Learn

Start here

Laya is ConvAI Innovations’ open decision model. The weights and the Python package are theirs. This page is a short on-ramp that links back to the canonical docs.

  1. Read the model card so you pick a checkpoint on purpose.
  2. Install the package from PyPI.
  3. Try the official Space if you want a browser demo before a local install.
  4. Browse the builds when you want a game, a bench, or a laptop port.

A small local call

The router is the path ConvAI documents for mixed languages. Preload keeps the checkpoints in memory so a language change does not reload weights.

pip install laya
from laya import Router

router = Router(preload=True)
state = {"subject": "Duplicate charge on invoice 4411", "body": "Please refund the second charge."}
questions = {
    "department": {
        "type": "choice",
        "instructions": "Which desk should take this?",
        "criteria": {"billing": "invoices and refunds", "technical": "bugs and outages"},
    },
    "refund": {"type": "noul", "instructions": "Does the writer ask for money back?"},
}
print(router.predict(state, questions)["answers"])

The full quickstart, including Hindi routing and the typed-decisions override, is on the model card. Source sits in NandhaKishorM/laya.

If you are not on a GPU server

The gallery’s on-device shelf points at MLX, Core ML, ONNX, and a PyTorch Metal runtime. Those are other people’s ports. Start from their README, not from this page.