The Tuteliq Roblox module is a pure Luau library that monitors in-game chat for grooming, bullying, and unsafe content via the Tuteliq API. Drop it into any experience and start protecting players with two lines of code.
Installation
Wally (recommended)
Add to your wally.toml:
[dependencies]
TuteliqSafety = "tuteliq/tuteliq-safety@1.0.0"
Rojo model file
Download TuteliqSafety.rbxm from the latest GitHub release and insert it into ServerScriptService in Roblox Studio.
Roblox Creator Store
Search for TuteliqSafety in the Creator Marketplace toolbox inside Roblox Studio, then insert the model into ServerScriptService.
Manual
Copy the TuteliqSafety/ folder into ServerScriptService in Roblox Studio.
Prerequisites
- Enable HTTP Requests — In Roblox Studio go to Game Settings → Security → Allow HTTP Requests and turn it on.
- Get an API key — Sign up at tuteliq.ai and create a key starting with
tq_.
Quick start
Create a Script in ServerScriptService:
local TuteliqSafety = require(game.ServerScriptService.TuteliqSafety)
TuteliqSafety.start({
apiKey = "tq_live_YOUR_API_KEY_HERE",
sensitivity = "high",
detect = { "bullying", "grooming", "unsafe" },
})
Never hardcode API keys in production. Use Roblox Secrets Store or a server-side configuration service to load keys at runtime.
How it works
Every chat message goes through a three-stage pipeline:
- Crisis keywords — Phrases like “kill myself” or “send me pictures” trigger an immediate API call to
/safety/unsafe for instant analysis.
- Grooming buffer — Messages are accumulated per channel. When the buffer reaches 5+ messages from 2+ unique senders, a conversation analysis is sent to
/safety/grooming.
- Batch queue — All other messages are batched (25 per request or every 5 seconds) and sent to
/batch/analyze for efficient bullying and unsafe content detection.
When a threat is detected above your sensitivity threshold, the module takes automatic actions based on severity:
| Severity | Actions |
|---|
| Low | Log to output |
| Medium | Log + notify admins via BindableEvent + webhook |
| High | Log + notify + mute player for 60s + webhook |
| Critical | Log + notify + mute + kick player (if enabled) + webhook |
Listen for detections
Connect to the TuteliqAlert BindableEvent to build custom admin UI or notifications:
local alertEvent = TuteliqSafety.onDetection()
alertEvent.Event:Connect(function(data)
-- data.type: "bullying" | "grooming" | "unsafe"
-- data.severity: "low" | "medium" | "high" | "critical"
-- data.confidence: number (0–1)
-- data.riskScore: number (0–1)
-- data.rationale: string
-- data.recommendedAction: string
-- data.playerName: string
-- data.playerId: number
-- data.message: string
-- data.timestamp: number
warn(string.format(
"[ALERT] %s | %s | Player: %s",
data.type, data.severity, data.playerName
))
end)
Integrate muting with your chat system
Muted players have the attribute TuteliqMuted set to true. Use this to filter messages in your chat UI:
player:GetAttributeChangedSignal("TuteliqMuted"):Connect(function()
if player:GetAttribute("TuteliqMuted") then
-- Block this player's messages in your UI
end
end)
Manual analysis
Analyze a specific message on demand, bypassing the batch queue:
TuteliqSafety.analyzeNow(
"some suspicious message",
"bullying", -- "bullying" | "unsafe"
player -- optional Player instance
)
analyzeNow does not support grooming detection directly. Grooming requires conversation context, which is handled automatically by the conversation buffer.
Configuration
TuteliqSafety.start({
apiKey = "tq_live_...", -- Required. Your Tuteliq API key.
sensitivity = "medium", -- Threshold: low (0.7) | medium (0.5) | high (0.3) | maximum (0.1)
detect = { "bullying", "grooming", "unsafe" },
actions = {
mute = true, -- Temporarily mute flagged players
kick = false, -- Kick on critical severity (disabled by default)
notifyAdmins = true, -- Fire TuteliqAlert BindableEvent
log = true, -- Log detections to output
},
muteDurationSeconds = 60, -- How long muted players stay muted
batchSize = 25, -- Messages per batch API call (1–50)
batchFlushIntervalSeconds = 5, -- Seconds between batch flushes
httpBudgetPerMinute = 400, -- Max HTTP requests per minute
webhookUrl = nil, -- Optional external webhook URL
})
Sensitivity levels
| Level | Threshold | Best for |
|---|
low | 0.7 | Minimal false positives, only flags high-confidence threats |
medium | 0.5 | Balanced — good default for most experiences |
high | 0.3 | Catches more subtle threats, may increase false positives |
maximum | 0.1 | Maximum protection — flags nearly everything suspicious |
API
| Method | Description |
|---|
TuteliqSafety.start(config) | Start monitoring chat |
TuteliqSafety.stop() | Stop monitoring and clean up connections |
TuteliqSafety.onDetection() | Returns the TuteliqAlert BindableEvent |
TuteliqSafety.analyzeNow(text, type?, player?) | Manually analyze a message |
TuteliqSafety.isRunning() | Check if the module is active |
Architecture
TuteliqSafety/
├── init.lua — Main entry point
├── Types.lua — Luau type definitions
├── Config.lua — Validation, defaults, thresholds
├── Logger.lua — Structured logging
├── HttpClient.lua — API communication via HttpService
├── RateLimiter.lua — Token bucket rate limiter
├── BatchQueue.lua — Message batching with timer/size flush
├── ConversationBuffer.lua — Ring buffer for grooming context
├── CrisisDetector.lua — Keyword matching for immediate analysis
└── ActionDispatcher.lua — Mute/kick/notify based on severity
Rate limiting
The module uses a token bucket rate limiter defaulting to 400 requests per minute. This is conservative and leaves room for your game’s other HTTP calls. Roblox enforces a global limit of 500 HTTP requests per minute per server.
Batch processing lets you analyze up to ~10,000 messages per minute within this budget.
Next steps
