Gaming Chat Moderation

Gaming moderation is fundamentally different from moderating a blog or forum. Players use aggressive banter that’s perfectly normal in context, spam RMT offers using creative obfuscation, and communicate in rapid-fire short messages. Sieve’s Gaming pipeline mode is built for this.

Why Gaming Is Different

“I’m gonna kill you” is normal PvP trash talk. “I’m gonna find where you live” is a real threat. Context matters more than keywords.

Gold sellers, boosting services, and account traders use heavy obfuscation: “g0ld 4 s4le” and Unicode substitution to evade filters.

Players deliberately obfuscate slurs and banned terms with character substitution, spacing tricks, and Unicode lookalikes.

Chat messages need sub-second moderation. Players notice delays, and slow moderation breaks the flow of gameplay.

Recommended Setup

For most games, use Gaming mode with chat context:

curl -X POST https://api.getsieve.dev/v1/moderate/text \
  -H "Authorization: Bearer mod_live_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "player message here",
    "context": "chat",
    "username": "PlayerOne"
  }'

Tip

Pass the username field alongside the message. Sieve uses this to detect username-based harassment and track per-user behavior patterns.

Code Examples

const response = await fetch('https://api.getsieve.dev/v1/moderate/text', {
  method: 'POST',
  headers: {
    'Authorization': 'Bearer mod_live_your_key',
    'Content-Type': 'application/json',
  },
  body: JSON.stringify({
    text: message,
    context: 'chat',
    username: player.username,
  }),
});

const result = await response.json();

if (result.action === 'block') {
  // Don't show the message, warn the player
  player.warn('Your message was blocked for: ' + result.primary_category);
} else if (result.action === 'flag') {
  // Show the message but log it for review
  chat.send(message);
  moderationQueue.add(result);
} else {
  chat.send(message);
}

import requests

response = requests.post(
    "https://api.getsieve.dev/v1/moderate/text",
    headers={
        "Authorization": "Bearer mod_live_your_key",
        "Content-Type": "application/json",
    },
    json={
        "text": message,
        "context": "chat",
        "username": player.username,
    },
)

result = response.json()

if result["action"] == "block":
    player.warn(f"Message blocked: {result['primary_category']}")
elif result["action"] == "flag":
    chat.send(message)
    moderation_queue.add(result)
else:
    chat.send(message)

using var client = new HttpClient();
client.DefaultRequestHeaders.Add("Authorization", "Bearer mod_live_your_key");

var payload = new {
    text = message,
    context = "chat",
    username = player.Username
};

var response = await client.PostAsJsonAsync(
    "https://api.getsieve.dev/v1/moderate/text", payload);
var result = await response.Content.ReadFromJsonAsync<ModerationResult>();

switch (result.Action) {
    case "block":
        player.Warn($"Message blocked: {result.PrimaryCategory}");
        break;
    case "flag":
        Chat.Send(message);
        ModerationQueue.Add(result);
        break;
    default:
        Chat.Send(message);
        break;
}

var http = HTTPRequest.new()
add_child(http)

var headers = [
    "Authorization: Bearer mod_live_your_key",
    "Content-Type: application/json"
]

var body = JSON.stringify({
    "text": message,
    "context": "chat",
    "username": player.username
})

http.request("https://api.getsieve.dev/v1/moderate/text",
    headers, HTTPClient.METHOD_POST, body)

# In the response callback
func _on_moderation_complete(result, response_code, headers, body):
    var json = JSON.parse_string(body.get_string_from_utf8())
    match json["action"]:
        "block":
            player.show_warning("Message blocked: " + json["primary_category"])
        "flag":
            chat.send_message(message)
        "allow":
            chat.send_message(message)

Handling the Response

Every moderation response includes an action field with one of three values:

Action	What It Means	Recommended Handling
allow	Content is clean	Display the message normally
flag	Content is borderline	Display but log for human review
block	Content violates rules	Suppress the message, warn the user

The response also includes per-category scores so you can implement custom logic:

{
  "action": "flag",
  "primary_category": "toxicity",
  "scores": {
    "toxicity": 0.72,
    "harassment": 0.45,
    "hate_speech": 0.12,
    "sexual": 0.01,
    "violence": 0.08,
    "self_harm": 0.00,
    "spam": 0.15
  },
  "tier": 1,
  "latency_ms": 145
}

Custom Rules for Your Game

Every game has specific slang, ability names, and community terms that need special handling. Use custom rules to fine-tune moderation for your game.

Common custom rules for games:

# Allowlist game-specific terms that trigger false positives
curl -X POST https://api.getsieve.dev/v1/config/rules \
  -H "Authorization: Bearer mod_live_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "type": "allowlist",
    "words": ["assassinate", "execute", "headshot", "killstreak", "nuke"],
    "priority": 100
  }'

# Block game-specific RMT spam
curl -X POST https://api.getsieve.dev/v1/config/rules \
  -H "Authorization: Bearer mod_live_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "type": "wordlist",
    "action": "block",
    "category": "spam",
    "words": ["buy gold", "cheap currency", "account boost", "power level service"],
    "priority": 10
  }'

Latency Considerations

For real-time game chat, latency matters. Here’s what to expect:

Scenario	Typical Latency
Tier 0 catches it (60-70% of requests)	< 5ms
Escalates to Gaming AI tier	200-400ms
Blended average	~80ms

Tip

If you need the absolute lowest latency, consider Edge mode (~140ms blended) which runs Llama Guard 3 on Cloudflare’s edge network. You’ll lose gaming-specific context awareness but gain speed. See Pipeline Modes.

Architecture tips for low-latency integration:

• Fire and forget: Send the moderation request async. Display the message immediately and retroactively hide it if blocked. This is the pattern most large games use.
• Pre-send check: Moderate before displaying. Adds latency but prevents any blocked content from ever appearing. Better for younger audiences.
• Hybrid: Use fire-and-forget for chat context and pre-send for username context.