What it does
Agents spend LLM calls on small classification jobs: which queue a ticket goes to, whether a message is urgent, which tags apply, how severe something is.POST /v1/decide
answers those without a generation. You put the labels in the request, and no training
run is needed. When you want a model trained on your own text, Fine Tune
builds one.
Four question types
choice, multilabel, bool and score, up to 16 questions per request.A probability for every option
Not just a label: the distribution over options, so you can threshold, abstain
or escalate.
Open weights
The hosted models are the same Apache-2.0 OpenDecision weights published on
Hugging Face. Run them yourself if you prefer.
Not stored
Text sent to
/decide is read by the model and not stored or logged. Only counters
are kept.Next
Quickstart
Your first decision in about two minutes.
Decisions API
Request shape, models, prices and limits.