Jev — Decisions from Text, with Confidence Scores
A decision model that turns messy input into a clear choice, score, or yes/no— for routing, triage, and high-volume classification.
Fast enough for product UX
70–500ms latency. Put intent detection, urgency scoring, and queue routing inline without a spinner.
Typed outputs you can branch on
selected + confidence + full distribution. No parsing prose, no regex, no “as an AI…”.
Built for high-volume classification
Input-only pricing at $0.042 / 1M tokens with free output. Ideal for support, trust & safety, and ops automation.
Where teams use Jev
Short examples of the jobs Jev is built for
How it works
Three steps from unstructured text to a decision your code can use
Pass a state
Any unstructured text: an email, a ticket, a chat log, a DOM snapshot. Keep tone and punctuation — they matter.
Define questions
Choice, Score, or Noul. Short, exclusive options that map to the branch you will write in code.
Read probabilities
Get selected, confidence, and distribution. Threshold low-confidence cases to a human instead of guessing.
Need open-ended generation? Use an LLM. Need a label you can branch on, with a probability? See how Jev differs from GPT →