Understanding Our Moderation Process

Safety by Design: Our Moderation Philosophy

At Qerios, we believe that a social platform should be safe by default. We do not rely solely on user reports to maintain community standards. Instead, we have engineered a proactive, multi-layered moderation system designed to catch harmful content—including profanity, harassment, threats, sexual content, spam, and hate speech—before it ever reaches your inbox.

Our Promise: We employ a "silent filtering" approach. If a message is blocked by our systems, the sender is not notified, preventing them from attempting to bypass our filters or "game" the system.

The Three-Layer Protection System

We utilize a sophisticated, three-tier architecture to ensure maximum protection with minimal latency.

Layer 1: Custom Banned Words

The first line of defense is highly personalized. We understand that "harmful" can be subjective. You can define your own list of custom banned words in plain text. This layer operates instantly, ensuring that you never see the specific terms you have chosen to block.

Layer 2: Deterministic Regex Filter

Our second layer is a platform-wide deterministic regex (Regular Expression) filter. This layer processes content instantly and identifies known harmful patterns, sequences, and character combinations that violate our core community safety standards.

Layer 3: The AI Classifier (The Intelligent Layer)

This is the heart of our moderation engine. Once content passes the first two layers, it is analyzed by an AI classifier. This system examines the context and intent of the content across six critical harm categories:

1. Profanity
2. Harassment
3. Threat
4. Sexual Content
5. Spam
6. Hate Speech

This layer is asynchronous, meaning it processes in the background to ensure your user experience remains fast and fluid.

Redundancy Through Multi-Model Fallback

Safety is too important to rely on a single point of failure. If one AI model is unavailable or encounters a processing error, our system automatically initiates a 5-tier AI provider fallback strategy. The system sequences through:

1. OpenRouter
2. Groq
3. HuggingFace
4. Cloudflare
5. Gemini

This ensures that regardless of external service status, your safety remains uncompromised.

Your Role in Moderation

While our automated systems are robust, they are designed to work in tandem with our community. If our AI misses a subtle nuance, you can always utilize our reporting system. We encourage you to use the provided 11 reporting categories to flag issues that our automated filters might not fully contextualize.