Laya package – overview and setup

The Aeovi.Laya package brings fast, local AI decisions into your Robomotion flows. It talks to Laya, a small AI model that is built for deciding, not for writing. Laya never invents text. It only spreads probabilities over the answers you give it. That makes it predictable, cheap and very fast: a request with several decisions usually takes 20–40 ms on a graphics card.

The four nodes at a glance

Setup

What you need

1. A running Laya server. Laya runs on your own machine or server. Nothing is sent to a cloud service.
2. The Aeovi.Laya package in your Robomotion workspace.
3. A robot that can reach the Laya server over the network (by default on the same machine).

Step 1 – Install and start the Laya server

Install Laya as described in the Laya repository. Then start the HTTP server. To understand German and other languages as well as English, load both models.

Windows (PowerShell)

$env:LAYA_MODELS = "english,multilingual"
laya-serve

Linux / macOS

LAYA_MODELS=english,multilingual laya-serve

By default the server listens on port 8000. Check that it runs by opening this address in a browser:

http://127.0.0.1:8000/health

You should see "status":"ok" and the loaded models.

Step 2 – Add the package to your flow

Open your flow in the Designer, search for Laya in the package panel and drag one of the nodes onto the canvas. The package is added to the flow’s dependencies automatically.

Step 3 – Your first decision

1. Add an Inject node and a Function node with this code:

   msg.text = "You charged my invoice twice, I want my money back!";
   return msg;

2. Connect a Noul node. Keep the default question “Does the customer ask for a refund?”.
3. Connect a Debug node and run the flow.
4. In the debug output you see msg.laya.noul (about 0.9) and msg.laya.route = yes.

📷 Image placeholder: Screenshot of the canvas with Inject → Function → Noul → Debug, and the debug panel showing msg.laya with noul and route.

Settings that every node has

Every input field can be filled from Custom (fixed value), Message (msg.…) or JS (a JavaScript expression), like in the standard Robomotion nodes.

Reading the result

The nodes write Laya’s answer unchanged into Result. Field names are Laya’s own names. Only the Noul node adds one extra field, route, to make branching easy. Fields you will meet in every answer:

To branch, use a standard Switch node after the Laya node, for example:

msg.laya.route == "yes"          // Noul
msg.laya.choice == "Billing"     // Choice

Batch mode

Turn on Batch when you have many records and want to ask every record the same question(s). Instead of a loop that calls Laya once per record, the node sends up to 64 records per request. With 1,000 e-mails and 7 questions each, our demo flow went from 45 ms to 17 ms per e-mail.

In batch mode, Text must be one of these:

A list of records – the result is a list in the same order (msg.laya[0], msg.laya[1] …):

msg.text = ["first text", "second text"];

An object with IDs – the result has the same IDs (msg.laya.T1, msg.laya.T2 …). This is the easiest way to match answers back to your records:

msg.text = {
  "T1": { "subject": "…", "body": "…" },
  "T2": { "subject": "…", "body": "…" }
};

Every entry in Result looks exactly like the result without batch mode. More than 64 records are split into several requests automatically. Empty records are rejected with an error that names the record.

Good questions, good results

Limits

Errors that all nodes can report

Next steps

• Noul – yes/no decisions
• Choice – pick one of several choices
• Score – rate on a scale
• Advanced – many questions in one request