Skip to content
browser-llm-nexus

Quickstart

Install

Section titled “Install”
Terminal window
npm install browser-llm-nexus
npm install @huggingface/transformers   # the engine — an injectable peer dependency

@huggingface/transformers and fflate are optional peer dependencies. If you don’t install them, the library loads Transformers.js from a CDN at runtime and only pulls in fflate when you do zip work. Install them when you want a pinned, bundled, offline-safe build — which you usually do.

Load a model

Section titled “Load a model”

The source is always explicit. There is no default host and no /models/ convention:

import { NexusChat } from 'browser-llm-nexus';


const chat = await NexusChat.load({ hub: 'onnx-community/Qwen3-0.6B-ONNX' });


console.log(chat.device);   // 'webgpu' or 'wasm' — whichever this browser can do
console.log(chat.dtype);    // 'fp16' | 'q4' | 'q8' | 'fp32'

Loading weights takes a while the first time. Show progress:

const chat = await NexusChat.load(
  { hub: 'onnx-community/Qwen3-0.6B-ONNX' },
  { onProgress: p => console.log(p.status, p.file, p.progress) },
);

Stream tokens

Section titled “Stream tokens”

Events are hooks, not callbacks — subscribe as many as you like:

chat.on('token', t => output.append(t));
const answer = await chat.chat('Explain WebGPU in two sentences.');

Add a tool

Section titled “Add a tool”

A tool is a name, a description, a parameter shape, and a handler. Parameters take a shorthand — { city: 'string' } expands to a full JSON Schema property:

chat.tool(
  'get_weather',
  'Get current weather for a city',
  { city: 'string' },
  async ({ city }) => (await fetch(`/api/weather?c=${city}`)).json(),
);


chat.on('toolCall', (call, result) => console.log(call.name, result));


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

The loop is automatic: the model emits a call, the library parses it, runs your handler, feeds the result back, and asks again — up to chat.maxRounds (default 4) — until you get a grounded answer. See Tool calling.

Check what it cost

Section titled “Check what it cost”
console.log(chat.metrics.summary());
// { load_ms_avg, tokens_per_second, tool_calls_ok, … }

Next

Section titled “Next”