Jev = fast, typed decisions for AI agents and software — choice, score, or yes/no probability, in 70–500ms.
You hand Jev the context and the questions. It hands your code back structured, calibrated values — no prose to parse.
No — Jev is a hosted model from TypeSafe AI, called over an API. But the tooling around it is open: awesome-jev, jev-mcp and the LangChain integration. The model itself runs on TypeSafe's servers.
Three routes: the official TypeSafe waitlist, a gateway (Vercel AI Gateway, OpenRouter or Cloudflare), or a hosted metered key from us with no waitlist. Full walkthrough on the Get Jev page.
Most intelligence shouldn't have to become a chat window. Jev makes the small, fast, high-volume decisions that used to need brittle rules or an overkill LLM call.
Send a ticket, email, event or document to the right place with a single typed choice and calibrated probabilities.
Turn fuzzy judgments — risk, urgency, quality, lead fit — into an ordered score you can threshold on.
Ask a yes/no noul and get a probability, so you can auto-approve the easy cases and escalate the uncertain ones.
The same three primitives cover a huge range of the small, fast judgments software makes all day. A few:
Route & prioritize inbound tickets
Gate user posts before they publish
Score candidates across dimensions
Qualify inbound sales leads
Flag risky pull requests
Catch jailbreaks & prompt injection
Turn free text into typed fields
Read sentiment & intent to leave
if your agent was missingAgents spend most of their time on tiny judgments — is this urgent, is this risky, which model, is this still needed. Too fuzzy for a hand-written if, too small for a frontier LLM. That's Jev's lane.
Grade a request's complexity and send it to a fast, balanced or frontier model — the pattern LangChain and routers like pi-jev-router ship.
Before an agent runs a shell command or edits a file, score the risk and allow, confirm, or block — a fast safety check in the loop.
Decide keep / truncate / drop for each stale tool result so long sessions shrink — verbatim, no lossy summary rewrite.
You describe the state and the questions. Jev evaluates them all in a single round trip and returns typed values with probabilities.
Pick one of up to 255 labelled options. Returns the winning key, per-option probabilities and a confidence.
Place the input on a 2–10 level ordered spectrum. Returns a fractional score plus the full distribution.
A calibrated yes/no as a probability from 0 to 1 — perfect for gates, filters and guardrails.
Paste your own state and questions into the playground — it runs against the real API, free.
Open the playground ▶