rev-decision

TypeScript / JavaScript Client for rev: Fast System 1 Decision Engine with Calibrated Probabilities.

rev-decision is the official, zero-dependency TypeScript/JavaScript client for rev. It connects to your local or hosted rev-server to execute typed decisions (choice, noul, score) in a single forward prefill pass with prefix KV-caching.

Installation

npm install rev-decision
# or
pnpm add rev-decision
# or
yarn add rev-decision

Works out of the box in Node.js (18+), Bun, Deno, Next.js, Vite, and all modern browsers.

Quickstart

1. Basic Decision

import { rev } from "rev-decision";

const res = await rev.predict({
  state: "Customer received broken parcel and demands an immediate refund.",
  questions: {
    intent: {
      type: "choice",
      instructions: "Classify primary customer intent",
      criteria: {
        refund: "Customer demands money back",
        support: "General support assistance",
        feedback: "Product feedback",
      },
    },
    urgent: {
      type: "noul",
      instructions: "Does this require urgent human escalation?",
    },
  },
});

console.log(res.answers.intent.choice);      // "refund"
console.log(res.answers.intent.confidence);  // 0.984
console.log(res.answers.urgent.noul);        // 0.96 (probability of true)
console.log(`Evaluated in ${res.latency_ms} ms (cached: ${res.cached})`);

2. Custom Client Configuration

import { RevClient } from "rev-decision";

const client = new RevClient({
  baseUrl: "http://127.0.0.1:8000", // Default or hosted instance
  apiKey: process.env.REV_API_KEY,   // Optional
  timeoutMs: 15000,
});

const info = await client.getInfo();
console.log(info.run, info.device, info.mode);

3. Using Built-in Presets

import { rev, presets } from "rev-decision";

const res = await rev.predict({
  state: "System database connections exhausted on production-us-east-1",
  questions: presets.triageQuestions(),
});

console.log(res.answers);

Question Types

| Question Type | Output Shape | Description | | :--- | :--- | :--- | | choice | { choice, confidence, probabilities } | Multi-class selection over criteria mapping. | | noul | { noul } | Calibrated continuous probability $\in [0, 1]$ ($p(\text{true})$). | | score | { score, legend, probabilities } | Expected ordinal level across graded tiers. |

License

Apache-2.0 © Jaswanth Sanjay