Stellarapp

Create and Run AI in the Browser

Train and run AI models in your browser. No code, no debugging, just model design. Everything stays on your device.

Stellarapp guarantees the safety of your training data by never uploading it in the first place. But if you want full control, you can fire up a local version of Stellarapp.

Getting Started

Requirements

• Node.js
• Git

Clone the repository

git clone https://github.com/rkuang9/stellarapp.git

Install all packages

cd stellarapp

npm install

Start the server

# development server
npm run dev

# production server
npm run build && npm start

Open the local web page at http://localhost:3000

Jest Unit Testing

This project uses Jest for unit testing. To run the tests, use

npm test

Playwright E2E Testing

This project uses Playwright for end-to-end testing. Install the extension Playwright Test for VSCode, then the test browsers

npx playwright install

Run the tests via the VSCode testing panel

Creating Custom Models

Stellarapp uses @stellarapp/tfjs-stellar to implement LLMs. You can use it interoperably with @tensorflow/tfjs to create your own GPT-like or Llama-like model class. The web app utilizes web workers to train models and avoid locking the web interface.

1. Create a new web worker and client in the src/lib/webworker folder (recommended to copy and modify gpt_client.ts and gpt_worker.ts)
2. In src/features/training/gpt swap out all instances of GPTModelWorker and its imports with your worker client

To create your own model class, extend the tfs.models.LlmModel class. See the GptModel class as an example (remember to tf.serialization.registerClass your class).