AI decision API · Swiss-hosted · Jev-compatible
Laya Studio reads an email, a patient message or a review and answers your questions about it with a calibrated probability, not prose. In about a tenth of a second, processed in Switzerland.
{"state": "pixel swallow, 6x5 grid, wing raised, red beak", "questions": { "is_laya_logo": {"type": "noul"}, "tone": {"type": "choice", "criteria": ["light", "dark"]}}}{"answers": { "is_laya_logo": {"noul": 0.99}, "tone": {"choice": "light"}}, "usage": {"input_tokens": 58}}~120 ms
per warm request, from Europe
100+
languages, routed automatically
$0.0357
per 1M input tokens, 15% below Jev
Apache-2.0
open model weights
How it works
Send a message, a ticket or any JSON with the questions you care about. Laya reads it once and returns a probability for every answer: no generated text, nothing to parse.
See it decide
Five everyday messages: a refund email, a patient portal message, a product review, an incident ticket and a sales lead. See the questions Laya asks and the real answers it returned, each with how sure it is. Sign up to ask your own; every workspace starts with 6k free tokens.
Hi, I was charged twice for order #4821 ($129.00). Please refund the duplicate charge before Friday, our books close then. This is the second billing mistake this quarter and we're starting to look at other vendors.
Laya’s verdict
reading…
Waiting for every answer…
reading
waiting
waiting
Calibrated
Laya is trained for calibration: across many answers, the ones it scores 0.8 come true about 80% of the time. So you automate above a threshold you choose and send the rest to a person.
Three question types
Most decisions fit one of three shapes. Define them in plain words in the request; Laya returns a probability for every option.
choice
Pick one label: intent, category, the team to route to.
What does the customer want?
returns {"choice": "refund"}
score
A level on your own scale: urgency, sentiment, risk.
How frustrated do they sound?
returns {"score": 1.97} (≈ annoyed)
noul · yes / no
Yes or no, as a probability: a complaint, an adverse event, spam.
Is it urgent?
returns {"noul": 0.71}
Use cases
Hover or tap a card to see the kind of question Laya answers there.
Automate support ticket triage: one API call reads each ticket and returns intent, urgency, frustration and churn risk, with a confidence score per answer.
Route inbound email to the right team automatically and flag spam, phishing and urgency in one call, with probabilities that decide what a person sees.
Detect what users want in chatbots, voice menus and search. Define intents per request, get calibrated probabilities and hand off to a person when unsure.
Moderate posts for toxicity, harassment, threats and spam in one call. Calibrated probabilities decide what is removed, sent to review or published.
Score inbound leads from form fills, emails and CRM notes. Get fit, buying stage and persona in one API call, with calibrated confidence to prioritise sales.
Route each AI agent step to the right tool or model tier with a calibrated decision model: difficulty, domain and tool choice in one fast, typed API call.
Swiss data residency
Laya Studio answers your questions on GPUs located in Switzerland, keeps its account database in Zurich, and never writes the content of a request to disk. Turn on Swiss-only mode and a request is only ever answered on GPUs in Switzerland, never by a failover abroad. On the way it is handled in memory by our Cloudflare API gateway at the edge location nearest the caller, which can be outside Switzerland and stores none of the content.
A Jev alternative
TypeSafe Jev and Laya both answer typed questions without generating text. An existing Jev client works after two changes: the base URL and the key.
Speed · one decision
Not like-for-like: the first is GPU time in the process; the others include the network. Scale 0–300 ms.
Price · per 1M input tokens, same unit
15% less at every message length: both bill the tokens read.
Openness
For developers
One POST, typed questions, calibrated probabilities. Copy a request and run it.
No tokens are generated, so latency is short and predictable: a warm request takes about 120 ms end to end from Europe.
Every answer is a distribution over its options. Threshold on it, route on it, escalate on it.
The model scores the options you wrote. It can be wrong; it cannot invent a label or a sentence.
Language detection routes each request to the English or multilingual checkpoint automatically.
Same /v1/systemone request and answer shape. Migrate by changing a base URL.
$0.0357 per 1M input tokens, against Jev’s $0.042 per 1M input tokens. Same unit, lower price, and failed requests are free.
Read the docs What calibrated probabilities are Jev alternatives compared
Laya Studio products
The same account, API key and credits run every Laya Studio product.
Check an agent’s tool call before it runs: allow, ask or block.
$0.20 per 1k checks
A small model specialised in tool use and agent steps.
In training
Pricing
Slide to your volume. $0.0357 per 1M input tokens — 15% below Jev. 6k free tokens to start, then credits or a monthly plan from $19.
You pay for the input tokens the model reads: 1 credit = 1 input token, at $0.0357 per 1M. Each question reads the text once (one question per request here). Failed requests are free.
Laya Studio · Pay as you go
$9.46/ month
265M input tokens · $0.0000095 per request
Jev at its list price ($0.042 per 1M input tokens): $11 / month · $0.000011 per request. 15% below Jev.
Every new workspace starts with 6k free tokens.
See all plansFAQ
Create a key, paste the curl, and have calibrated answers in under two minutes.