Jev

What it is

Jev is a decision API accessed through two endpoints. The first accepts raw state and typed questions with custom choice options and score rubrics. The second accepts text and a ruleset name, choosing from eight pre-made rulesets tuned on public benchmarks. Both endpoints bill from the same balance and share latency: multiple questions in one call run in parallel and cost less than separate calls.

Responses return a probability distribution over choices or a probability-weighted score, with confidence on each answer. The API accepts a 64k token context window and has no output tokens (pricing is input-only at $0.042 per 1M input tokens). A single measured choice call over three options cost $0.0000144. Rate limits are 250,000 tokens per second and 1,200 requests per minute.

Access requires a key obtained by email or Google sign-in. Free tier includes 5 credits per month (one credit equals 1,000 input tokens); additional credits are available as one-time purchases. Keys can be rotated or deleted by the holder. The API can also accept a TypeSafe key to run calls on a TypeSafe account instead, with TypeSafe billing directly and no credits consumed here.

Quick answers

How is this Jev API access different from calling Jev directly?

The difference is the waitlist, not the capability. Both accept the same raw state and typed questions (choice options and score rubrics). The key advantage here is immediate access without waiting, and the same endpoints run both custom criteria and eight pre-measured rulesets from the same balance.

How much does one Jev decision cost?

A real one-choice call costs $0.0000144, about 20 times less than the $0.0004 per-decision figure commonly circulated, because typical state sizes are hundreds of tokens rather than thousands. Pricing is $0.042 per 1M input tokens with no output charge. Multiple questions in one call cost only 57% more than a single question because the state is sent once.

Why is output free?

Jev does not generate any output tokens. It returns a decision and a probability distribution from a single forward pass, unlike chat models which bill per token written. There is no stream of tokens to meter, so the entire cost sits on the input side.

Can I use my own TypeSafe key instead of credits?

Yes. Save a TypeSafe API key to run calls on your own TypeSafe account instead. TypeSafe will bill you directly and no credits are consumed here. The key stays in your browser and is passed to TypeSafe with each call, never stored on this site's servers.

What happens to my text sent through the API?

Your text is forwarded to TypeSafe to obtain the decision, and the response is returned. The text is not stored. Only usage data (credits, tokens, timing) is kept. Do not send other people's personal data through it.