Quickstart

Get your API key

You’ll get two keys:

mod_test_ — test mode, returns deterministic scores without calling AI models
mod_live_ — production mode, full pipeline

Make your first request

curl -X POST https://api.getsieve.dev/v1/moderate/text \
  -H "Authorization: Bearer mod_live_xxxxx" \
  -H "Content-Type: application/json" \
  -d '{
    "content": "I will find you and end you",
    "context": "gaming_chat"
  }'

const response = await fetch("https://api.getsieve.dev/v1/moderate/text", {
  method: "POST",
  headers: {
    "Authorization": "Bearer mod_live_xxxxx",
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    content: "I will find you and end you",
    context: "gaming_chat",
  }),
});

const result = await response.json();
console.log(result.action); // "flag" or "block" or "allow"

import requests

response = requests.post(
    "https://api.getsieve.dev/v1/moderate/text",
    headers={"Authorization": "Bearer mod_live_xxxxx"},
    json={
        "content": "I will find you and end you",
        "context": "gaming_chat",
    },
)

result = response.json()
print(result["action"])  # "flag" or "block" or "allow"

Understand the response

{
  "id": "mod_r_9x7mw2",
  "flagged": true,
  "action": "flag",
  "tier_used": 1,
  "latency_ms": 142,
  "categories": {
    "toxicity": {
      "score": 0.72,
      "threshold": 0.7,
      "flagged": true
    },
    "threat": {
      "score": 0.85,
      "threshold": 0.6,
      "flagged": true
    },
    "harassment": {
      "score": 0.45,
      "threshold": 0.6,
      "flagged": false
    },
    "profanity": {
      "score": 0.05,
      "threshold": 0.8,
      "flagged": false
    }
  },
  "context": "gaming_chat",
  "metadata": {
    "pipeline_mode": "general",
    "shadow_validated": true
  }
}

Here’s what each field means:

Field	Description
`id`	Unique moderation result ID. Use this for appeals or audit logs.
`flagged`	`true` if any category score exceeds its threshold.
`action`	The recommended action: `allow`, `flag` (send to human review), or `block`.
`tier_used`	Which pipeline tier handled the request (0 = local, 1 = standard AI, 2 = advanced AI).
`latency_ms`	End-to-end processing time in milliseconds.
`categories`	Per-category breakdown with `score` (0.0 - 1.0), `threshold`, and whether it was `flagged`.
`context`	The content context you provided. Affects scoring behavior.

Choose a pipeline mode

Pipeline modes control how Sieve classifies content. Set this per-request or as an account default.

Best for: Forums, comments, social apps

The default mode. Balanced scoring across all categories with standard thresholds.

{
  "content": "your message here",
  "context": "comment",
  "mode": "general"
}

Best for: In-game chat, gaming communities, Discord bots

Higher tolerance for competitive banter (“gg ez”, “you’re trash”), stricter on real threats and hate speech. Understands gaming-specific slang.

{
  "content": "gg ez get rekt noob",
  "context": "gaming_chat",
  "mode": "gaming"
}

Best for: Speed-critical paths, high-volume ingestion

Tier 0 only — local Rust processing, no AI calls. Sub-5ms latency, zero per-request cost. Use this when you need speed over nuance.

{
  "content": "message to check",
  "context": "chat",
  "mode": "edge"
}

Install an SDK

Pick your language and install the official SDK.

npm install @sieve/sdk

pip install sieve-mod

dotnet add package Sieve.Moderation

Then initialize the client:

import Sieve from "@sieve/sdk";

const sieve = new Sieve("mod_live_xxxxx");

const result = await sieve.moderate.text({
  content: "check this message",
  context: "gaming_chat",
});

if (result.action === "block") {
  // reject the message
}

from sieve_mod import Sieve

sieve = Sieve("mod_live_xxxxx")

result = sieve.moderate.text(
    content="check this message",
    context="gaming_chat",
)

if result.action == "block":
    # reject the message
    pass

using Sieve.Moderation;

var client = new SieveClient("mod_live_xxxxx");

var result = await client.Moderate.TextAsync(new TextRequest
{
    Content = "check this message",
    Context = "gaming_chat",
});

if (result.Action == "block")
{
    // reject the message
}

Next Steps

Understand the 3-tier pipeline and how scoring works under the hood.

Best practices for moderating real-time gaming communication.

Add your own blocklists, allowlists, and regex patterns.

Drop-in replacement for the Google Perspective API.