Sage docs | Levanto

Set top-level reasoning on /decide or /decide/batch. On batch it applies to every question.

Request

{
  "content": "Refund request: order #4411, $180, bought 41 days ago. Policy: refunds within 30 days; store credit up to 60 days for unopened items. Customer says the box is unopened.",
  "question": {
    "id": "refund_ok",
    "kind": "yesno",
    "instructions": "Under the policy, should we issue a cash refund?"
  },
  "reasoning": "on"
}

Modes

`reasoning`	What happens	Use for
`off` (default)	Answers right away	Most calls, tight latency
`auto`	Reasons only when the question needs it	Mixed workloads, long policies
`on`	Always reasons; up to a few seconds	Policy, eligibility, refunds

Response

Each answer’s meta.reasoning reports what happened:

{
  "id": "refund_ok",
  "kind": "yesno",
  "result": { "answer": "no", "probability": 0.06 },
  "meta": {
    "model": "levanto-sage-2",
    "latency_ms": 391.0,
    "reasoning": {
      "ran": true,
      "finished": true,
      "tokens": 19
    }
  }
}

Field	Meaning
`ran`	Sage reasoned before answering
`finished`	`false` if cut off; `null` if Sage didn’t reason
`limited`	Why it was cut off: `cap` (length), `timeout` (6 s) or `budget` (no time to start)
`forced`	When cut off: `true` answered from partial reasoning, `false` gave the quick answer
`tokens`	Thinking tokens, counted as output

Time limit

Reasoning stops at 6 seconds or a length limit; the call still answers. With auto or on, set your client timeout above 10 seconds.

Pricing

Reasoning adds input and output tokens at your plan’s rates; meta.usage shows the totals. See Pricing.