Decisions API

Choose a decision model

Existing requests keep using Jev. The endpoint, API key, decision types and Jev response shape are unchanged. Add model: "clef" only when you want Clef; omit model or use "jev" for Jev.

Jev: 1 credit per started 1,000 input tokens. Clef: 1 credit per started 100 input tokens. Minimum charge: 1 credit. Actual input usage includes the question and criteria, not just your context.

Jev reserves up to 32 available credits and requires at least 1. Clef dynamically reserves credits estimated from the complete text and image dimensions; the full estimate must be available. Actual input usage determines the charge, capped at the reservation. Unused credits return; iajev covers underestimates. Failed requests are not charged.

Estimate text from the JSON encoding of state and the mapped upstream questions, including all choice labels and descriptions: max(Unicode characters × 2, UTF-8 bytes) + 200 tokens. Add image tokens estimated from 32-pixel grids after a minimum 65,536-pixel upscale, plus a 25% margin and 16 tokens per image. Cap at 65,536 tokens and round up to credits per 100. This is a heuristic, not a tokenizer. GET /api/v1/models exposes the estimation version, parameters and maximum hold; Clef billing now uses reservationMode, maxReservationCredits, estimation and chargeCappedAtReservation instead of fixed reservationCredits/minimumAvailableCredits.

Jev keeps its 8,000-character / 16,000-byte limits. Clef accepts up to 1 MiB of JSON-encoded text and a 65,536-token model window shared by text, images and the question; overlong state may be truncated upstream. Both accept one question and up to 8 choices or score levels. Clef requests may be up to 13 MiB; video is not supported.

Clef images: optional images array of data:image/png;base64,... data URLs (JPEG/WebP also supported), or {content_type, base64} objects. Up to 4 images; each up to 4 MiB and 16 MP, 8 MiB decoded total. Remote URLs are rejected. Images affect idempotency, are not retained in history or saved presets, and must be resubmitted on retries. Clef timeout: 90 seconds; Jev: 15 seconds. The playground loads three fixed iajev CDN examples through an internal helper and resizes them to 512px before submission. API/MCP clients still send embedded images; remote URLs remain unsupported.

GET /api/v1/models uses your existing API key and costs no credits. Read available before calling. MCP clients can use the free list_models tool, then pass model to run_decision. Authentication and scopes stay unchanged.

Use the same idempotency key after an uncertain result. Changing model under the same key returns IDEMPOTENCY_CONFLICT. Completed calls replay their original result and charge even if Clef is later disabled. No automatic model fallback occurs.

Only Clef responses add meta.model and billing.pricingVersion, estimationVersion, estimatedInputTokens, reservedCredits and waivedCredits. waivedCredits is the difference iajev covers if usage exceeds the hold; billing.credits is the actual debit. Jev responses stay unchanged. MODEL_UNAVAILABLE (503) means Clef is disabled.

curl https://iajev.com/api/v1/models -H 'Authorization: Bearer YOUR_API_KEY'

curl https://iajev.com/api/v1/decisions/run \
  -H 'Authorization: Bearer YOUR_API_KEY' \
  -H 'Idempotency-Key: clef-example-001' \
  -H 'Content-Type: application/json' \
  --data '{
  "model": "clef",
  "state": {
    "message": "My payouts have failed for three days. Please help today."
  },
  "images": [
    "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR42mP8/x8AAwMCAO+jRZkAAAAASUVORK5CYII="
  ],
  "decision": {
    "type": "noul",
    "question": "Does this request need urgent attention?"
  }
}'

{
  "model": "clef",
  "state": {
    "message": "My payouts have failed for three days. Please help today."
  },
  "images": [
    "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR42mP8/x8AAwMCAO+jRZkAAAAASUVORK5CYII="
  ],
  "decision": {
    "type": "noul",
    "question": "Does this request need urgent attention?"
  },
  "idempotency_key": "clef-example-001"
}

Boolean probability and scoring examples

{
  "state": {
    "message": "My payouts have failed for three days. Please help today."
  },
  "decision": {
    "type": "noul",
    "question": "Does this request need urgent attention?"
  }
}

{
  "decision": 0.95,
  "confidence": null,
  "results": [
    {
      "value": "true",
      "probability": 0.95
    },
    {
      "value": "false",
      "probability": 0.05
    }
  ],
  "meta": {
    "latencyMs": 218,
    "requestId": "…"
  },
  "billing": {
    "credits": 1,
    "inputTokens": 380,
    "outputTokens": 45
  }
}

{
  "state": {
    "message": "My payouts have failed for three days. Please help today."
  },
  "decision": {
    "type": "score",
    "question": "How urgent is this request?",
    "criteria": [
      "Low urgency",
      "Moderate urgency",
      "High urgency"
    ]
  }
}

{
  "decision": 1.8,
  "confidence": 0.9,
  "results": [
    {
      "value": "Low urgency",
      "probability": 0.05
    },
    {
      "value": "Moderate urgency",
      "probability": 0.1
    },
    {
      "value": "High urgency",
      "probability": 0.85
    }
  ],
  "meta": {
    "latencyMs": 218,
    "requestId": "…"
  },
  "billing": {
    "credits": 1,
    "inputTokens": 380,
    "outputTokens": 45
  }
}