About Jev Lab
An open playground and course for TypeSafe AI's Jev: send a state and typed questions, read the probabilities, turn them into policy.
Where a run goes
No sign-up, no key to paste. Lyzr pays, so every run passes our guard rails first.
History and progress stay here
Rate limits · daily $ cap · never logs your text
Jev answers
Only when you compare
- runs / minute
- 30runs / minute, per network
- runs / day
- 300runs / day, per network
- comparisons / minute
- 10comparisons / minute, per network
- comparisons / day
- 5comparisons / day, per network
When the day’s budget runs out nothing breaks: unedited examples show their recorded answer, and live runs return at 00:00 UTC. Want no limits? Get your own key at console.typesafe.ai — at $0.042 per million input tokens, a session costs a fraction of a cent.
Every number is one of two kinds
A single run, just now
From our server in local, network included, with the versioned model id that answered.
A response we recorded
Shown with its source, model id and date. Some come from TypeSafe’s small cookbook demo sets.
Neither is a benchmark. They show the shape of a result; TypeSafe’s own claims are quoted as theirs.
Not affiliated with TypeSafe AI. Their documentation is the authority — where we differ, the docs are right. And not a place for real data: see privacy.
How it’s builtshowhide
- Next.js on Vercel. One Zod schema mirrors the TypeSafe contract for both the editor and the proxy.
- Rate limits and the spend cap run on Upstash Redis. Spend is reserved from an estimate, then settled against the real bill.
- The error fixtures in content/recorded/errors are real API rejections captured on purpose, so the linter knows which limits are enforced.
Built and paid for by
Start somewhere
Numbers here are single runs (live) or recorded from docs.typesafe.ai with model id and date. They are not benchmarks.