Skip to main content
Every branch in an automation is a decision: which team gets this lead, how urgent is this ticket, which topics does this message cover. Branch Pilot turns each of those into a decision endpoint you define without code, test with real payloads, and call from your workflow.

​
What you get back

A typed answer, never free text:
  • Route decisions pick one option among up to 255.
  • Score decisions return a 0–100 score over ordered levels.
  • Classify decisions return every label whose probability reaches its threshold.

​
How it works

Branch Pilot runs on Jev, a System One model by TypeSafe AI. Unlike a language model, Jev does not generate text: it evaluates your options in parallel and returns calibrated probabilities. It cannot produce a malformed answer. When Jev is unavailable, Branch Pilot falls back to Mistral with the same typed schema, and the response tells you which engine answered. Before any model call, personal data in your payload is replaced by neutral tokens. Only the pseudonymized payload is sent to the engines and kept in your history. Read how pseudonymization works.

Quickstart

Your first decision in five minutes, from a template to a curl call.

Decisions

Route, score and classify: what each type expects and returns.

Confidence

Thresholds, the uncertain branch and the human review queue.

API reference

Authentication, request and response formats, errors.