Laya, behind a typesafe/jev compatible API

Free, public endpoint for the open-weights Laya decision models (Apache-2.0). It mirrors the request and response format of Cloudflare Workers AI's typesafe/jev: one state, typed noul / choice / score questions in, calibrated probabilities out, in a single forward pass. No account, no API key.

Requests are stored and used to train models. Do not send personal data, secrets or confidential information. By using the API you accept the Terms of Service.

curl https://zaitlabs--laya-jev.eu-west.modal.run/client/v4/accounts/any/ai/run \
  --header 'Content-Type: application/json' \
  --data '{
  "model": "typesafe/jev",
  "input": {
    "state": "Help! My payouts have been failing for 3 days.",
    "questions": {
      "is_urgent": {
        "type": "noul",
        "instructions": "Does this convey urgency?"
      },
      "department": {
        "type": "choice",
        "instructions": "Which team should handle this?",
        "criteria": {
          "billing": "Payments, invoicing, refunds",
          "technical": "Bugs, outages, integrations"
        }
      }
    }
  }
}'

• Point a Cloudflare client at https://zaitlabs--laya-jev.eu-west.modal.run instead of https://api.cloudflare.com; the account id and any bearer token are ignored (and never stored) - except a token issued to a gateway, which only selects that gateway's rate-limit budget.
• "model": typesafe/jev or laya picks a checkpoint by language; or force one with laya/english, laya/multilingual, laya/typed-decisions.
• Context window: 1024 tokens per question. Longer requests are rejected with max_tokens_exceeded, never silently truncated - and, up to 256k tokens, still stored, because they are what a longer-context successor has to learn from. Response headers X-Laya-* tell you which checkpoint answered and why.
• Exact repeats are answered from a cache (X-Laya-Cache: hit) that holds a hash and numbers, none of your text. Hosted in the EU.
• There is an OpenAI-compatible surface too: POST https://zaitlabs--laya-jev.eu-west.modal.run/v1/chat/completions with model convaiinnovations/laya and a response_format of type json_schema, whose boolean, enum and bounded-number fields become the questions. It streams, and it answers nothing else - this model does not write text.
• This is Laya, not Jev - expect different answers. Not affiliated with TypeSafe AI or Cloudflare.

OpenAPI docs · Terms · input schema · output schema · models · health