Perspective API Compatibility

The Perspective API compatibility endpoint accepts the same request format and returns the same response format as the Google Perspective API. With Google sunsetting Perspective on December 31, 2026, this endpoint lets you migrate with a single URL change.

Migration Guide

Migrating from Google Perspective to Sieve takes one change — swap the URL and auth:

import requests

response = requests.post(
    "https://commentanalyzer.googleapis.com/v1alpha1/comments:analyze",
    params={"key": "YOUR_GOOGLE_API_KEY"},
    json={
        "comment": {"text": "You are an idiot"},
        "requestedAttributes": {
            "TOXICITY": {},
            "IDENTITY_ATTACK": {},
            "INSULT": {},
            "THREAT": {}
        }
    },
)

import requests

response = requests.post(
    "https://api.getsieve.dev/v1/perspective/analyze",
    headers={"Authorization": "Bearer mod_live_abc123..."},
    json={
        "comment": {"text": "You are an idiot"},
        "requestedAttributes": {
            "TOXICITY": {},
            "IDENTITY_ATTACK": {},
            "INSULT": {},
            "THREAT": {}
        }
    },
)

const response = await fetch(
  "https://commentanalyzer.googleapis.com/v1alpha1/comments:analyze?key=YOUR_GOOGLE_API_KEY",
  {
    method: "POST",
    headers: { "Content-Type": "application/json" },
    body: JSON.stringify({
      comment: { text: "You are an idiot" },
      requestedAttributes: {
        TOXICITY: {},
        IDENTITY_ATTACK: {},
        INSULT: {},
        THREAT: {},
      },
    }),
  }
);

const response = await fetch(
  "https://api.getsieve.dev/v1/perspective/analyze",
  {
    method: "POST",
    headers: {
      "Authorization": "Bearer mod_live_abc123...",
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      comment: { text: "You are an idiot" },
      requestedAttributes: {
        TOXICITY: {},
        IDENTITY_ATTACK: {},
        INSULT: {},
        THREAT: {},
      },
    }),
  }
);

Request

The request body follows the Perspective API format exactly:

The comment to analyze.

The text content to analyze.

A map of Perspective attribute names to empty objects. Each key requests scoring for that attribute.

Supported attributes: TOXICITY, SEVERE_TOXICITY, IDENTITY_ATTACK, INSULT, PROFANITY, THREAT.

ISO 631-1 language codes. Optional. Defaults to auto-detection.

Response

The response matches the Perspective API format:

{
  "attributeScores": {
    "TOXICITY": {
      "summaryScore": {
        "value": 0.91,
        "type": "PROBABILITY"
      }
    },
    "IDENTITY_ATTACK": {
      "summaryScore": {
        "value": 0.12,
        "type": "PROBABILITY"
      }
    },
    "INSULT": {
      "summaryScore": {
        "value": 0.87,
        "type": "PROBABILITY"
      }
    },
    "THREAT": {
      "summaryScore": {
        "value": 0.08,
        "type": "PROBABILITY"
      }
    }
  },
  "languages": ["en"]
}

Attribute Mapping

Perspective API attributes are mapped to Sieve’s internal categories:

Perspective Attribute	Sieve Category	Description
TOXICITY	toxicity	Rude, disrespectful, or unreasonable language.
SEVERE_TOXICITY	toxicity	Very hateful or aggressive language (higher threshold).
IDENTITY_ATTACK	hate_speech	Attacks based on identity attributes.
INSULT	harassment	Insulting, inflammatory, or negative language toward a person.
PROFANITY	toxicity	Swear words, curse words, and other obscene language.
THREAT	violence	Threats of harm or violence against an individual or group.

Full Example

curl -X POST https://api.getsieve.dev/v1/perspective/analyze \
  -H "Authorization: Bearer mod_live_abc123..." \
  -H "Content-Type: application/json" \
  -d '{
    "comment": {
      "text": "You are an idiot and nobody likes you"
    },
    "requestedAttributes": {
      "TOXICITY": {},
      "INSULT": {},
      "IDENTITY_ATTACK": {},
      "THREAT": {}
    }
  }'

const response = await fetch("https://api.getsieve.dev/v1/perspective/analyze", {
  method: "POST",
  headers: {
    "Authorization": "Bearer mod_live_abc123...",
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    comment: { text: "You are an idiot and nobody likes you" },
    requestedAttributes: {
      TOXICITY: {},
      INSULT: {},
      IDENTITY_ATTACK: {},
      THREAT: {},
    },
  }),
});

const result = await response.json();
console.log(result.attributeScores.TOXICITY.summaryScore.value);

import requests

response = requests.post(
    "https://api.getsieve.dev/v1/perspective/analyze",
    headers={
        "Authorization": "Bearer mod_live_abc123...",
        "Content-Type": "application/json",
    },
    json={
        "comment": {"text": "You are an idiot and nobody likes you"},
        "requestedAttributes": {
            "TOXICITY": {},
            "INSULT": {},
            "IDENTITY_ATTACK": {},
            "THREAT": {},
        },
    },
)

result = response.json()
print(result["attributeScores"]["TOXICITY"]["summaryScore"]["value"])

Differences from Google Perspective

While the request and response formats are compatible, there are a few differences:

• Authentication — Sieve uses Authorization: Bearer or X-API-Key headers instead of a key query parameter.
• Span scores — Sieve returns summaryScore only. Per-span scores are not currently supported.
• Additional attributes — Perspective attributes like FLIRTATION and SEXUALLY_EXPLICIT are mapped to Sieve’s sexual category internally.
• Rate limits — Follow your Sieve plan’s rate limits, not Perspective’s QPS limits.