Skip to main content

Introduction

⚡️ EnergeticAI makes it easy to use open-source AI models in your app.

It's a distribution of TensorFlow.js optimized for serverless functions:

  • Small module size (~3 MB vs. 146 MB - 513 MB for stock TensorFlow.js)
  • Fast cold-start inference (~50 ms vs. 2000+ ms for stock TensorFlow.js)
  • Incredible ease-of-use (pre-trained models for common cases)

It's intended to be a condensed replacement for TensorFlow.js, so you can typically use it with existing code and models.

Getting started​

EnergeticAI is available on NPM.

Install the core package, along with any model libraries you want to use (e.g. embeddings):

npm install @energetic-ai/core @energetic-ai/embeddings @energetic-ai/model-embeddings-en

Here's an example of using the embeddings model:

import { initModel, distance } from "@energetic-ai/embeddings";
import { modelSource } from "@energetic-ai/model-embeddings-en";
(async () => {
  const model = await initModel(modelSource);
  const embeddings = await model.embed(["hello", "world"]);
  console.log(distance(embeddings[0], embeddings[1])));
})();

Once you've given that a spin, head over to our tutorial for an in-depth example of using embeddings to power product recommendations.

Models & libraries​

Currently EnergeticAI offers two pre-trained model libraries out of the box:

We're excited to add more soon!

Design principles​

When designing EnergeticAI, we're guided by the following principles:

  • Serverless first. We optimize for serverless functions, as that's the most constrained environment. If it works there, it'll work anywhere.
  • Minimal configuration. We focus on having a small API surface area that's easy to use with minimal configuration options. For example, we accelerate through WebAssembly, as that's fast and widely available, but don't aim for GPU acceleration.
  • Support incremental adoption. The EnergeticAI core module is a drop-in replacement for @tensorflow/tfjs-core, so you can use it with existing code and models.
  • Business friendly. We use the business-friendly Apache 2.0 license license for EnergeticAI, and select pre-trained models and dependencies that are under similarly business-friendly licenses.

Comparison with other tools​

Tensorflow.js​

TensorFlow.js is a fantastic open-source library, but not well optimized for the contraints of serverless functions:

  • Module size. Serverless functions have tight limits around module size, and TensorFlow.js is a large library on its own (146 MB - 513 MB uncompressed) and frequently not deployable in serverless environments at all without careful attention to packaging.
  • Cold start. Packages like the Universal Sentence Encoder for TensorFlow.js assume you'll able to download the model weights on first use. This adds unacceptable latency to cold starts, which are critical to minimize for serverless functions.
  • Warm start. TensorFlow.js by default uses a CPU backend that's pretty slow. There's alternate backends, but it requires further configuration.

EnergeticAI is an optimized distribution of TensorFlow.js that solves for these problems.

OpenAI​

OpenAI's APIs provide a fantastic proprietary set of AI models. They're easy to use in serverless functions, and the models are more powerful than EnergeticAI's, but come with a few differnet risks:

  • Privacy risk. You'll have OpenAI as a data processor.
  • Financial cost. You'll need to budget for OpenAI spend for your product.
  • Vendor lock-in. Building a business on top of OpenAI means that they have significant leverage over your business. If they decide to raise prices, or change their terms, you'll have to adapt.

Cohere​

Cohere's APIs provide similar functionality to OpenAI's, and comes with similar risk to OpenAI's. At time of writing Cohere adds extra review steps required to go to production, and comes with similar privacy, billing, and vendor lock-in risks as OpenAI.

Staying informed​

Something missing?​

If something is missing, please open an issue or send a pull request. We're excited to work with you to get it added!