On-device AI Models for JavaScript

npm packages that run models in the browser and in Node, with TypeScript types included.

The JavaScript and TypeScript packages run in Node and the browser, scoped under @desert-ant-labs on npm. The packages run the models through transformers.js, ONNX Runtime Web, or LiteRT.js.

Everything runs locally, so data stays on the device. Install with npm and import with full TypeScript types.

Available models

RedactPII redaction

PII redaction in the browser via LiteRT.js, no server round-trip.

EmoEmoji suggestions

Multilingual emoji suggestions from text, 22 languages, in the browser or Node.

GistContent topic tagging

Content topics from any text across 101 languages, on device via WebAssembly and LiteRT.js.

ShapesShape recognition

Single-stroke shape recognition in WebAssembly with LiteRT.js.

ClearSpeech enhancement

Speech enhancement in the browser through WebAssembly and LiteRT.js.

TongueLanguage identification

Language identification for short text across 84 languages, runs client-side with no inference runtime to load.

EarSpoken language detection

Spoken language identification across 99 languages, in the browser and in Node.

Pricing

Every model is free up to 100k monthly active devices per SDK. Unlimited inference per user.