Moderation API

https://api.aiearth.dev/v1
https://api.aiearth.vip/v1

用户输入 -> 审核分类 -> 通过后进入业务逻辑或模型调用 -> 输出审核 -> 返回用户

import json
import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["AIO_API_KEY"],
    base_url=os.environ.get("AIO_BASE_URL", "https://api.aiearth.dev/v1"),
)

moderation_prompt = """
你是内容安全审核器。请判断用户文本是否存在风险。
只输出 JSON，不要输出解释。
字段：
- flagged: boolean，是否需要拦截或人工复核
- categories: object，包含 hate, harassment, self_harm, sexual, violence, illegal, privacy
- scores: object，各类别 0 到 1 的风险分数
- action: string，取值 allow, review, block
- reason: string，简短说明
"""

text = "这是一段需要审核的用户输入。"

response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[
        {"role": "system", "content": moderation_prompt},
        {"role": "user", "content": text},
    ],
    temperature=0,
)

result = json.loads(response.choices[0].message.content)
print(result)

texts = [
    "第一段待审核文本。",
    "第二段待审核文本。",
    "第三段待审核文本。",
]

batch_prompt = """
你是内容安全审核器。请审核用户提供的 JSON 数组。
返回 JSON 数组。每个元素包含 index, flagged, categories, scores, action, reason。
action 只能是 allow, review, block。
"""

response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[
        {"role": "system", "content": batch_prompt},
        {"role": "user", "content": json.dumps(texts, ensure_ascii=False)},
    ],
    temperature=0,
)

results = json.loads(response.choices[0].message.content)
for item in results:
    print(item["index"], item["action"], item["reason"])

{
  "flagged": true,
  "categories": {
    "hate": false,
    "harassment": true,
    "self_harm": false,
    "sexual": false,
    "violence": false,
    "illegal": false,
    "privacy": false
  },
  "scores": {
    "hate": 0.02,
    "harassment": 0.86,
    "self_harm": 0.01,
    "sexual": 0.0,
    "violence": 0.12,
    "illegal": 0.0,
    "privacy": 0.0
  },
  "action": "review",
  "reason": "包含较强攻击性表达，建议人工复核。"
}

POST https://api.aiearth.dev/v1/moderations

from openai import OpenAI

client = OpenAI(
    api_key="sk-your-api-key",
    base_url="https://api.aiearth.dev/v1",
)

response = client.moderations.create(
    model="omni-moderation-latest",
    input="这是一段需要审核的文本内容。",
)

result = response.results[0]
print(result.flagged)
print(result.categories)
print(result.category_scores)

def moderate_text(text):
    response = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[
            {"role": "system", "content": moderation_prompt},
            {"role": "user", "content": text},
        ],
        temperature=0,
    )
    return json.loads(response.choices[0].message.content)

def handle_user_message(text):
    result = moderate_text(text)
    if result["action"] == "block":
        return "这条内容无法处理，请修改后再提交。"
    if result["action"] == "review":
        return "这条内容需要人工复核。"
    return "审核通过，可以进入后续处理。"

def safe_chat(user_text):
    input_result = moderate_text(user_text)
    if input_result["flagged"]:
        return "抱歉，这个请求暂时无法处理。"

    answer = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[{"role": "user", "content": user_text}],
    ).choices[0].message.content

    output_result = moderate_text(answer)
    if output_result["flagged"]:
        return "抱歉，生成结果需要复核，暂时无法展示。"

    return answer

THRESHOLDS = {
    "hate": 0.7,
    "harassment": 0.75,
    "self_harm": 0.4,
    "sexual": 0.8,
    "violence": 0.7,
    "illegal": 0.5,
    "privacy": 0.35,
}

def decide_action(scores):
    high_risk = [key for key, value in scores.items() if value >= THRESHOLDS[key]]
    if high_risk:
        return "block"
    if any(value >= 0.3 for value in scores.values()):
        return "review"
    return "allow"

{
  "request_id": "mod-20260515-001",
  "user_id": "user_123",
  "action": "review",
  "categories": ["harassment"],
  "scores": {
    "harassment": 0.86
  },
  "created_at": "2026-05-15T08:00:00+08:00"
}

这条内容暂时无法发布，请修改后再提交。