Laya local

What you need

• Python 3.10 or newer.
• A CPU is enough for one checkpoint. Set LAYA_DEVICE=cuda only when this machine has an NVIDIA GPU.
• Disk for the weights. The English checkpoint is about 800 MB. Multilingual is a separate download.

Install and start Laya locally

1. Create a virtual environment and install the server extra: pip install "laya[serve]".
2. Bind the process to 127.0.0.1. The package default listens on every interface at port 8000.
3. Preload one checkpoint. Local names are english, multilingual, and typed-decisions. laya-english is not a checkpoint name.
4. Set LAYA_API_KEY before any other machine calls this server.

python -m venv .venv
.venv\Scripts\python -m pip install "laya[serve]"

$env:LAYA_HOST = "127.0.0.1"
$env:LAYA_PORT = "8000"
$env:LAYA_DEVICE = "cpu"
$env:LAYA_MODELS = "english"
$env:LAYA_PRELOAD = "1"
$env:LAYA_API_KEY = "choose-a-local-secret"
laya-serve

On macOS or Linux, create the same virtual environment, activate .venv/bin, and export those variables instead of using $env:.

Call the local API

Check GET /health, then send state plus named questions. Use model english or multilingual so the router keeps that checkpoint.

curl http://127.0.0.1:8000/health

curl http://127.0.0.1:8000/v1/systemone \
  -H "Authorization: Bearer choose-a-local-secret" \
  -H "Content-Type: application/json" \
  -d '{
    "state": "Payment failed since yesterday.",
    "model": "english",
    "questions": {
      "team": {
        "type": "choice",
        "instructions": "Which team should handle this?",
        "criteria": {
          "Billing": "Payments and invoices",
          "Technical": "Bugs and outages"
        }
      }
    }
  }'

Laya local and the hosted playground

Laya local is the open-weight server on your computer. To try the same question in the browser, open the playground with a Laya model already selected:

• Playground with Laya English
• Playground with Laya Multilingual