How Praetoid Works
Understand the moderation pipeline from content to verdict.
Praetoid is AI-powered content moderation as a service. You send it your users' content — text, images, or both — and it evaluates that content against your moderation rules using AI. You get back structured verdicts with enforcement actions, and every decision is traceable back to the specific rule that triggered it.
Content In
Praetoid accepts your content as cases. A case is a container that holds the pieces of content you want moderated — a chat message, a profile photo, a marketplace listing, or any combination of text and images together. You can associate each case with a user and a conversation thread, giving the AI full context when evaluating content and enabling tracking across related submissions.
Cases enter the pipeline through checkout sessions. Your server creates a checkout by calling the API, specifying which moderation rules to apply. The checkout returns a short-lived token that your client uses to submit content directly. This keeps your API key on your server while letting clients submit from the browser or mobile app.
Split Authentication
Checkouts separate authentication from submission. Your server creates the checkout using your API key. Your client submits content using the checkout token. Your API key never leaves your server.
Evaluate and Enforce
Once a case is submitted, Praetoid evaluates each content item against your moderation rules. AI agents reason about whether content violates each rule. For images, vision models analyze visual content before evaluation. The result is a structured verdict — not just pass or fail, but a detailed breakdown of which rules were violated, which content triggered them, and why.
When violations are found, Praetoid produces enforcement actions based on workflows you configure. Actions like ban, mute, warn, or custom actions you define are collected into a sentence — a structured set of outcomes you can review, approve, or override. Sensitive actions can require human approval before taking effect.
Structured Reasoning
Every moderation decision includes which rule was violated, which content triggered it, and a natural-language explanation of why. This gives you a complete audit trail for every decision Praetoid makes.
Define Your Rules
Your moderation rules live in a Praetor — a versioned moderation profile. Think of it as a named judge with a specific rulebook. You create praetors for different content types: one for chat messages, another for profile photos, a third for marketplace listings.
Each praetor contains a set of policies. You write natural-language instructions like "content must not contain threats of violence" and the AI reasons about whether submitted content violates that instruction. Policies can also use pattern matching and classification for deterministic detection. Each policy defines what enforcement actions should be taken when violated through a configurable workflow.
Immutable Versioning
Praetors version immutably. When you deploy changes, previous versions are preserved. You can always see exactly which rules were in effect when a piece of content was evaluated.
Results Out
Praetoid delivers results through webhooks and callbacks. Configure webhook endpoints that subscribe to specific event types — case completions, violations, enforcement actions. Every webhook payload is cryptographically signed so you can verify authenticity.
You also configure how moderated content flows back to your application. Praetoid supports multiple transfer methods depending on your architecture: direct delivery in the response, presigned download URLs, or uploads to your storage endpoint. You choose the method that fits your infrastructure when creating each checkout.
Flexible Delivery
You control how results reach your application. Webhooks notify you of events in real time. Content transfer methods let you choose the tradeoff between latency, storage, and implementation simplicity.