Laya Node Wrapper

The Node.js wrapper for running the Laya Text Classification Model, the multilingual, non-autoregressive System 1 decision model for AI (for artificial intelligence).

Laya is a non-autoregressive decision model designed for text classification, email triage, moderation and classification.
This package exposes the HTTP API to allow you to easily integrate it into your Node.js applications and environment.
It can be configured to require a secret key and includes a Studio UI you may use to try and validate the schemas and prompts.

In the end, this works as an alternative to Jev.

Features

• Zero-Config Installation: Just follow the instructions during the install process.
• API Key Protection: Built-in authorization token support to secure your endpoint.
• Daemon Mode: Easily run the inference server in the background as a daemon.
• Automated Warmup: Pre-loads the model weights upon startup so your first HTTP request is lightning fast.
• Clean CLI UX: Smooth cli experience.
• Studio: A full featured Studio UI to test your prompts and schemas

Installation

git clone [email protected]:felipenmoura/laya-node.git
cd laya-node

pnpm install

During installation, an interactive prompt will ask you:

1. Which Port the Express server should bind to (default: 4000).
2. Whether to generate or set a custom Secret API Key for route protection.
3. Whether to automatically set up the Python venv environment.
4. Whether to download the Laya model weights immediately.

Your preferences will be saved securely to a .env file (which is automatically git-ignored).

Additional Environment Variables

You can manually edit your .env file to configure additional parameters:

• MAX_LEN: Controls the maximum token length passed to the Laya model during predictions. Defaults to 8192 if not provided.
• PORT: The port for the Express server. Defaults to 4000 if not provided.
• SECRET_API_KEY: The SECRET API key required to access the API.

Usage

Running the Server

To start the server:

npm start
# or
pnpm start

The server will initialize, bind to the configured port, require the correct API Secret for any incomming message, automatically wake up the Laya model in the background (warm up).

It will show you a curl sample on how to call it, too.

Running in the Background (Daemon Mode)

To start the server as a background daemon process:

pnpm start -D
# or
npm run start -- -D

The wrapper will boot up, perform the startup warmup routines, print the success logs to your terminal, and detach.

To shut down the background daemon:

pnpm stop
# or 
npm run stop

Starting the Studio UI

To start the built-in visual Studio UI for testing prompts and configurations:

pnpm start:studio
# or
npm run start:studio

You can also start the studio at the same time with the main service:
pnpm start --studio 4001

Embedding Studio in an iframe (CORS & CSP)

By default, the Studio blocks iframe embedding for security reasons. If you need to embed it in another dashboard, or access its API from another origin, you can supply a comma-separated list of allowed origins. This injects the proper CORS headers and Content-Security-Policy: frame-ancestors directives.

Via CLI flag:

pnpm start:studio 4001 --cors "http://localhost:3000,https://my-dashboard.com"

Or via environment variable:

STUDIO_CORS="http://localhost:3000,https://my-dashboard.com"

Securing the Studio with a WAF

Important Security Notice: The Studio UI automatically injects your SECRET_API_KEY into requests to the main Laya service to make testing seamless. This means that anyone who can access the Studio port can run predictions without knowing the secret key!

If you are exposing the Studio to the internet or embedding it into an external dashboard, it is highly recommended to place it behind a Web Application Firewall (WAF) or a secure reverse proxy (like NGINX, Cloudflare Access, or AWS WAF).

Best practices for shielding the Studio:

1. IP Whitelisting: Restrict access to the Studio port so only authorized internal IPs or dashboard servers can reach it.
2. Basic Authentication: Use a reverse proxy to add HTTP Basic Authentication or Single Sign-On (SSO) in front of the Studio UI.
3. Rate Limiting & WAF Rules: Configure your WAF to block malicious payloads and rate-limit requests to the Studio endpoint to prevent abuse or denial-of-service.

Running with Docker

You can easily containerize this wrapper and run it anywhere using Docker. The included Dockerfile will automatically fetch the model weights during the build phase so container startups remain fast.

To build the image:

docker build -t laya-node-wrapper .

To run the container (exposing port 4000):

docker run -p 4000:4000 -e SECRET_API_KEY="my-secure-key" laya-node-wrapper

Making Predictions

Once running, send a POST request to /predict. Be sure to include your Bearer token if you enabled API Key protection during setup.

Example Request

curl -X POST http://localhost:4000/predict \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <your-secret-api-key>" \
  -d '{
    "state": "The app crashes on startup",
    "questions": {
      "issue": {
        "type": "choice",
        "instructions": "What is the issue?",
        "criteria": {
          "bug": "crash or error",
          "feature": "new request"
        }
      }
    }
  }'

Example Response

{
  "model": "laya-rl-agent",
  "answers": {
    "issue": {
      "type": "choice",
      "choice": "bug",
      "probabilities": {
        "bug": 0.9764,
        "feature": 0.0236
      },
      "confidence": 0.8388,
      "action": {
        "act_probability": 1.0
      }
    }
  },
  "usage": {
    "input_tokens": 28,
    "output_tokens": 0
  },
  "routing": {
    "model": "english",
    "repo": "convaiinnovations/laya",
    "reason": "English Latin text",
    "detection": {
      "script": "latin",
      "script_profile": {
        "latin": 1.0
      },
      "language": "en",
      "is_english": true,
      "non_latin_fraction": 0.0
    },
    "workflow": null
  }
}

Using the laya-node-client

You may want to use the laya-node-client to help you make predictions to your laya-node server.

License

This project is licensed under the MIT License.

Get to know a little more about us:

• BrazilJS channel and conference
• DEx01 company focused on web development, design/ux, research & insights and more.
• Felipe N. Moura