Are the benchmark numbers real?

They are 100-question samples run in the app request format: 94% BoolQ, 93% QuALITY, 90% on the hard subset, with 156 of 157 correct at 90%+ confidence; the author flags possible training exposure.

Does loading a file execute code inside it?

No. It only reads text from accepted types, refuses executables and oversized or malicious shapes, disables PDF code, and renders names and content as plain text.

Where do my API key and document text actually go?

Your own Vercel AI Gateway key stored only on device; file text goes to Vercel for inference, OCR stays local, and the app shows reported cost and credits with a pre-charge stop.

How do the Windows and Android versions differ?

Same interface on both; Windows uses Node 22 plus Edge and Windows OCR, while Android talks to Vercel directly with bundled ML Kit OCR and no PC needed.

How does it handle long documents?

Jev takes about 32,000 tokens per call, so the app splits long documents automatically and merges the answers, handling up to 400,000 characters; part answers are averaged for choice and scale and maxed for yes or no.

Which file types and limits apply?

Text, PDF, Word docx, and images; scanned PDFs and photos use on-device OCR. Limits are 100 MB per file, 1,000 PDF pages, and 400,000 characters, and unsupported types are refused before reading.

Where does the shown cost come from?

The app shows the actual cost reported by Vercel and your remaining free credits and stops sending before you would be charged; Jev's list price at the time was $0.042 per million input tokens with free output.

CommunityJev Demos & Experiments

Jev Chat — Manta-Boardgame

Windows and Android chat app for Jev via Vercel AI Gateway that answers document questions with per-option probabilities and costs.

See highlights

Open github.com

Stars: 0
Language: HTML
Published by: Community
Author: Manta-Boardgame
Last verified: Oct 2026

Manta-Boardgame/jev-chat

User repository on GitHub

View on GitHub

Ask questions about your documents and get answers with probabilities. Windows & Android app for Jev (TypeSafe AI) via Vercel AI Gateway — reads whole books, PDFs, Word and images, with on-device OCR. Bring your own key.

Forks: 0
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

This is one of several Jev listings for the same project. The canonical entry is Jev Chat — w3cj.

About this project

Drop in a document, ask questions, get answers with probabilities — on Windows and Android. Jev Chat is a chat-style app for Jev, TypeSafe AI's typed decision model, running through Vercel AI Gateway. Unlike a chatbot, Jev doesn't write paragraphs. It gives you a decision — and tells you how sure it is.

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

Tagscommunity github

Works forJev for Classification

Used inJev for Software & DevTools

Highlights

Asks many questions per line in one pass with full probability distributions instead of generated paragraphs.
Splits long documents automatically up to large limits with chunked averaging and max strategies.
Reads text, PDF, Word, and images with local OCR plus explicit size, page, and type safeguards.
Surfaced Vercel cost, remaining credits, and pre-charge stop with your own key and budget limit.

Watch out

Benchmark figures are indicative on public datasets with 100 samples each and possible training exposure; reproduce via benchmarks/ before relying on them. 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

TypeSafeJevPlayground

Unofficial .NET port of the JavaScript SDK covering SystemOne noul, choice and score questions with env-based config, retries and typed errors.