Jev Knows is a free yes-or-no tool that weighs a should-I question against recorded human experience, for anyone stuck on one decision.
The number is a lean, not the odds it works out
The homepage keeps a labeled example read, and the image on this page is that card. “Should I take the job in Lisbon?” comes back Yes 64% and No 36%, with the line that 64% of the human experience Jev draws on leans yes. The about page says the percentage is how strongly that knowledge points one way for a choice like yours. It is not the chance that things will work out, and asking again can return a slightly different number.
The same example lists six scores: goal fit 78%, downside manageable 61%, easy to reverse 29%, timing is right 58%, enough information 55%, and opportunity cost worth it 67%. The strongest reason to do it is “It fits what you want.” What gives it pause is “It’s hard to undo.”
Ask once, then add a detail
- One decision, framed as yes or no. The English ask box uses “take the job in Lisbon” as the placeholder, with shortcuts such as text them back, go to the party tonight, take the weekend trip, adopt the dog, and learn Rust.
- You can add a detail and have it read again, up to five times. Each read stays stacked, so you can swipe back and compare.
- The questions page publishes thirty everyday decisions with no extra context. On October 28, 2026, “Should I text my ex?” was listed at No 73%, “Should I learn to code?” at Yes 61%, and “Should I start a side hustle?” at No 51%. Those pages say your own situation is what moves the read.
The language menu includes English, Simplified Chinese, Japanese, Spanish, French, German, Korean, and Portuguese. The sample prompt changes with the language. In Simplified Chinese it asks whether to change jobs and move to Shenzhen after the New Year.
A front end for the Jev model, run separately
The about page says Jev Knows runs on Jev, a decision model from TypeSafe AI, because that model returns how likely something is instead of writing an essay. It also says Jev Knows is an independent product and is not affiliated with, sponsored, or endorsed by TypeSafe AI. The model already has its own listing here: Jev.
Quotes on that about page, including posts from Vercel, LangChain, and Simon Willison, are labeled as comments on the model, not on Jev Knows. The product’s own stats page, checked the same day, showed 45 decisions and two reactions, both marked “Knows” and both filed under “other.” From those two reactions the page prints “100% of reactions say Jev knows.” Career, health, travel, buying, and business each showed zero reactions. The disclaimer says user reactions are opinions, not an accuracy benchmark.
What stays in the browser
- The latest 12 questions, each with up to six reads and the details you added, stay in this browser. They are not synced to an account or another device. Clearing site data deletes that local history. It does not take down a link you already shared.
- Sending a question passes the question and context to model providers. The privacy page names Vercel AI Gateway as the primary route and TypeSafe as the fallback, and says the server does not keep the question or the answer unless you share. When enabled, MoleAPI receives the question and result signals for a short association.
- A public share uploads the question, the read, and details from later rounds. A private share leaves those later details out. A card made with “No link” is drawn in the browser and uploads nothing. Shared pages ask search engines not to index them. Anyone with the link can still copy what it shows. Removal requests go to [email protected].
- Records without the question text are kept for quotas and abuse controls, including topic category, time, model version, and hashed tokens. IP addresses are stored as daily keyed hashes. Umami records coarse events such as “asked” or “shared,” not the words you typed. Google Analytics 4 loads for visitors outside the EU, EEA, UK, and Switzerland.
Ten questions a day, and not advice
The terms, updated September 25, 2026, limit anonymous use to 10 decisions per IP per UTC day, with extra burst controls. Requests may be refused, and the limit can change. You must be at least 13, and under 18 only with a parent or guardian. The terms also say not to submit secrets, passwords, payment credentials, or someone else’s confidential information.
The disclaimer calls the service entertainment and reflection. It says a percentage is a model’s leaning, not a validated chance of success, and that output can be incorrect, incomplete, biased, or outdated. It is not medical, mental-health, financial, legal, or emergency advice. NoRank Labs LLC operates the site. The footer that day showed version v0.1.5. The site does not link a code repository.