TypeSafe (Jev) v1.18.0+

TypeSafe’s System One models (Jev) answer typed questions with calibrated probabilities. They power NodeLLM judgments.

Table of contents

1. Configuration
2. Models
3. Jev-compatible servers
4. Errors
5. Pricing

---

Jev models do not generate text. They take an input and a set of typed questions and return a probability, a choice, or a score for each. NodeLLM uses them for Judgments; chat and the other operations are not available on this provider.

---

Configuration

TYPESAFE_API_KEY=...   # JEV_API_KEY is accepted too

import { NodeLLM, probability } from "@node-llm/core";

// Jev models route to TypeSafe automatically, from any instance.
const judgment = await NodeLLM.judge("The site is down.", {
  questions: { urgent: probability("Does this need attention today?") }
});

Or scope an instance explicitly:

const typesafe = NodeLLM.withProvider("typesafe", {
  typesafeApiKey: process.env.TYPESAFE_API_KEY
});

---

Models

Model
jev-latest	Default. TypeSafe’s current System One model
jev-preview	A preview of the next jev-latest

judgment.model reports the exact model that answered (for example jev-1.13.0). listModels() returns the live catalog.

---

Jev-compatible servers

Any server implementing the System One API (POST /v1/systemone) can stand in for TypeSafe:

const local = NodeLLM.withProvider("typesafe", {
  typesafeApiBase: "http://localhost:8001",
  typesafeApiKey: "local"
});

await local.judge(input, { questions, model: "my-local-jev" });

---

Errors

TypeSafe errors map to NodeLLM’s error classes: an invalid request or unknown model is a BadRequestError, a bad key an AuthenticationError, and a rate limit a RateLimitError carrying retryAfter when the server sends one.

---

Pricing

$0.042 per million input tokens; output is free. judgment.usage reports input_cost, output_cost (always 0) and cost. Models without known pricing, such as one on a local server, report tokens but leave cost undefined.