@askif/jev

Let Jev decide.

The batteries-included askif bundle for TypeSafe's Jev: the jev() backend factory, plus a ready-to-use ask.

npm install askif @askif/jev
export TYPESAFE_API_KEY=...

import { ask } from "@askif/jev";

await ask.if("cat", "is animal", () => {
  console.log("cat is animal");
});

Full documentation, the ask.if/ask.switch/ask.score guide, and runnable examples live in the main repo README.

Other Jev-compatible servers

@askif/jev speaks TypeSafe's Jev wire API, so it also works with OpenJev, an open Jev-compatible server (DiffusionGemma via vLLM or MLX; hosted on Codiv). Only baseURL and model change:

import { jev } from "@askif/jev";

const backend = jev({
  apiKey: process.env.OPENJEV_API_KEY, // not needed by a server you run without auth
  baseURL: "https://api.codiv.ai", // or your own server, e.g. "http://127.0.0.1:8080"
  model: "openjev-latest",
});

• OpenJev uses its own model ids (openjev-latest, openjev-0.1); jev-latest and jev-preview are accepted as aliases, but a pinned Jev id like jev-1.13.0 returns 400 Unknown model.
• Scales take 1–10 levels, matching this backend's limit. Some of OpenJev's other models allow fewer choices than this backend's 255 (Verdict: 24), which the server rejects with a 400.
• Checked live against the hosted Codiv API (2026-09-30): yes/no, choice and scale in one batch (about 0.5–0.7 s), openjev-latest and the jev-latest alias, scales of 2 and 10 levels, and the unknown-model error. Not tested: OpenJev's other models, or a self-hosted server with a real model.
• OpenJev's image and think/samples extensions aren't exposed by askif.

License

MIT