huggingface.js: a suite of JS/TS libraries for managing Hugging Face repositories and executing model inference

What it solves

This project provides a suite of JavaScript and TypeScript libraries that allow developers to programmatically interact with the Hugging Face ecosystem. It eliminates the need to write raw API calls to manage models, datasets, and spaces, and provides a streamlined way to integrate AI models into web applications, Node.js servers, and other JS runtimes like Bun and Deno.

How it works

The project is organized as a collection of specialized packages:

• @huggingface/hub: Handles the management of repositories (creating, deleting, and uploading/downloading files).
• @huggingface/inference: Provides a client to call over 100,000 ML models via serverless Inference Providers or dedicated Inference Endpoints, supporting tasks like chat completion, text-to-image, and translation.
• @huggingface/mcp-client: Implements a Model Context Protocol (MCP) client and a small agent library for tool-using AI agents.
• @huggingface/tiny-agents: A model-agnostic library for building AI agents that can use tools.
• @huggingface/gguf & @huggingface/dduf: Parsers for specific model formats (GGUF and DDUF).
• @huggingface/jinja: A minimalistic Jinja templating engine for ML chat templates.

Who it’s for

JavaScript and TypeScript developers who want to build AI-powered applications using models hosted on Hugging Face without needing to manage themselves the underlying infrastructure or complex API requests.

Highlights

• Multi-runtime support: Works across modern browsers, Node.js (>=18), Bun, and Deno.
• Broad model access: Access to a massive library of models via a unified interface.
• Provider flexibility: Ability to switch between different serverless inference providers (e.g., Sambanova, Together, Fal-ai, Replicate, Cohere).
• Zero-dependency approach: Uses modern JS features to avoid polyfills and heavy dependencies.