API
Everything the demos do is one HTTP call. Send Laya a state and typed questions;
it answers with a choice, a score or a yes/no probability. It runs on an Android phone in the Acurast Cloud.
Quick start
Question types
| type | you send | you get | good for |
|---|---|---|---|
choice | "criteria": {"id": "description", ...} | choice + probabilities | routing, classification, picking a move |
score | "criteria": ["level 0", "level 1", ...] | score (0 to n-1) + probabilities | ratings, severity, sentiment |
noul | just instructions | noul: probability of yes (0 to 1) | yes/no checks, flags |
Several questions in one request are answered in one pass. Refer to fields of state with backticks in
instructions, like `message`.
Try it
The response appears here.
Other endpoints
GET /health: status and loaded model, no auth.GET /feed?url=<rss url>: fetches an RSS/Atom feed on the phone (same auth). Used by the Newsroom.GET /llms.txt: this page as plain text, for AI agents and coding assistants.
Good to know
- About 0.2-1 s per question on a phone CPU, depending on the length of the state. Every question re-reads the whole state, so fewer questions per request is faster.
- Context is about 512 tokens and English works best. Keep
stateshort. - Probabilities are not perfectly calibrated. Scores tend to be more stable than yes/no.
- This instance is temporary: the URL changes with every redeploy.
- Model: convaiinnovations/laya (Apache-2.0). Run your own on Acurast.