Sage Console | Decisions you can act on

Yes/No, Choice, Scale, and Tags accept an image as content, on /decide and on each /decide/batch group. This is a beta and the first step towards a multimodal Sage.

Request

{
  "content": {
    "kind": "image",
    "media": "data:image/png;base64,iVBORw0KGgo...",
    "text": "Ticket: the checkout button is missing on mobile."
  },
  "question": {
    "id": "shows_bug",
    "kind": "yesno",
    "instructions": "Does the screenshot show the reported problem?"
  }
}

Field	Description
`kind`	`"image"`
`media`	Base64 `data:` URI (`image/png`, `image/jpeg`, or `image/webp`). Remote URLs are not accepted
`text`	Optional. Omit it to judge the image alone, or set it to judge the image together with its context (for example a screenshot plus the ticket that reported it)

Response

Results have the same shape as for text. meta.usage adds image_count and image_tokens.

{
  "id": "shows_bug",
  "kind": "yesno",
  "result": { "answer": "yes", "probability": 0.93 },
  "meta": {
    "model": "levanto-sage-v1",
    "latency_ms": 412.0,
    "usage": { "billed_input_tokens": 16, "image_count": 1, "image_tokens": 850 }
  }
}

On image content, Tags always decide: applies is true or false, never null. Yes/No and Choice can still answer null when Sage isn’t sure.

Supported kinds and limits

Supported for Yes/No, Choice, Scale, and Tags.
At most 4 MiB decoded, longest edge at most 8192 px, width × height at most 4096².
Choice with an image accepts at most 20 options, because each option costs its own pass over the image. Text-only Choice still allows 120.
Not supported with sort, inside a list item, or together with grounding. Those combinations return 400.

Pricing

Each unique image counts as one extra document: one decision unit on top of the question/token count. One question about text plus one image is 2 units; the same image repeated in a request or batch call counts once. See Pricing.