Start with what Jev does, make your first typed decision, then learn when to use Choice, Score, or Noul in a real workflow.
Submit a projectJev is TypeSafe AI's first public System One Model. typesafe-ai/jev returns Choice, Score, and Noul answers with probabilities; high confidence can suggest action, while low confidence routes to review.
TypeSafe Jev is a System One model for software judgment tasks. Instead of generating free-form text, it returns typed Choice, Score, or Noul answers to questions you declare, together with probabilities or confidence signals.
[1] TypeSafe introduction ↗An application puts decision evidence in state and defines bounded questions with explicit criteria. Jev returns structured distributions; application code then validates the answer, applies thresholds, and chooses automation, human review, or abstention.
[2] TypeSafe question primitives ↗This is not a testimonial wall for this site. Every card points to a public repository, separates reviewed implementation facts from outcomes we have not independently tested, and links to the evidence.
A Rust Git semantic gate that runs as pre-commit and pre-push hooks. It reads a staged or piped diff and asks two Jev Noul questions about exposed credentials and destructive payloads; it is neither a code generator nor a conventional syntax linter.
README, Rust implementation, and tests at pinned commit d1b6732 were reviewed on 2026-09-26. This site did not run the binary or measure latency. README claims such as about 80 ms and a sub-2 MB static binary are author-reported, not independent benchmarks.
A market-making experiment on Kuru’s MON-USDC book on Monad, with optional per-block Jev buy/sell decisions.
Defaults to a mock model and simulates fills without a wallet key. Live trading, latency and profitability were not verified.
A MuJoCo drone experiment deriving scene features from camera buffers for Jev tactical advice.
The 2.5/50/500 Hz figures describe the author’s simulation architecture. They are not a flight-safety guarantee or validation on a real aircraft. README and integration source reviewed at a fixed commit; not independently run or benchmarked by this site.
Jev is TypeSafe AI's first public System One Model: it returns typed Choice, Score, and Noul answers with probabilities — not free-form prose.
Choice picks one named option; Score places state on an ordered rubric; Noul (boolean) estimates P(true). Confidence comes from the option distribution.
Low confidence → human queue with no default action. Money, refunds, and irreversible side effects stay human-owned — suggestions only.
Pick one named option. Returns the key, full probability map, and confidence.
Place state on an ordered rubric. A level index — not an accuracy promise.
Keep act / confirm / human bands in your code. Calibrate on your labeled data.
Jev is TypeSafe AI's first public System One Model. It returns typed Choice, Score, and Noul answers with probabilities instead of free-form text. Vercel AI Gateway is the access layer demonstrated by this site.
If you use the Gateway path demonstrated here, your app or serverless function can call it without an extra relay. That does not make Gateway the only way to access Jev.
No. Templates and examples only suggest routes or labels. Money and refund decisions stay with humans or other systems.