents hosted on Almond: one prompt, five timed events scored by the page, plus Almond-fastloop, a ~200-line Chrome DevTools loop where Jev picks one bounded action per tick.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
eriestra/browser-use-olympicsHTML
eriestra/browser-use-olympicsHTML0GitHub stars
A FastAPI app that turns a text prompt into 16x16 pixel art by asking Jev 256 parallel Choice questions, one per pixel over a 16-colour palette, then refining the grid.
Jev Demos & Experiments#community#jevbest#catalog-import
2GitHub stars
Tetris where Jev picks every move from the seven legal options, but the game pre-computes landings, holes, roughness and a best plan so the model only ever judges outcomes.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
Yasserbhb/Agent-JEV-TetrisHTML
Yasserbhb/Agent-JEV-TetrisHTMLFrom the Jev guides
Original write-ups on what this entry is for.
A game tick is a state plus a legal action list. How Clash Royale, Minecraft, and the earlier Mario, Pokémon, StarCraft, and drone loops ask Jev one step at a time.
Jev Demos & Experiments#gaming#agents#control
Read article
From the Jev community
Posts from builders shipping with Jev right now.