Transformers.js (JavaScript ML)

What it does

Transformers.js enables agents to run state-of-the-art Hugging Face models directly from JavaScript/TypeScript environments without a Python backend. The skill documents the pipeline API, device selection (WASM vs WebGPU), quantization options, model selection guidance, and memory management patterns (including disposing pipelines). It covers a wide range of tasks: text classification, generation, translation, summarization, NER, image classification, object detection, audio transcription, and feature extraction.

When to use it

Use this skill when you need to embed ML inference into browser or Node/Deno/Bun apps, implement client-side model inference, or prototype multimodal features without running separate model servers. It's suitable for building on-device features, demos, or production services that can fetch models from the Hugging Face Hub or use local caches.

What's included

• Scripts: none bundled — has_scripts=false.
• References: extensive references and example guides under references/ (progress callbacks, model registry, configuration).
• Instructions: step-by-step examples for installing, creating pipelines, handling quantization, choosing models, tracking progress, and avoiding memory leaks.

Compatible agents

Useful for agents that can write JavaScript/TypeScript and integrate with web runtimes (Copilot-like code agents, web dev automation agents). Also valuable to operators building browser demos or Node services that need local model inference.