Custom Rules

Custom rules let you extend Sieve’s built-in moderation with rules specific to your platform. Block game-specific slurs that general models miss, allow terms that are false positives for your community, or match complex patterns with regex.

Rule Types

Match exact words or phrases. Sieve handles Unicode normalization and leetspeak variants automatically.

Match patterns using regular expressions. For complex patterns that wordlists can’t capture.

Prevent specific words or phrases from being flagged. Override false positives from the AI tier.

Wordlist Rules

Wordlist rules match exact words or phrases. Sieve automatically normalizes Unicode and decodes leetspeak before matching, so you don’t need to add every variant manually.

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", "account for sale", "boosting service"],
    "priority": 10
  }'

Use cases:

Game-specific slurs or offensive terms not in general datasets
Known scam phrases specific to your platform
Brand-specific terms that should always be blocked
Competitor names in contexts where they indicate spam

Regex Rules

Regex rules match content against regular expressions. Use these for patterns that can’t be expressed as simple word lists.

curl -X POST https://api.getsieve.dev/v1/config/rules \
  -H "Authorization: Bearer mod_live_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "type": "regex",
    "action": "flag",
    "category": "spam",
    "pattern": "\\b\\d{2,}k\\s*gold\\s*(for\\s*)?\\$\\d+",
    "priority": 5
  }'

Use cases:

Matching price patterns in RMT spam (“10k gold $5”)
Detecting URLs with specific domain patterns
Complex obfuscation patterns unique to your platform
Matching structured data patterns (phone numbers, crypto wallet addresses)

Allowlist Rules

Allowlist rules prevent specific terms from triggering moderation. Use these to fix false positives where legitimate terms in your community get flagged by the AI tier.

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": ["assassin", "killer bee", "execute order"],
    "priority": 100
  }'

Use cases:

Game character names that contain violent terms (“Assassin’s Creed”)
In-game abilities with aggressive names (“Execute,” “Backstab”)
Community slang that sounds offensive out of context
Technical terms that trigger false positives

Priority Ordering

Rules are evaluated in priority order (higher number = evaluated first). This matters when rules conflict:

Priority	Rule	Action
100	Allowlist: “assassin”	Allow
10	Wordlist: offensive terms	Block
5	Regex: spam patterns	Flag
0	Default pipeline	Evaluate

A higher-priority allowlist rule will override a lower-priority wordlist block. Use this to create nuanced rule sets:

High priority (100+): Allowlists for known false positives
Medium priority (10-50): Wordlists for must-block terms
Low priority (1-9): Regex patterns for soft matching

Managing Rules

List Rules

curl -X GET https://api.getsieve.dev/v1/config/rules \
  -H "Authorization: Bearer mod_live_your_key"

Delete a Rule

curl -X DELETE https://api.getsieve.dev/v1/config/rules/rule_abc123 \
  -H "Authorization: Bearer mod_live_your_key"

Update a Rule

curl -X PUT https://api.getsieve.dev/v1/config/rules/rule_abc123 \
  -H "Authorization: Bearer mod_live_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "words": ["buy gold cheap", "account for sale", "boosting service", "power leveling"],
    "priority": 15
  }'