Playground
Paste a state, keep the typed questions, and press Decide — free, no account.
Edit the question schema — 2 typed questions
Decision
Paste a state and press Decide. Your decision appears here — nothing is stored without your consent.
Repeat this exact request
This is the input behind the result above. Run it in a terminal through the same free endpoint and session allowance; no API key is needed. Probabilities can vary between calls.
Run this decision from your own app
The same state-plus-schema contract is a hosted API: a probability for every option, 2,000 decisions a day, and an API key sent by email.
Support ticket → label probabilities → API request
Load the example to restore the fixed billing ticket and its intent/urgency questions. Press Decide, inspect every candidate probability and the fallback label, then copy the exact request shown with your result. This classifies the ticket; it does not issue a refund or close a case.
Cloudflare published 38.8 ms median (about 39 ms) and 122.4 ms p95 across 43 evaluations on 1 October 2026. Those historical benchmark figures are not this public demo’s latency or SLA. The result above measures this browser request, including network and server time; a fallback result is marked separately. Published benchmark
Free tier: 25 decisions per day per browser session. Paid plans raise the limit and issue an API key.
Run each model in its own frame
Two community-hosted playgrounds, one per model size. The frames load lazily so they never slow the rest of the page.
Community-hosted demo of the 9B model. Runs live in your browser session.
Community-hosted demo of the 27B model. Larger, slower, more precise.
Frequently asked questions
Do I need an account?
No. The playground and a daily allowance of free decisions work without signing up.
Where does the answer come from?
The embedded demos run on community-hosted infrastructure; the panel above calls this site’s own decision API. Both use the same state-plus-schema contract.
What happens if the live model is down?
The API answers with a deterministic scorer and marks the response as degraded, so you always get a distribution rather than an error.
Do you store what I paste?
We store the count, the model and the latency — not the state. See the privacy policy for the exact list.
Put it in your stack
The docs page has the request shape, the error list and copy-paste examples in JavaScript and Python.
Model notes
One short email when a new decision model lands or a benchmark changes. No spam, unsubscribe in one click.