Jev

AI/agents · ship Sep 21, 2026

Jev

Typed choice, score, and yes/no decisions for agents and software — 70–500ms, not prose.

Hosted System One model: choice/score/noul questions return calibrated typed values for routing, gates, and scores.

Listed Sep 23, 2026 · Ship Sep 21, 2026

Why on the board

Verified live 2026-09-23. Product Hunt featured 2026-09-21 — brand-new typed-decision API for in-loop micro-judgments, not a household LLM brand.

What it is

Jev is TypeSafe AI’s hosted “System One” model for fast, typed decisions inside software and agents — not a chat model that returns prose. We verified jevtypesafeai.com on 2026-09-23 (HTTP 200): you send messy state plus one or more questions typed as choice (up to 255 labelled options), score (ordered levels), or noul (calibrated yes/no probability), and get structured values with probabilities in roughly 70–500 ms so code can branch without parsing free text.

The site positions it for high-volume micro-judgments: support triage, moderation gates, lead scoring, tool-call guardrails, model routing, and shrinking stale tool context. Access paths include TypeSafe’s API, gateways (OpenRouter / Vercel AI Gateway / Cloudflare), plus open tooling around it (awesome-jev, jev-mcp, LangChain). Batch Analyzer accepts CSV/rows for many decisions in one pass. Output tokens are described as free; input is priced per million tokens on the marketing page.

Who it’s for

Builders wiring agents or backends who keep burning frontier LLM calls on tiny route/score/gate decisions, or who fight brittle regex rules. If every judgment already fits a hand-written if, you do not need Jev. If you need calibrated probabilities your code can threshold on — and latency under half a second — this shape is the fit.

Why we listed it

Fresh public launch: Product Hunt featured day 2026-09-21 (#2 Product of the Day on PH trackers), live site verified 2026-09-23. Category gap — we have agent security layers and feature flags, but not a typed-decision API for in-loop micro-judgments. Early hosted model, not a household LLM brand.

Caveats

Closed hosted model (not open weights). We have not load-tested latency or calibrated the probabilities ourselves; “0 hallucinations by construction” is a vendor framing of the typed output contract, not a guarantee about real-world labels. Pricing figures on the page differ slightly across sections — confirm current rates on the console before budgeting. Evaluate on your own labels before replacing human review on high-stakes gates.

Next step

Try Jev

External product site. We write our own notes — we don’t paste vendor marketing.

Open Jev →

For the maker

Built Jev? Add the listed badge

Drop this in your site footer or README so visitors can read our independent notes on your product. Optional and free — a listing never depends on a backlink.

Jev — listed on sudoai.net
<a href="https://sudoai.net/launches/jev/" target="_blank" rel="noopener"><img src="https://sudoai.net/badges/listed-on-sudoai-light.svg" alt="Jev — listed on sudoai.net" width="162" height="54" /></a>

The badge links to this page. Use it only for products on the board — we don’t hand out rank or “featured” badges.

Notes are original sudoai.net curation for makers. Facts checked against the live product URL on list day; features and pricing change — verify on the vendor site before you buy or integrate.