Jev AI Tools

Jev AI Tools is an independent playground for TypeSafe AI’s Jev evaluation model, designed for structured decision-making from your inputs. Instead of producing prose that you later need to parse, Jev returns direct, typed signals your application can use immediately: a category from a named set, a score on an ordered rubric, or a yes-or-no probability.

Work with a prepared customer message example or paste your own input. Run interactive experiments to see how Jev classifies topics, estimates urgency, and determines whether follow-up is needed. The labs also cover decision patterns like startup assumption review, draft tone and clarity scoring (without rewriting), agent next-step simulation, routing tasks, selecting relevant context blocks, and comparing a claim to supplied evidence—without any web search.

A key idea: Jev outputs signals, but your application controls what happens next. Review the returned choice/score/probability values and apply your own thresholds and policies in code—deciding whether to take action, require human review, or do nothing. For boolean questions, the value is a probability of truth, not a binary verdict.

You can explore sample results without signing up. Live runs use anonymous trial credits: 3 trial credits for visitors, and when signed in you get 30 per day up to 300 per month, with 1 credit per 5,000 billed input tokens. You’ll be prompted before any multi-credit run.

What answers does Jev AI Tools generate for my input?

Jev can return (1) a choice from a named set, (2) a score on an ordered rubric (2–10 levels), or (3) a yes/no probability (a number between 0 and 1). These typed outputs are meant to feed directly into your decision logic.

Do I need an account to try Jev AI Tools?

You can try the sample results without an account. Live runs use anonymous trial credits, and signed-in accounts receive additional daily trial credits.

Can Jev rewrite my text when I run the draft-related experiments?

No—draft experiments return scores such as tone and clarity rather than rewriting your content.

How should I use Jev’s probability and score outputs in my app?

Treat them as decision signals and apply your own thresholds in your code. Jev indicates what to consider, while your application decides whether to act, ask for review, or take no action.

Does the claim-vs-evidence experiment search the web for sources?

No. It compares your provided claim and the evidence you supply only, without web searching.

How do trial credits work for live runs?

Visitors receive 3 trial credits for live runs. Signed-in accounts get 30 per day up to 300 per month, and live runs cost 1 credit per 5,000 billed input tokens; the tool will prompt before multi-credit runs.