Isidorus

Isidorus is a high-performance machine learning library for Node.js featuring TensorFlow-backed CPU execution, shared tensor IR, and graph abstractions.

The project is named in honor of St. Isidore of Seville, the patron saint of the internet and computer users, who was known for his efforts to compile and preserve the world's knowledge.

For Production Edge ML

If you're deploying ML inference on edge devices with Node.js and need both:

• Predictable latency
• Stable throughput
• Non-blocking event loop behavior

Then Isidorus handles the hard scheduling problems you shouldn't have to solve yourself.

Project Structure

This is a monorepo managed with npm workspaces and Changesets.

• packages/core: Shared types, Tensor Intermediate Representation (IR), and graph abstractions.
• packages/cpu: TensorFlow CPU graph construction, training, and inference for Node.js.

Getting Started

Prerequisites

• Node.js (v18, v20, v22, or v24 recommended for native compatibility)
• TypeScript ^5.4.0

Installation

npm install

Build

Build all packages in the workspace:

npm run build

Testing

Run tests across all packages:

npm test

Versioning and Publishing

This project uses @changesets/cli for versioning.

• To create a new version: npm run changeset
• To bump versions: npm run version
• To publish to npm: npm run publish-packages

License

Apache-2.0