REST - Branch Pilot

One endpoint per decision: POST https://api.branchpilot.ai/v1/decide/{slug}. The full contract is in the API reference.

Request

{
  "input": { "email": "anna@acme.com", "message": "Could we get a quote for 40 seats?" },
  "options": { "version": 3, "engine": "jev" }
}

input — the payload your decision expects: an object matching the declared fields, or free text.
options.version — optional; run this version instead of the live one.
options.engine — optional; "jev" or "llm" to force an engine for this call.

Response

{
  "run_id": "3f0b2c1e-…",
  "decision": "lead-routing",
  "version": 3,
  "status": "ok",
  "choice": "sales",
  "probabilities": { "sales": 0.87, "support": 0.09, "spam": 0.04 },
  "confidence": 0.87,
  "confidence_source": "calibrated",
  "engine": "jev",
  "model": "jev-1.13.0",
  "latency_ms": 212,
  "pii_redacted": 1
}

Field	Present for	Meaning
`status`	all	`ok`, `uncertain` or `pending`
`choice`	route	Selected option key, or `__uncertain__`
`score`, `raw_score`, `legend`	score	0–100 score, raw level position, level labels
`labels`	classify	Retained label keys
`probabilities`	all	Per option, level or label
`confidence`, `confidence_source`	all	See confidence
`engine`, `model`	all	`jev` or `llm_fallback`, and the exact model version
`fallback_reason`	fallback only	Why the primary engine did not answer
`review_url`	pending only	Where a human resolves the run

Timeouts

Give your HTTP client at least 10 seconds: a decision through Jev usually completes under one second, but a fallback after a Jev timeout can take a few seconds.