文鉴 AI · AI 文本检测 API 与图片检测 API

# 1. 提交检测任务（提交即扣费，务必携带 Idempotency-Key）
#    幂等键任意 <=64 字符的随机串即可。这里读内核提供的 UUID，不用 uuidgen ——
#    它属于 uuid-runtime 包，最小化安装的 Debian/Ubuntu 上默认没有。
#    is_merge=false 让每个段落独立出置信度（分段检测）；省略或 true 则合并为整体判定
curl -X POST https://your-domain.com/v1/text/classify \
  -H "Authorization: Bearer sk-zq-你的密钥" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: $(cat /proc/sys/kernel/random/uuid)" \
  -d '{"text":"这是一段待检测的文本","is_merge":false}'

# 2. 用返回的 task_id 轮询结果（pending / running / succeeded / failed）
curl https://your-domain.com/v1/tasks/<task_id> \
  -H "Authorization: Bearer sk-zq-你的密钥"

{
  "code": "ok",
  "data": {
    "task_id": "01J8Z9K3M7X2Q4R6T8V0W1Y5BC",
    "provider": "zhuque-text",
    "status": "succeeded",
    "cost": "0.0600",
    "result": {
      "status": "success",
      "softmax_confidence": 0.9987,
      "ratio_confidence": 0.9821,
      "labels_ratio": { "0": 0.35, "1": 0.65 },
      "segment_labels": [
        {
          "text": "前半段人工撰写的内容",
          "label": 0,
          "conf": 0.0121,
          "order": 1,
          "position": [0, 10]
        },
        {
          "text": "后半段由模型生成的续写",
          "label": 1,
          "conf": 0.9976,
          "order": 2,
          "position": [10, 11]
        }
      ]
    },
    "error_code": null,
    "error_msg": null,
    "latency_ms": 1832,
    "created_at": "2026-10-10T15:32:14+08:00",
    "finished_at": "2026-10-10T15:32:16+08:00"
  }
}

segment_labels[].position 是 [start, length]：第二项是段长（按 Unicode 字符计），不是结束下标。