Basics

Three tools whose answers a model cannot guess — so a correct answer proves a real call happened.

Try it live → — load a model once in the demo, then ask one of the questions below.

What it shows

Weather, the time in a city, and exact multiplication. The values are deliberately fixed and checkable: 31C, humid, light haze in Chennai, 1096637 for 4831 × 227. A model that answers correctly had to call the tool; a model that answers anything else guessed. That makes this the example to run first on any new model.

The tools file

Deliberately unguessable values, so a correct answer proves a real call happened.

// Every tool the model can call. Edit freely, then "Apply tools".
// tool(name, description, params, handler)

tool('get_weather', 'Get the current weather for a city. Use this for any weather question.',
  { city: 'string' },
  async ({ city }) => {
    // A real app would fetch() here. Fixed values keep this demo offline and checkable.
    const table = { chennai: '31C, humid, light haze', tokyo: '18C, clear', london: '9C, drizzle' };
    return { city, conditions: table[String(city).toLowerCase()] ?? '24C, partly cloudy' };
  });

tool('get_time', 'Get the current local time in a city. Use this for any question about the time.',
  { city: 'string' },
  async ({ city }) => {
    const zones = { chennai: 'Asia/Kolkata', tokyo: 'Asia/Tokyo', london: 'Europe/London' };
    const tz = zones[String(city).toLowerCase()] ?? 'UTC';
    return { city, timezone: tz, time: new Date().toLocaleTimeString('en-GB', { timeZone: tz }) };
  });

tool('multiply', 'Multiply two numbers exactly. Use this for arithmetic — never calculate it yourself.',
  { a: 'number', b: 'number' },
  async ({ a, b }) => ({ product: Number(a) * Number(b) }));

// Try adding your own — it is live on the next question:
//
// tool('get_stock', 'Look up a stock price by ticker.',
//   { ticker: 'string' },
//   async ({ ticker }) => ({ ticker, price: 431.20 }));

Each tool(name, description, params, handler) is one callable. The file is plain JavaScript, not a module — no imports, no build step, and the demo lets you edit it and apply it live.

Wire it up

import { NexusChat } from 'toolnexus-web';

// loadForTools checks the model can really call a tool before handing it over.
const chat = await NexusChat.loadForTools({ hub: 'onnx-community/Qwen3-0.6B-ONNX' });
await chat.loadTools('./tools.js');   // the file above, saved as tools.js

chat.on('token', (t) => render(t));
chat.on('toolCall', (call, result) => console.log(call.name, result));

const answer = await chat.chat("What's the weather in Chennai?");

Questions to try

• What’s the weather in Chennai?
• What time is it in Tokyo right now?
• What is 4831 multiplied by 227?

What to expect

Measured by the repo’s harness (npm run test:models, real weights, CPU): Qwen3-0.6B and Qwen2.5-0.5B both call all three tools and answer from the result at q4 (3/3 each). Below ~0.5B it stops working — SmolLM2-360M scored 0/3 at every quantization. See Which models actually work.

Going further

Replace the fixed tables with fetch() calls to your own API. The handler can be async and return any JSON; the model reads it back and answers from it.