Skip to content
MicroResolve

Node.js / TypeScript Quickstart

MicroResolve ships a Node.js package with pre-built native binaries. No build step, no native compiler needed.

Install

Terminal window
npm install microresolve
# or
yarn add microresolve
# or
pnpm add microresolve

Pre-built binaries for Linux, macOS, and Windows (x86_64 and arm64).

Basic usage

import { MicroResolve } from 'microresolve';


const engine = new MicroResolve();
const ns = engine.namespace('support');


ns.addIntent('cancel_order', [
  'cancel my order',
  'I want to cancel',
  'stop my order',
]);
ns.addIntent('track_order', [
  'where is my package',
  'track my order',
  'shipping status',
]);


const matches = ns.resolve('I need to cancel');
for (const { id, score } of matches) {
  console.log(`${id}: ${score.toFixed(2)}`);
}
// cancel_order: 0.91

CommonJS

const { MicroResolve } = require('microresolve');
const engine = new MicroResolve();

Persist to disk

import { MicroResolve } from 'microresolve';


const engine = new MicroResolve({ dataDir: '~/.local/share/microresolve' });
const ns = engine.namespace('support');
// ... add intents and phrases ...
await engine.flush();

Custom threshold per namespace

const security = engine.namespaceWith('security', { defaultThreshold: 1.3 });

Override threshold per call

import type { ResolveOptions } from 'microresolve';


const opts: ResolveOptions = { threshold: 0.3, gap: 1.5 };
const matches = ns.resolveWith('cancel my order and get a refund', opts);

Add phrases at runtime

const result = ns.addPhrase('cancel_order', 'abort my order');
console.log(result.added);      // true
console.log(result.redundant);  // false if new


// Specify language
ns.addPhrase('cancel_order', '取消我的订单', 'zh');

Correct a wrong classification

// "stop shipment" classified as cancel_order, should be track_order
ns.correct('stop shipment', 'cancel_order', 'track_order');


// Teach from a miss (engine returned nothing)
ns.correct('where did my parcel go', null, 'track_order');

TypeScript types

import type { Match, PhraseResult } from 'microresolve';


const matches: Match[] = ns.resolve('cancel my order');
// Match = { id: string; score: number }


const result: PhraseResult = ns.addPhrase('cancel_order', 'abort order');
// PhraseResult = { added: boolean; redundant: boolean; newTerms: string[]; warning?: string }

Vercel AI SDK integration

import { MicroResolve } from 'microresolve';
import { generateText } from 'ai';


const engine = new MicroResolve({ dataDir: '~/.local/share/microresolve' });
const ns = engine.namespace('support');


async function handleMessage(message: string) {
  const matches = ns.resolve(message);


  if (matches.length > 0 && matches[0].score > 0.7) {
    // High confidence — handle directly
    return handleIntent(matches[0].id, message);
  }


  // Low confidence — fall back to LLM
  return generateText({ model: /* ... */, prompt: message });
}

Next